press /
Time Tax A movement
§ 7 — Chronicle

Newsfeed

A running chronicle of the regulation, research, and critique bearing on the extraction of digital time. We collect, we annotate, we link out. We do not editorialise where the source speaks for itself.

Twice weekly · Mon & Thu Updated 22 October 2026

October 2026

19 entries
21 Oct 2026
Regulation·PPC Land / Meta Q3 results

Meta books a $9.89bn charge for its teen settlement as ad revenue rises 26%

Meta’s Q3 results show advertising revenue of $63.06bn, up from $50.08bn, driven mostly by price: average price per ad rose 14% worldwide and 23% in the US and Canada, against impression growth of 10% and 8%. It booked a $9.89bn charge tied to an August agreement with state attorneys general over teen mental health, which Meta says provides for payments of up to about $18.59bn. PPC Land notes that settlement totals and jurisdiction counts do not fully reconcile across documents, and several derived figures are its own calculations.

Why we logged it. The charge and the growth sit in one filing. Price per ad, not volume, drove most of the revenue increase, and the settlement puts a figure on harm to a group of minors.

21 Oct 2026
Critique·Tech Policy Press / CELE

Unpacking the public health turn in internet regulation

Ramiro Álvarez Ugarte of the University of Palermo argues that Meta’s settlement with US state attorneys general, and a growing set of national and state laws, point to a health framing for limits on platform design and access, mostly for minors. He says such limits must still meet necessity and proportionality tests, that evidence of harm is contested, and that courts are beginning to push back. He predicts the approach will eventually reach adults, as tobacco and alcohol rules did. The adult extension is his prediction, not a development.

Why we logged it. Regulators are starting to treat design as a dose. That is closer to our frame than a speech frame, because it names a cost borne by the user. It also stops at harm, and a health regime has no mechanism for sharing the value taken.

21 Oct 2026
Critique·Pluralistic / Cory Doctorow

Adversarial interoperability is a team sport

Doctorow argues that because general-purpose computers can run any program, owners can counter companies that degrade or monetise their devices by installing alternative software, but anti-circumvention law such as DMCA Section 1201 makes that risky. He says AI coding tools now let ordinary people build such fixes, and that individual hacks are not enough without legalised reverse-engineering. Smart TVs full of ads and tracking are among his examples. This is polemic rather than reporting.

Why we logged it. Forced ads and device tracking are the extraction made visible on the screen in the living room. His remedy is exit, the right to remove the extraction yourself. Ours is return. The two are compatible, and both start from the premise that the user was never asked.

20 Oct 2026
Regulation·PPC Land / Apple WebKit

Safari 27 blocks identity firms that turn email addresses into tracking IDs

A WebKit post says Safari 27 blocks network access to third parties that convert email addresses from login and newsletter forms into persistent identifiers, at domain level and with no exceptions. Apple names no firms; PPC Land lists LiveRamp, ID5, UID2, OpenX, Permutive and Audigent as in the frame. The affected list is unconfirmed and the WebKit policy page is undated. PPC Land’s accompanying Chartbeat search-traffic figure is secondary, and it flags inconsistencies in it.

Why we logged it. The identity layer of ad targeting is being cut by the company that owns the browser, not by a regulator.

18 Oct 2026
Regulation·PPC Land / Google

Google Ads Data Manager stops requiring hashed customer data from 16 November

From 16 November, Data Manager will accept encrypted customer data by default and decrypt, normalise and match it inside a hardware-isolated trusted execution environment, covering Customer Match and enhanced conversions. Google cites better match accuracy. PPC Land notes that hashing never amounted to legal anonymisation. The source is an undated advertiser email posted by a third party, and questions about keys, opt-in and verification are unanswered.

Why we logged it. The privacy guarantee moves from hashing, which the source notes never amounted to legal anonymisation, to processing inside a vendor-run enclave.

17 Oct 2026
Regulation·PPC Land / Meta

Meta grants itself the right to crawl advertisers’ websites for AI

Meta’s revised Commercial Terms, effective 30 October, let it use automated tools and test credentials to access, copy and re-crawl websites linked to a business’s products, and train its own products on the content. Continued use counts as acceptance, with no partial opt-out, and the terms do not mention robots.txt. A Meta spokesperson says the clause is narrower than critics claim, and an attorney quoted says enforceability is unclear. PPC Land works from press coverage and excerpts, not the full contract.

Why we logged it. Advertisers pay for the audience and, under the new terms, would also supply content for Meta to train on. Acceptance is by continued use.

17 Oct 2026
Critique·Pluralistic / Cory Doctorow

Israeli ad-tech company was a delivery system for “mercenary spyware”

Doctorow uses the Citizen Lab findings to argue that programmatic advertising is insecure by design, that surveillance ads add little over contextual ones, and that regulators should ban surveillance advertising. Some figures, such as a click-through uplift of under 5 percent, are given without a source, and the claims about the firm’s certifications are relayed from Citizen Lab. This is advocacy, not reporting.

Why we logged it. The sharpest statement of the argument that the extraction is not even efficient. We log it as polemic and pair it with the report it draws on.

15 Oct 2026
Critique·Citizen Lab / University of Toronto

Exposing AdFection: how Discover-Tech turned targeted ads into a pipeline for zero-click spyware

Citizen Lab reports with high confidence that ads bought by Israeli firm Discover-Tech delivered Candiru spyware to at least 45 Al Jazeera devices with no click required, a technique it calls AdFection. PPC Land’s coverage adds that the firm bid on 3.77 billion impressions between March and June 2026 and had been listed by Google since 2020. Attribution is a research finding, not a legal one, and the figures beyond the report’s summary come from that coverage.

Why we logged it. The auction that sells our attention was used as a delivery route for spyware, by a vendor Google had listed since 2020.

14 Oct 2026
Regulation·PPC Land / IAB Tech Lab

IAB Tech Lab says Apple can block ad vendors remotely, with no notice or appeal

An essay by IAB Tech Lab’s Rowena Lam argues that Apple’s iOS 27 WebKit blocklist of ad-tech domains can be extended remotely, without notice or appeal. The remote-update mechanism is described as unreleased, the published list is small, and Apple has not commented. Lam leads the Trusted Server product the essay promotes.

Why we logged it. The platform that owns the screen sets the terms for everyone who sells against it. This is an interested party’s account, and we log it as one. The dispute is between intermediaries; the people whose attention is being traded are not parties to it.

13 Oct 2026
Regulation·PPC Land / Datatilsynet

Norway fines Xandr NOK 92m over Grindr user data

Norway’s data protection authority fined Microsoft-owned Xandr, its largest fine to date, for receiving and passing on Grindr user data between July 2018 and April 2020 without explicit consent. It held Xandr a joint controller with Grindr, responsible for checking that consent existed. Xandr has three weeks to appeal, and the case took nearly seven years from the 2020 complaint.

Why we logged it. The ad exchange, not only the app, is held to account for the transfer. Seven years from complaint to fine is the price of a compulsory transfer nobody agreed to. Nothing in the remedy returns anything to the users whose data moved.

12 Oct 2026
Critique·Tech Policy Press / EPIC

Spirit Airlines is selling its workers’ data. That can’t fly.

Defunct Spirit Airlines is seeking bankruptcy-court approval to sell about $10 million of employee records going back to 1986 to Google, reportedly for AI training. EPIC counsel argue that Bankruptcy Code privacy protection covers consumers, not workers. A later filing reportedly excludes the most sensitive records; talks on deidentifying the rest were continuing.

Why we logged it. Adjacent to attention, and the same logic. Value is taken from people who were never asked, and the sale is arranged between the parties who hold the data. The workers produced it and have no seat at the table.

7 Oct 2026
Regulation·PPC Land

San Diego County sues AppLovin over adult ads served in children’s games

San Diego County has filed a complaint in Superior Court alleging that AppLovin’s SDK served ads for sex products, gambling, cannabis, alcohol and vaping to children’s devices, removed a flag that let developers mark users under 13, and used interfaces built to provoke accidental clicks. It seeks civil penalties of up to $2,500 per violation under California’s False Advertising and Unfair Competition Laws, plus restitution and an injunction. These are allegations, not findings.

Why we logged it. The penalty is counted per ad served and per data collection, so every impression carries its own price. That is the unit of account this movement argues for, applied here to harm rather than to value. Nothing in the remedy returns anything to the child whose attention was sold.

6 Oct 2026
Regulation·PPC Land

Google and the DOJ file rival ad tech judgments: six months to open AdX, or twelve

Competing proposed final judgments were lodged on 2 October in the Eastern District of Virginia. Both sides accept a six-year monitor, a ban on first and last look, and AdX accepting bids from rival publisher ad servers. They split on timing (six months versus twelve), global scope, the monitor’s access and the standard that would trigger contempt. Judge Brinkema earlier declined to order a divestiture of AdX, and an appeal is widely expected. Reported secondhand; the filings themselves are the primary record.

Why we logged it. The fight is now over the clock and the enforcement teeth, not over whether the auction was rigged. Every month of delay is a month in which the price of attention is set inside a system a court has already found unlawful, and the remedy as framed concerns publishers and rivals, not the people being auctioned.

6 Oct 2026
Critique·Tech Policy Press

Meta’s multibillion-dollar settlement left a lot to be desired

Juvaria Shahid of Tech Justice Law reads the settlement with state attorneys general, described as up to $17.1 billion, as short of a “Big Tobacco moment.” She notes a default two-hour limit for under-18s that only a parent can change, a midnight-to-6am block and a non-personalised feed option, and argues that data collection and targeting carry on inside the limit and that opt-in protections depend on consent users rarely give. She reports that about 30 percent of the sum is withheld unless rivals follow. The terms are as she describes them.

Why we logged it. It states our objection from the policy side: a time cap treats the symptom and leaves the extraction engine running. A cap on hours does not touch what is taken in them.

5 Oct 2026
Critique·Pluralistic

Doctorow: “Scrutinized”

Cory Doctorow argues that camera glasses with facial recognition would mostly enable public shaming, and separates scrutiny, the felt sense of being judged, from surveillance as such. He holds that people need a protected space to make mistakes and grow, and notes that the executives pushing ubiquitous watching keep privacy for themselves. The essay links a 404 Media discussion, Kashmir Hill’s reporting and others.

Why we logged it. Only partly about attention, and logged for the surveillance strand. The point worth keeping is that watching is not neutral. It takes something from the person watched, whether or not anyone is charged for it.

5 Oct 2026
Regulation·Tech Policy Press

The EU’s tech regulation paradox is getting harder to ignore

Mark Scott argues the Commission is pursuing two agendas at once: simplifying digital rules for competitiveness through its omnibus proposals, while proposing new child-safety rules. He lists the Digital Fairness Act as forthcoming, DMA and DSA reviews likely in early 2027, and DSA preliminary findings against TikTok and Meta on minors and addictive design. This is analysis, not a regulatory act.

Why we logged it. The Digital Fairness Act is where dark patterns and addictive design are most likely to be written into law. Whether it arrives intact depends on which of the two agendas wins.

3 Oct 2026
Regulation·PPC Land

Judge Castel sends Google’s ad tech damages claims to a jury

On 30 September Judge P. Kevin Castel (S.D.N.Y., MDL 21-md-3010) denied most of Google’s summary judgment motions, sending private damages claims to a jury. Figures cited, including a $1.72 billion AdX class overcharge claim, are claimants’ expert numbers, not findings. This is a denial of summary judgment, not a verdict, and the source is a summary of the 88-page opinion.

Why we logged it. The first real attempt to put a price on what the auction took, and the claimants are publishers. The people whose attention was the underlying asset are not in the courtroom.

2 Oct 2026
Regulation·PPC Land

Italy fines IQVIA €7m over patient records it called anonymous

Italy’s data protection authority, the Garante, held that records tied to a stable 22-character code were not anonymous and said a hash protects only the part of a record directly linked to identity. It distinguished the Court of Justice’s 2025 recipient-based ruling on pseudonymised data, because IQVIA designed the process. This is a health data case; any read-across to advertising identifiers is by analogy.

Why we logged it. Ad tech leans on hashed identifiers as though hashing were anonymity. A regulator saying plainly that a stable code is still a person is the argument we would want repeated in the advertising context.

2 Oct 2026
Critique·Tech Policy Press

Francine Berman’s Better Tech is a call to arms

Nathan Sanders reviews Berman’s MIT Press book, which argues that digital technology is critical infrastructure and deserves the design scrutiny and governance of roads or the electrical grid, and which profiles public-interest technology practitioners. He praises its optimism and disputes some of its claims about AI.

Why we logged it. Not squarely about attention. Logged for the infrastructure framing, which sits close to the idea that a shared resource taken at scale needs a public account of who takes it.

September 2026

33 entries
30 Sep 2026
Regulation·PPC Land

Apple’s iOS 27 appears to block The Trade Desk’s ad delivery on Safari

Safari in iOS 27, released 14 September, blocks The Trade Desk’s core ad request domain and its UID2 identity domain, along with ID5, Audigent, LiveRamp and Permutive, while Google’s doubleclick.net kept working. The Trade Desk raised it with WebKit engineers. Apple has not said whether this is policy or a bug, and it is not a regulator action.

Why we logged it. A reminder that the plumbing of ad targeting is controlled by a few firms, and that whether your attention can be tracked is decided between them, not by you.

30 Sep 2026
Regulation·Tech Policy Press

Is the Kids Online Safety Act the “social reckoning” we are looking for?

Karolina Mackiewicz of George Washington University notes that the Senate Kids Online Safety Act (S.1748) would set protective defaults for known minors, including limits on infinite scroll and autoplay, plus a duty of care for design features, while the House version (H.R.7757) drops the duty of care. She argues the real fix is standards that put safety ahead of engagement for all users, without relying on age verification. This is her argument, not a reading of the bill text.

Why we logged it. Close to our own framing. The rule that matters is on engagement design for everyone, not a carve-out for children.

30 Sep 2026
Regulation·Tech Policy Press

A safer internet should not come with democratic shortcuts

Asha Allen of CDT Europe says the proposed EU Kids Act would turn non-binding Digital Services Act guidelines into binding law, including age verification, age assurance and parental monitoring, but is moving on an expedited track without proper consultation. She links it to earlier shortcuts on the AI Omnibus and Chat Control.

Why we logged it. Child-safety rules for attention products are arriving alongside more monitoring infrastructure. Protection that needs more surveillance of the user is a poor trade.

29 Sep 2026
Critique·Pluralistic

Doctorow: surveillance pricing and its mirror image in wages

Reviewing Lindsay Owens’s Gouged, Cory Doctorow argues firms use commercial surveillance to estimate each person’s ability to pay, citing Instacart raising prices by up to 20 percent on personal financial data, and that algorithmic wage setting is the mirror scheme for workers. He roots both in a shift from “fair price” to “whatever the market will bear.” A polemical review, not an independent test of the book.

Why we logged it. The most direct current statement of extraction through data. What is collected about you sets what you are charged and what you are paid.

29 Sep 2026
Critique·The American Prospect

A review of Gouged names the data trails behind individualised pricing

Hal Singer’s review of Owens’s book lists the cases it uses: carmakers sharing driving data that reaches insurers, McDonald’s app tracking visit frequency, and Starbucks selling customer data to more than 60 third parties. These are the book’s examples as relayed by the reviewer and were not checked against it.

Why we logged it. A second angle on the same book, with named examples of everyday data trails turned into price.

29 Sep 2026
Regulation·PPC Land

Publishers take a bill requiring AI scraping bots to identify themselves to Congress

Around 300 publishing executives, organised by the News/Media Alliance and America’s Newspapers, lobbied for a federal Stealth Bot Prohibition Act requiring AI scrapers to name themselves and their purpose. New York passed a state version in June 2026. The federal bill is introduced, not passed.

Why we logged it. Peripheral to attention, but the same pattern: value taken from people’s work with no declaration of who is taking it.

28 Sep 2026
Regulation·Tech Policy Press

Public Knowledge: Google ad tech remedies show the need for legislation

Patrick Gallaher summarises the behavioural remedies a Virginia federal judge imposed instead of a breakup, including AdX integration with Prebid, publisher data export, a ban on self-preferencing and a court-appointed monitor. His argument is that Google still owns both the ad server and the exchange, so the conflict of interest survives. This is an advocacy group’s reading.

Why we logged it. Read alongside the rival final judgments filed this month. The remedy is being argued over in detail while the design of the auction stays largely in place.

28 Sep 2026
Research·arXiv

Benn, Merz and Grüning: “Reclaiming the social in social media”

A preprint arguing that moderation, prosocial design and age restrictions leave the attention-driven model intact, and that the harms come from how platforms are built inside the attention economy. The authors derive four technical properties (operator blindness, algorithmic sovereignty, operational parity, verifiability) and propose an architecture meant to make attention extraction structurally impossible. Not peer reviewed, and the architecture is a proposal with no deployment evidence.

Why we logged it. It reaches our diagnosis from the engineering side: the problem sits in the business model, not in the content or the user’s age. Treat the proposed architecture as an argument, not a result.

25 Sep 2026
Regulation·PPC Land

A jury puts a number on Meta’s broken promises: 43.9 million of them

A jury in New Mexico’s First Judicial District Court found 26 of 29 Meta statements deceptive across five categories, among them “we do not and never will sell any of your information” and “we don’t profit from hate.” It counted 43,899,720 violations. Under a state law allowing $5,000 per willful violation, that implies a theoretical ceiling near $219.5 billion. The judge has yet to set the actual penalty, Meta has rejected the verdict and is expected to appeal, and Fortune reports that a $375 million child-safety penalty from the same litigation track has not yet been paid.

Why we logged it. It is the first time a court process has attached a unit price to what these promises were worth, and the unit is the statement rather than the user. Read the headline figure as a jury’s price tag, not a payment on its way: the number a judge sets will be far smaller, and the appeal will take years. What stays useful is the finding underneath it. A company was held to have told the public one thing about its data and done another, on a scale that can be counted. Nothing in the verdict routes a cent to the people whose data and time made the statements worth making.

25 Sep 2026
Regulation·PPC Land

Alabama takes its settlement partly in minutes

TikTok settled Alabama’s teen-safety suit for $116.2 million: $100 million in restitution, $14.2 million in fees, and a further $183.8 million contingent on other states signing equivalent deals. The consent decree also requires a 120-minute default daily cap for under-18s that resets at midnight, an overnight block, muted push notifications during school hours, removal of “cosmetic procedure” filters by March 2027, and a non-personalised chronological feed by June 2027. Several of the extra restrictions only activate if Meta, Snap and YouTube adopt equivalent rules.

Why we logged it. Most penalties in this chronicle are paid in money and leave the design untouched. This one is partly paid in time, ordered by a court rather than offered by the platform, which makes it the nearest thing yet to the argument turned inside out. The counter-reading deserves equal space. The EFF’s David Greene has argued that the wider Meta settlement pattern obliges platforms to collect more data, through age verification and proactive monitoring, and hands teens limits set by parents or the company rather than tools of their own. A cap on minutes helps; it does not settle who is owed for the minutes that remain. Reach is conditional and the default is a default, not a wall.

21 Sep 2026
Critique·PPC Land

Microsoft measured what its assistant took from publishers, and the figures were in a court filing

Internal Microsoft data disclosed in copyright litigation shows Copilot answers cutting clicks to New York Times properties by 87 to 93 percent, Daily News domains by 83 to 91 percent, and Ziff Davis properties by up to 94 percent, relative to a web-search referral. Separately, 87.78 percent of ChatGPT users visit no external website in a session, against 26.91 percent of people using Google search. A Microsoft executive is quoted acknowledging that the products are “largely substitutive, period, and will get more and more substitutive.”

Why we logged it. The click was the mechanism by which attention became publisher revenue. Here is a measurement of it being removed, taken by the company doing the removing. The attention has not gone anywhere; it is now answered, and kept, inside the assistant, with nothing flowing back to the source that supplied the answer. Caveats: these are internal figures surfaced through discovery and reported second-hand, the comparison is against search referral rather than an open web, and the “fair use” commentary quoted alongside them is an analyst’s view, not a finding.

21 Sep 2026
Regulation·Data Protection Commission (Ireland)

Ireland prices unlawful location tracking at €403 million

The Data Protection Commission found that Google’s “Web & App Activity” and “Location History” features unlawfully processed location data, that transparency obligations were not met across all three location features examined, and that data was retained longer than necessary. Deputy Commissioner Graham Doyle said people “could have been unaware that their location was being used to, for example, influence them with ads” and “could lose control over their personal data.” Google has six months to correct its practices. RTÉ has the news coverage.

Why we logged it. The regulator’s own sentence is the one to keep: the ads were the destination, and the person was never told the location was on the route. Note what the decision does not do. The fine is administrative, and using location to target advertising is not banned; it was found to have been collected and disclosed unlawfully. The remedy is a better notice. The transfer, a person’s whereabouts converted into ad revenue, continues once the notice is fixed.

21 Sep 2026
Regulation·Tech Policy Press

Canada proposes to make safety a condition of market access

John Matheson describes Bill C-34, the proposed Safe Social Media Act, as “a system of conditional market access” rather than an age ban. A new Digital Safety Commission would let platforms keep serving under-16s only if they meet defined standards: disabling addictive-by-default features, restricting targeted advertising to minors, and using privacy-preserving age verification. In Matheson’s framing the law “shifts the question from whether to keep children off social media to whether social media can be built to deserve them.”

Why we logged it. Most child-safety proposals police the child. This one polices the design, and attaches the penalty to the thing that earns revenue. That is a more interesting lever than a ban, and closer to how we would want a cost placed on extraction. Caveats: it is a bill, not law, the enforcement mechanics of the exemption standard are still to be written, and the age-verification requirement carries exactly the data-collection risk critics have raised against other settlements.

21 Sep 2026
Research·Journal of Happiness Studies

What the screen time does to you matters more than how long it lasts

Two studies of Canadian university students, a cross-sectional sample (N=348) and a week-long daily experience-sampling sample (N=645), tested whether screen time’s link to wellbeing depends on whether use satisfies or frustrates needs for competence, autonomy and relatedness. Need satisfaction during screen use predicted positive wellbeing and need frustration predicted negative affect, with effects holding “beyond screen time duration” at both the daily and between-person level. The authors conclude that research and policy “should include assessments of the quality of screen time rather than just how long people spend on the screens.”

Why we logged it. This complicates the simplest version of our case, and we would rather carry the complication. If harm tracks quality more than duration, an hours-only argument is weaker than it sounds. It does not touch the argument about design, which is that platforms are built to produce use that frustrates as readily as it satisfies, provided it continues. Caveats: university students in one country, self-report measures, and a study of wellbeing rather than of any platform’s business model.

20 Sep 2026
Regulation·PPC Land

A Frankfurt court decides that running an ad auction is editorial control

In Finanzfluss v. Meta Platforms Ireland Limited, a case over impersonation-fraud ads, the Frankfurt Regional Court rejected Meta’s claim to hosting-provider immunity under the Digital Services Act. It held that “operating an advertising auction according to criteria set by the defendant” — ranking by quality score, applying eligibility filters, picking winners — is editorial control, not passive storage. Meta must now disclose reach and revenue data on the fraudulent ads and faces up to €250,000 per future breach.

Why we logged it. The neutral-pipe defence has been the load-bearing wall of the ad economy. This is the first ruling we have logged that rejects it on the ad-serving mechanism itself, not on content moderation. The reasoning generalises: any platform that ranks, filters and distributes ads through an auction it controls is doing the same thing. Caveats matter here. It is a single national court, not tested on appeal, and not yet exported to other jurisdictions. If it holds, it is a bigger story than this week’s coverage suggests.

17 Sep 2026
Regulation·PPC Land

The seal comes off: what the court asked for instead of a sale

Judge Leonie Brinkema’s remedies opinion in the United States ad tech case was filed under seal on 2 September and became public, unredacted, on the 16th — 106 pages of reasoning behind a two-page order that had said only that Google would keep AdX. The obligations are engineering ones. Google must build API integrations connecting both AdX and DFP to Prebid, and those integrations must be functionally equivalent, a phrase that forecloses the familiar pattern in which a rival connection technically exists and technically underperforms. AdX must submit real-time bids into competing publisher ad servers on the same terms DFP receives them. Google must share win and loss bid data with publishers and publish technical documentation explaining how DFP picks a winner. AdWords may no longer bid directly into DFP. A court-appointed technical monitor serves six years with access to employees, systems and source code, and the order applies globally. Display & Video 360, the demand side, emerged untouched: the court found an insufficient connection to the conduct at issue. On the refusal to order divestiture, the companion reading of the opinion records the decisive finding — the plaintiffs “failed to identify a viable AdX acquirer” during the remedies trial, and their case for a breakup “boils down to a lack of trust that Google will comply with an Order from this Court.”

Why we logged it. We logged the order itself two weeks ago without being able to read it. Now we can, and the shape is clearer than the headline suggested: the court has converted a proven monopoly into a disclosure regime. Publishers will be told how the auction works; they will not own any part of it. Nothing in the judgment reduces the share taken in transit, and nothing in it returns anything to the people whose attention is the commodity being auctioned — they were never parties. The honest caveat runs the other way too: functional equivalence and source-code-level supervision are real obligations, unusually deep and unusually long, and a monitor may extract more than a forced sale to an unidentified buyer would have. What the record shows is that the remedy was chosen partly because no buyer appeared.

16 Sep 2026
Regulation·PPC Land

Chrome now counts the advertisements on every page — and owns the definition of an advertisement

On 15 September Google added four measurements to the Chrome User Experience Report, the public dataset already underpinning Core Web Vitals: ad count, meaning the average number of advertisements visible in the viewport; ad density, the share of viewport area they occupy; and ad weight in both network bytes and processor milliseconds. The viewport is sampled once per second, and figures are reported at the 75th percentile — a bad-but-not-worst-case visit rather than an average one. Two conditions shape what the dataset can describe. Aggregate reporting is restricted to sites carrying an ads.txt file naming at least one authorised seller, which quietly excludes subscription sites, direct sellers and sites with no third-party advertising at all: a measure of the programmatically monetised web, not the web. And every figure rests on a classifier — Google’s own proprietary ad filter list. Google has set no thresholds and says it has no plans to. Alex Cone, group product manager at Chrome, would say only: “We look forward to feedback from the ecosystem.” An unusual coalition — The Guardian, Mediavine, Index Exchange, Adelaide — has lined up behind it.

Why we logged it. For the first time there is a public, per-site number for how much of a page is taken. That is the measurement a tax needs, and we would rather it existed than not. But note who is holding the ruler. The company publishing the count of advertisements is the largest seller of advertisements on the web, the operator of the browser doing the counting, and the owner of a demand-side platform that has welcomed the signals — and the definition of what counts as an advertisement is a proprietary artefact publishers cannot audit. Note also what happens to an open question in this market. Google proposes no threshold; the first buyer to build a filter will set one, and a number nobody proposed becomes the standard. The metric arrives, finally, at the exact moment generative search is shrinking the traffic those pages monetise — which means a site earning less per visitor is being handed a scorecard for the only lever it has left.

16 Sep 2026
Critique·Tech Policy Press

The engineer who built the Like button says the button was never the problem

Justin Rosenstein helped create the Like button at Facebook. Writing after Meta’s settlement conceded that Instagram and Facebook will stop showing minors their like counts, he sets out why a feature designed to connect people became a central exhibit in a child safety trial: not bad intentions, but economic misalignment. “Addiction is more profitable than connection, so the button became a tool to capture our attention, drive unhealthy comparisons, and keep us obsessively refreshing our feeds.” He generalises the mechanism across the economy — treatment is profitable and prevention is not, sugar is profitable and fruit is not — and argues that technical AI alignment answers only whether a system does what we ask, while the economy decides what we ask it to do. “AI will supercharge whatever economy we put it in.” His remedy is not regulation alone, which he expects to be skirted and lobbied around, but what he calls economic alignment, arrived at through economic democracy: participatory budgeting, community land trusts, public banks, worker ownership. He cites over 11,000 communities already running some version of it, and $400 million invested through participatory budgeting across more than 40 US cities.

Why we logged it. This is the Conversion Dividend’s argument arriving from inside the building, made by the person who wrote the feature. The diagnosis is identical to ours: the extraction is not a bug in the product but the product doing what it was paid to do, and no amount of design reform survives an incentive pointed the other way. Where we part company is the instrument. Rosenstein reaches for democratic control of allocation; we argue for a declared share returned to the people whose attention produced the revenue, which is a narrower and more auditable claim. Treat the piece as an ally’s argument rather than our own — it is a perspective column by a founder promoting his own non-profit, and the mechanism by which participatory budgeting scales “from city budgets to whole economies” is asserted rather than shown.

16 Sep 2026
Critique·Tech Policy Press

Meta’s settlement, priced in days of revenue

Meta’s stock rose as much as 4% on the announcement of its settlement with 47 states, which is the first thing the economist Sumit Sharma asks readers to explain. His case that the design remedies will not bind runs on arithmetic. The two-hour daily teen limit sits above actual behaviour: a 2023 Gallup survey put average teen use across Instagram and Facebook at 1.2 hours combined, so for most teenagers the cap changes nothing, and the one study cited in the master complaint found benefits at ten minutes per platform, not two hours. The cap also excludes messaging, long-form video and interactions with AI companions and chatbots. A lower one-hour limit, contingent on competitors agreeing to equivalent constraints, still permits two hours across the two apps because it applies to each separately — and it would bite harder on Snap, TikTok and YouTube, where teenagers spend more time. Sharma reads the contingent payment of roughly $5 billion as an investment in a compliance perimeter every new entrant must then clear on terms Meta has set, and notes the ‘Meta Injunctive MFN’ clause permitting weaker designs if rivals adopt them. The $12.19 billion payable is, on his calculation, less than 20 days of Meta’s revenues. Default teen feeds remain algorithmic and infinite.

Why we logged it. The omission Sharma calls most glaring is the one that matters most to us: the settlement carries no requirement to publicly disclose the metrics used to measure online harm, or to report them. Auditing is done confidentially by an auditor on Meta’s payroll. A dividend cannot be calculated against numbers nobody outside the company is permitted to see, and this settlement establishes that a landmark child-safety resolution need not produce any. The caveat is that this is advocacy by an economist who runs a competition non-profit, and the competitive reading — that the remedy advantages Meta against rivals — is an inference, not a finding. The revenue arithmetic and the MFN clause are not.

15 Sep 2026
Critique·Digiday

A paywall that decides, reader by reader, whether you are worth more as a subscriber or as an impression

Speaking at the Digiday Publishing Summit, Phil Andraos, general manager for digital at Reuters, described the dynamic paywall the agency deployed earlier this year. Machine-learning models take three inputs — the behaviour of the individual user, the performance of the specific article measured by whether it has historically converted readers into subscribers, and the estimated advertising revenue that user represents — and then decide whether to show that person a paywall at all. The gate tightens during active news cycles, when both conversion likelihood and content performance rise, and loosens on slow days. Andraos on the launch: “It’s basically like magic. Overnight, the dollars start to flow.” Pricing is a flat $4 a month across 80 countries with no introductory rates, and subscribers do not receive an advertisement-free experience, so converting a reader removes no inventory. His summary of the trade: “A smaller pool of highly engaged users can drive so much more ad revenue than a high base of loosely engaged readers.” The context that claim has to survive, recorded in PPC Land’s fuller account: Similarweb estimates Reuters site traffic fell 19% year over year in August 2026.

Why we logged it. This is the clearest public description we have yet found of the calculation this project exists to name. A person arrives at an article and a model asks, in the milliseconds before the page renders, which of two extractions is worth more from this particular person right now — their money, or their attention. Both answers are revenue. Neither is a choice the reader makes, and the reader is never shown the number. We would note in fairness that Reuters has built something more honest than most: a real price, published, the same in 80 countries, with no discount trap. The extraction is being measured rather than hidden. But a valuation performed on a person, by a system they cannot see, using an estimate of what their attention is worth, is precisely the transfer we say goes unrecognised — and here it has been engineered, named, and described from a conference stage as magic.

15 Sep 2026
Regulation·Tech Policy Press

Brussels proposes to write “no addictive design” into primary law

Mark Scott reports on the EU Kids Act, presented on 17 September and trailed in a draft Communication obtained by Euractiv. The age structure is graduated: no access below 3, parent-controlled accounts on child-friendly services from 3 to 13, parent-opened introductory accounts with capped peer contacts and daily time limits from 13 to 15, and autonomous accounts from 15. Scope runs through a category the draft calls “Social Media+”, covering social platforms, video-sharing services, online games, and AI chatbots and companions. The design obligations are the part worth reading closely. The draft text mandates no addictive design features, naming infinite scrolling, artificial notifications and certain reward mechanisms, and requires recommender systems to go beyond engagement-based signals. Very large platforms would file compliance plans and could not launch a new feature without a positive Commission opinion inside 30 days. A supervisory fee would fund enforcement. The Commission’s own framing is unusually plain: platform business models “designed to maximise the time spent online… may exploit the vulnerabilities of children,” and “parents, not algorithms, should be raising Europe’s children.”

Why we logged it. A European institution has written our premise into a legislative text: that the extraction of time is the business model, not a side effect of it, and that the features performing it can be named and prohibited. “Go beyond engagement-based signals” is, if it survives, the most consequential sentence in European platform law. Three caveats, all load-bearing. This is a proposal: it needs Parliament and the Council, and would not apply before 2028 at the earliest. The draft does not define which platforms count as “risky”, nor quantify the contact and screen-time caps. And France offers the cautionary precedent — its own under-15 ban was struck down by the Constitutional Council on proportionality grounds eighteen days before it was due to apply. The mandatory age verification at account creation carries its own surveillance cost, which we will not pretend is small.

15 Sep 2026
Critique·PPC Land

The exam room becomes a media channel

PatientPoint operates the largest point-of-care advertising network in the United States: 145,000 digital screens across 32,000 medical offices — more locations than Starbucks and McDonald’s combined. The footprint is not one format but five, spanning waiting-room displays, digital check-in portals, screens inside exam rooms, back-office placements, and mobile targeting reaching the patient before and after the visit. Pharmaceutical point-of-care spending passed $1.2 billion last year, and consumer packaged goods and supplement brands are now buying into a channel built for prescription products. The company has assembled the apparatus of a conventional media business — a Manhattan upfront drawing 150 buyers, MRC accreditation, a general manager recruited from programmatic rather than health publishing. On privacy, senior SVP Kate Calabrese states the design in a sentence: “Everything is tokenized, so we don’t know who the individual is.” Its Precision product draws on consented electronic health record data that the company describes as separate from HIPAA authorisations. PPC Land sets it beside the same week’s advertising on car dashboards, in console games and on smart refrigerators, and names the pattern: advertising has run out of conventional inventory and is annexing physical surfaces that were never sold before.

Why we logged it. The waiting room was the last reliably unsold interval in an ordinary day — time you are obliged to spend, in a place you did not choose, in a state of mind that makes you unusually persuadable. It has now been inventoried and priced. We record the company’s privacy position without endorsing it: a tokenised identifier is still an identifier, and the protection depends entirely on what can be joined to it downstream. The distinction between consented EHR data and HIPAA-covered data is legally real, which is the point — it means the safeguards travelling with a patient’s data here are contractual, set by the seller, rather than statutory. No formulary governs which brand may address a patient in the room where the prescribing happens.

14 Sep 2026
Research·arXiv (econ.TH)

A model in which the platform decides how much of you to spend

Yi Chen, Fei Li and Marcel Preuss model revenue-maximising attention allocation on an ad-funded social media platform governed by recommendation algorithms. The framing is the part to carry: attention is treated as costly and finite, and the platform can either monetise it through advertising or spend it increasing a creator’s exposure — an explicit trade-off between monetisation and production incentives. In a two-sided model with heterogeneous viewers and creators under private information, the optimal recommendation mix “includes content that is ex post suboptimal for some viewers to leverage network externalities.” The distortions are targeted rather than uniform: low-ability creators are excluded outright, high-ability creators subsidised through exposure or direct payment. The authors draw out implications for personalisation regulation and advertising markets. Open access, posted 14 September.

Why we logged it. Here is the extraction written as a constrained optimisation, by economists, without polemic. Attention appears in the model as a resource the platform allocates — a budget it holds and spends — and the optimum it derives deliberately serves some people worse in order to hold the system together. That is our argument stated in the field’s own notation. Two honest limits. This is theory, not measurement: it describes what a revenue-maximising platform would do, not what any named platform does. And it is a preprint, not yet peer-reviewed. What it supplies is the accounting frame — once attention is a quantity on someone’s balance sheet, the question of whose balance sheet becomes a legitimate one to ask.

14 Sep 2026
Critique·404 Media

The contribution nobody priced: humans are reading the chats

Joseph Cox reports, from leaked internal documents and real prompts, that OpenAI is hiring hundreds of contractors to read a stream of genuine ChatGPT conversations and rate the model’s replies — work internally referred to as Project Lily. The reviewers do not see usernames, and OpenAI says it attempts to strip personal information before prompts reach them, but the company acknowledged sensitive details can still get through. Whole conversations are in scope, from a user base above 900 million weekly, many of whom treat the product as a therapist, an assistant or a friend. Anthropic confirmed to 404 Media that it also uses human review. One worker, asked whether users understand this: “I don’t think they would imagine some contractor somewhere […] is analyzing the conversations.” Cox’s framing is that the story dispels a misconception — that these models improve through scraping, engineering talent and scale alone. Paywalled after the opening section, with a free tier available.

Why we logged it. The same structure as the attention transfer, one layer down and in a different currency. The product improves because people supplied something of value — here their conversations rather than their viewing time — without being told what was taken, what it was worth, or who would handle it. It is booked as a cost of goods sold on the contractor side and as nothing at all on the user side. Our interest is the accounting rather than the scandal: if the improvement of a commercial model is partly built from an unpriced contribution by the people using it, the absence of a line item is the finding. We note in fairness that reviewing outputs to reduce sycophancy is a safety practice with a real purpose, and that OpenAI does disclose human review in policy documents most users will never open. Disclosure that nobody reads is the oldest form of unrecognised consent there is.

09 Sep 2026
Research·arXiv (econ.TH)

What limited attention costs, measured in matches never made

Victor H. Aguiar and Dian Hong model large markets in which scarce attention limits how many partners anyone can evaluate, and show the resulting allocation loss is magnified rather than absorbed by the requirement that matches be stable. In an independent random market with average executable degree d, unmatched shares fall at a rate of e−√d under stability against e−d under maximum matching on the same graph. They propose limited-attention deferred acceptance, which preserves valid authorisations and re-engages eligible pairs that a provisional screen had rejected, and test the framework against two datasets: bilateral reports from speed dating, which let them separate missed opportunities from same-graph stability loss, and Chilean school choice, where they document 9,502 applicants accepting higher-ranked or new placements through retained rankings. Open access, posted 9 September.

Why we logged it. The connection to advertising is indirect and we will not overstate it — this is matching theory, not media economics, and no platform appears in it. We log it because of where the cost lands. Formal economics here treats limited attention as a real welfare loss borne by the person doing the attending: 9,502 Chilean schoolchildren received a placement they preferred, not because anything about the schools changed, but because the system stopped throwing away what their attention had already been spent on. Attention is finite; spending it badly has a measurable price; and that price is paid by the individual while the efficiency gain accrues to whoever designed the mechanism. That is the ledger we keep.

09 Sep 2026
Critique·PPC Land

A news app shipped a feed that stops on purpose

MS NOW, the cable network formerly called MSNBC, opened its first paid application on 9 September at $7.99 a month, and the architecture is the story. The personalised feed, called For You, is deliberately finite: once a member reaches the end of the curated content, the app offers a move into a guided-meditation section built with Calm rather than another tranche of material. The company frames the choice explicitly against infinite feeds built to hold attention indefinitely. PPC Land’s reading is the one worth keeping — a finite session has a ceiling on impressions by construction, which cuts against the engagement logic underpinning most advertising-funded media. The product also carries fourteen scheduled live appearances a week with named anchors, a moderated community across five categories, and a 24/7 television feed.

Why we logged it. Someone built the counterfactual and put a price on it. An ending is not a feature a platform funded by impressions can afford, and here it arrives bundled with a subscription that replaces the impression count — which is the whole argument in miniature: the feed stops when somebody other than the advertiser is paying. Caveats, and they are real. The company has not said whether the Membership carries advertising at all, so the ceiling is theoretical for now; the introductory year at $39.99 renews at $79.99; and a product that ends its feed and then offers meditation has arguably relocated the session rather than closed it.

05 Sep 2026
Critique·Pluralistic

The intermediaries’ cut went from 15% to 51% — and the remedy is sealed

Cory Doctorow’s reading of the Brinkema remedy, which requires Google to sell no part of its ad-tech stack and which stays under seal for roughly two weeks, so nobody yet knows what it actually demands. He walks the three-part stack — sell-side platform, demand-side platform, exchange — and argues that owning all three while also acting as advertiser and publisher produces a conflict no behavioural rule can reach: the same firm represents the buyer, the seller and the house. The figure to carry: the intermediaries’ share of ad-industry revenue rose from around 15% before Google and Facebook took 80% of display advertising, to 51% today, moving hundreds of billions of dollars away from publishers and advertisers. He notes a bill to end the arrangement co-sponsored by Elizabeth Warren and Ted Cruz, and calls Google a thrice-convicted monopolist.

Why we logged it. This is the tax stated as a percentage. Fifty-one pence of every pound spent to reach a person’s attention never reaches anyone who made anything — it is taken in transit, by firms the person has never heard of, in an auction that closes before the page finishes loading. Read it against the ruling itself, logged below. Caveat: this is commentary, not reporting, and the sealed remedy may yet be more demanding than its critics expect.

05 Sep 2026
Research·TVision (via PPC Land)

Nine attentive minutes out of two hundred

TVision’s Ad Scoreboard 2.0, announced 25 August, puts a number on the thing the whole market is priced against. Of more than 200 minutes of daily video consumption, roughly 24 minutes represent advertising opportunity — and about nine minutes qualify as attentive advertising time. That is approximately one attentive advertising minute for every 22 minutes of viewing, under five percent of the total. The method is observational rather than modelled: a panel of 5,000 US households fitted with computer-vision sensors, automatic content recognition and digital meters, capturing four variables every second, including who is in the room and at what level of attention. Set beside the Video Advertising Bureau’s finding that video occupies 51% of American leisure time for a fourth straight year — 2 hours 37 minutes daily out of a 5 hour 10 minute total — it gives the extraction a denominator.

Why we logged it. An industry that sells attention has now measured how little of it it actually gets, and published the number. Two hundred minutes of a life go in; nine minutes of watched advertising come out; the whole apparatus of surveillance, auction and intermediation is built to harvest that sliver. Caveats, and they are serious: this is vendor research, not peer-reviewed; TVision was bought by the demand-side platform Viant in May 2026 for $40 million, so a measurement panel owned by a buyer is supplying the data its own buyers cite; and the release gave no measurement window, no demographic breakdown, no MRC accreditation and no statement of whether the nine-minute figure comes from the panel or a third party.

05 Sep 2026
Regulation·PPC Land

Four authorities looked at the ad auction in six days. Nobody had to sell anything.

A week-in-review that assembles four separate decisions into a single pattern. A federal judge in Virginia found on 2 September that Google had illegally monopolised two advertising markets, then declined to take either away, ordering instead that the auction be run differently. The European Commission fined Alphabet €460 million for ranking its own services first, and gave it sixty days to change the ranking rather than the ownership. The FTC and 22 state attorneys general sued Amazon on 31 August over what advertisers were told about a pricing rule, not over the fact that Amazon runs the auction, sells the inventory and reports the results. The General Court in Luxembourg confirmed on 2 September that a browser shipped with Windows need not offer a choice screen, because it is not used enough. Meanwhile the independent demand-side platform that spent a decade arguing it could compound against the walled gardens cut about 575 people, roughly 15% of staff.

Why we logged it. The author’s own line is the one worth keeping: nobody had to sell anything, everybody has to explain something. Four regulators, four jurisdictions, one week, and in every case the remedy landed on conduct and disclosure while ownership of the auction stayed exactly where it was. This is what it looks like when a system is willing to name an extraction and unwilling to stop it. Caveat: trade press aggregating four rulings, each of which deserves its own reading — the €460 million fine in particular is a self-preferencing case, not an ad-tech one.

04 Sep 2026
Research·JAMA Health Forum

Health warnings on social media work — and the framing that works is the one platforms will never choose

A randomised clinical trial with 1,012 teenagers and young adults, run by researchers at Stanford and UNC Chapel Hill, tested whether warnings displayed on social media discourage people from wanting to use it. They do, relative to a neutral control message. More usefully, warnings that named health harms — depression, anxiety — outperformed a warning that only suggested taking a screen-time break without saying why. Open access; DOI 10.1001/jamahealthforum.2026.3014.

Why we logged it. Tobacco took decades to arrive at the label. This is the first properly powered test of the same instrument pointed at an attention product, and it finds that the honest framing is the effective one. It also draws the obvious line: a disclosure regime that lets a platform choose its own wording will get the screen-time-break version, which is the version that does not work. Caveat, and it matters: the outcome measured is stated intention to use, not measured use. Nobody has yet shown a warning label reducing time actually spent.

03 Sep 2026
Regulation·CalPrivacy

California puts a price on a registry error: $200 a day

CalPrivacy’s Enforcement Advisory 2026-01 states that data brokers filing incorrect information in their annual Delete Act registration are liable for a $200 fine for every day the information stays wrong, and notes the Enforcement Division has already brought multiple actions over reporting errors. Executive director Tom Kemp ties this directly to the Delete Request and Opt-out Platform: DROP only functions if brokers report their activities accurately, so an inaccurate registration is not a paperwork failure but a deletion right that silently does not work. The agency brought a separate action against Virginia broker SalesIntel Research earlier the same week, one of more than a dozen against unregistered brokers.

Why we logged it. A regulator that has stopped asking and started metering. The interesting move is not the size of the fine but its shape — per day, per error, accruing without an enforcement action having to be brought first. That is the structural form a Conversion Dividend would need: a continuously running obligation rather than a periodic settlement. Caveat: this is an advisory setting out the Enforcement Division’s reading of an existing statute, not a rulemaking or a decision, and it reaches registration accuracy only — not what brokers do with the data.

02 Sep 2026
Regulation·AdExchanger

The monopoly was proven, and then left standing

US District Judge Leonie Brinkema, who found in April 2025 that Google illegally monopolised the publisher ad server and ad exchange markets, has ruled on the remedy — and rejected the Department of Justice’s proposal that Google divest its sell-side AdX exchange. Instead she ordered most of the parties’ proposed behavioural remedies: making AdX real-time bid amounts available to rival ad servers, deprecating the Unified Pricing Rules so publishers can set different price floors for individual bidders, and abandoning the “first look” and “last look” privileges Google says it stopped using years ago. Both sides must file a jointly proposed final judgment within 30 days. It reads as a near-mirror of Judge Mehta’s search ruling, which also found monopoly and also declined divestiture.

Why we logged it. AdX is the exchange where, in the milliseconds after a page loads, a moment of someone’s notice is auctioned. A court has now established twice that the house rigged that auction in its own favour, and then decided the market is too entangled to unwind. The through-line worth naming: proving the extraction is the easy part. Stopping it is the part the system will not do. Caveats: the full ruling was not public at the time of reporting, the behavioural remedies are not final until the joint judgment is filed, and Google continues to contest the underlying liability finding on appeal.

01 Sep 2026
Critique·a16z

The industry has published its defence, and it is worth reading properly

Andreessen Horowitz ran a guest essay by Ruby Justice Thelot, a professor at NYU, arguing that enshittification “does not describe actual reality and user patterns.” The evidence offered is threefold: social media users grew from 2.5 billion in 2016 to close to 6 billion; average daily time on Instagram rises year on year; and American Customer Satisfaction Index scores for social media moved from 70 to 75 out of 100 between 2020 and 2025. Thelot renames the phenomenon “platform narcissism” — the belief that a platform should keep serving the needs one had at sixteen — attributes the feeling to “mostly geriatric millennials” and the media class, and closes: “if you don’t like it, uninstall them, and move on. Vote with your thumbs.” 404 Media covered it the following day.

Why we logged it. This is the revealed-preference defence in its cleanest published form, and we would rather carry it than route around it. Note what it does: rising time-on-platform is offered as evidence that nothing is being taken. That is the whole disagreement in one move — the hours are the transfer, not the refutation of it, and an argument that measures consent by counting the hours has assumed its own conclusion. Note too what it never mentions: ad load. The essay picks two products that are free at the till and then declines to price what the user pays instead. Caveats in both directions. It is published by a venture firm with holdings across the sector, and five points on a hundred-point satisfaction index over five years is thin load-bearing evidence for “there is no such thing.” But the underlying observation is correct and inconvenient: people do use these products more each year, by choice, and any case for a Conversion Dividend has to survive that fact rather than wish it away.

August 2026

15 entries
31 Aug 2026
Regulation·FTC and 22 states

The FTC sues Amazon over a hidden ad "surcharge" — and the word the company chose to bury it

The Federal Trade Commission and 22 state attorneys general filed a 181-page complaint (Case 2:26-cv-03097, W.D. Wash.) alleging that from 2018-2019 Amazon ran a second-price ad auction, computed the price it would produce, then overrode it with an undisclosed "soft reserve" capped only by the advertiser's own winning bid — while telling some 1.2 million advertisers, most of them small businesses, that winners pay "one penny more than" the runner-up. Internal reports cited in the filing show the share of clicks charged at the advertiser's full bid rising from 4% in late 2020 to 79.1% in 2024. When staff needed a word for the charge, they were told to stop calling it a "surcharge," asked ChatGPT for alternatives, weighed "profit extraction lever" — and settled on "performance premium." Amazon rejects the case, arguing inflation-adjusted cost-per-click stayed flat and that its relevance weighting saved advertisers more than $8 billion.

Why we logged it. The extraction and the euphemism in a single document: a marketplace inserting a hidden charge after the auction has closed, then hunting for language to keep anyone from naming it. This is the attention tax at the point of sale, and the internal record of a company that knew exactly what it was doing. Caveat: an unproven complaint, with much of the sizing arithmetic redacted; Amazon's counter-measurement compares its auction against a bid-only baseline rather than the second-price figure its own training materials promised.

27 Aug 2026
Critique·CPJ & Harvard's Carr-Ryan Center

The advertising machine is a surveillance machine — and journalists are the test case

The Committee to Protect Journalists, with the Carr-Ryan Center for Human Rights at the Harvard Kennedy School, documents how the roughly $1 trillion online-advertising ecosystem doubles as mass-surveillance infrastructure. Software development kits embedded in apps harvest location data; real-time bidding auctions broadcast a user's data to every potential bidder whether or not anyone buys the ad; and because that data is resold after it leaves the device, it becomes near-impossible to know when someone has been tracked. The concrete illustration: New York Times reporters were able to trace a Washington Post journalist using SDK location data drawn from more than 12 million phones. A companion guide sets out defensive measures.

Why we logged it. The machinery built to auction attention is the same machinery that surveils — not two systems but one, here turned on the people whose job is to watch power. It is the clearest recent statement that the ad exchange and the tracking apparatus are the same object. Caveat: the report's risk scenarios are drawn from journalists and human-rights defenders specifically; the ad-tech mechanics it describes are general.

27 Aug 2026
Research·arXiv preprint

The next auction does not sell you a slot. It sells you a sentence.

A mechanism-design preprint by Liu, Zhang, Yu, Ye and Qi proposes the Latent Advertiser Mixture Auction, which places advertiser bids inside the token-generation process of a generative model rather than in predefined advertising slots. The authors prove incentive-compatibility and individual-rationality properties for the mechanism, and report on commercial-search query splits that it improves platform welfare and revenue while maintaining user-facing response quality. In plain terms: the advertiser no longer buys a rectangle beside the answer, it bids on the composition of the answer itself, word by word, and the reader has no boundary to point at. DOI 10.48550/arXiv.2608.27382; open access.

Why we logged it. This is the clearest published description we have seen of where the extraction goes next, and it is written without embarrassment because it is addressed to engineers rather than to the public. Read the welfare accounting closely: the objective function counts the platform and the advertiser, and the user appears only as a quality constraint — a variable to be held above a floor, not a party to the transaction. The banner could at least be identified and ignored. A bid resolved inside the sentence cannot be. Caveats: a preprint, not peer reviewed; the results are proof-of-concept experiments on query splits rather than live users; and “welfare” here is a term of art in mechanism design that measures no user-side cost whatsoever. That last point is the finding, not a limitation of it.

21 Aug 2026
Regulation·California Legislature

California's surveillance-pricing ban reaches the Senate floor

On 21 August, AB 2564 was read a third time and amended in the California Senate, moving toward a floor vote before the legislative session closes on 31 August. The bill would prohibit "surveillance pricing" — a customised price for a consumer or group "based, in whole or in part, on personally identifiable information collected through electronic surveillance technology" and determined through statistical modelling, data analytics or AI — and it also reaches random price variation shown to different online shoppers. It is framed as furthering the California Privacy Rights Act. Business opposition is organised: the California Chamber of Commerce ran a "myths vs facts" rebuttal on 5 August.

Why we logged it. The same dossiers that decide what you see now set what you pay — a legislature deciding whether a price built from your surveilled data is lawful at all. It picks up the AB 2564 thread we logged in July at the point it reaches the Senate floor. Caveat: a state bill still in progress, not a law; a grocery-and-retail pricing measure, and a pass, amendment or failure could all still happen before month's end.

19 Aug 2026
Regulation·FTC (via Consumer Finance Monitor)

The FTC puts surveillance pricing on notice — as a third state bans it outright

On 19 August the FTC issued for public comment a Proposed Enforcement Policy Statement Regarding Personalized Pricing, warning that a business using consumers’ personal data to set individualised prices without adequate disclosure may violate Section 5 of the FTC Act. It is the first federal move to meet the state wave this feed has tracked through Connecticut’s disclosure-label law: New Jersey has now become the third state to ban surveillance pricing for groceries outright, after Maryland (first, April 2026) and Connecticut, while New York took the disclosure route. Congressional oversight runs in parallel — a House Oversight AI-pricing inquiry opened in March, an Energy and Commerce grocery inquiry in May — and California AG Bonta’s January investigative sweep continues.

Why we logged it. The same argument this site makes about attention, arriving at the checkout: value a company extracts by watching a person is turned back on that person as a higher price, and the federal consumer regulator has now said, provisionally, that doing so silently may be unlawful. Note what it does not say — that doing it openly is fine. Caveats: a proposal out for comment, not a rule or an action; it targets non-disclosure rather than the practice; and most state bills remain unpassed and contested.

18 Aug 2026
Critique·OpenAI

Advertising moves inside the chatbot — ChatGPT Ads goes international

OpenAI confirmed on 11 August that ChatGPT Ads has launched in the United Kingdom, Mexico, Brazil, Japan and South Korea, and on 18 August that it is expanding across Europe — the rapid internationalisation of a US test that only began in February 2026. Ads run for logged-in adult users on the Free and Go tiers and, by OpenAI's own account, are matched using "the topic of your conversation, your past chats, and past interactions with ads." OpenAI stresses that answers stay independent, that advertisers receive only aggregate performance data, and that personalisation is user-controllable (and, in the EU, consent-gated).

Why we logged it. The surveillance-advertising model — attention captured, profiled and sold — is now being rebuilt inside the conversational interface hundreds of millions of people increasingly use in place of search, a surface with far more intimate signal than a feed. The tax follows the attention wherever it goes. Caveat: the company's own framing of a product still in test; it is evidence of the model's expansion, not of a specific abuse. Read it alongside the independent audit logged below.

17 Aug 2026
Research·Pediatric Exercise Science

More childhood screen time, better cognitive processing — a result we are logging against ourselves

An eight-year follow-up of Finland’s PANIC study (124 girls, 136 boys, average age 15.8) found that greater screen time accumulated from childhood was associated with better cognitive processing in adolescence on the CogState battery — a result the researchers themselves flag as cutting against common assumptions. Their reading is that what a child does on a screen (problem-solving, creativity, active thinking) may matter more than how long the screen is lit, and that the aim is balance between physical activity and mentally engaging screen use rather than minimisation. DOI 10.1123/pes.2025-0083.

Why we logged it. We are logging this because it disagrees with us. Read against the Millennium Cohort study logged this month, it makes the point better than either does alone: “screen time” is the wrong unit. Our quarrel is with engineered compulsion — products built so that leaving is harder than staying — not with minutes in front of a display, and an honest ledger keeps that distinction sharp. Caveats the authors stress: small sample, associational not causal, sex-dependent and mixed physical-activity results, and an explicit call for intervention studies before anyone infers cause.

12 Aug 2026
Critique·404 Media

A searchable map of the advertising surveillance pipeline

DecryptAds, built by Zach Edwards and reported by Joseph Cox, aggregates three types of file the ad industry publishes for anti-fraud compliance — ads.txt, app-ads.txt and sellers.json — into the largest publicly searchable corpus of advertising supply-chain data yet assembled, roughly 284 million records at launch. Its purpose is to make legible the pipeline by which real-time bidding turns an ordinary app open into a surveillance record. The mechanism is the one this site keeps returning to: when an auction fires, the bid request — device ID, precise GPS, browsing profile, app in use — is broadcast to every potential bidder before anyone wins, and nothing requires the losers to delete it. A DHS document obtained under FOIA is, per the report, the first time a federal agency explicitly acknowledged buying RTB-sourced location data: a CBP pilot from 2019–2021 drawing on SDKs in tens of thousands of apps. Wider context here.

Why we logged it. The extraction of attention and the surveillance of a person are not two systems. They are one auction. A firm can join real-time bidding not to buy ads but to collect the data every participant receives by design, sell it to a broker, and let the broker sell it to the state — and until now there was no way for an outsider to trace the chain. Caveats: much of the framing is about Fourth Amendment law rather than the advertising economics we foreground; the tool maps the supply chain, it does not by itself prove any specific broker sold to any specific agency.

11 Aug 2026
Regulation·Connecticut (SB 4)

Surveillance pricing arrives at the checkout with a warning label

From 1 October, Connecticut's SB 4 bars retail sellers and third-party delivery services from setting customised prices on personal data — and requires any other business using a "price setting device" to display the words "THIS PRICE WAS INCREASED BY A PRICE SETTING DEVICE USING YOUR PERSONAL DATA." It is one route among many: more than 60 pricing bills are moving across some 33 states, with Maryland the first to flatly prohibit the practice. Enforcement is widening beyond the statutes — California's Attorney General has opened an investigative sweep treating surveillance pricing as a purpose-limitation violation of the CCPA, on the theory that data handed over for one purpose cannot be repurposed to decide what a person pays. The bills are drafted to close the obvious escape: pricing a behavioural "group" still counts.

Why we logged it. The value extracted by watching a person, turned back on them at the till — and at least one state now forces that transfer to announce itself on the price tag. It carries forward the California AB 2564 thread from a jurisdiction that has already enacted. Caveat: this is Connecticut specifically; most of the 60-plus bills are unpassed, and business groups are contesting both the disclosure and prohibition models. AG enforcement of SB 4 begins 1 July 2027.

07 Aug 2026
Research·arXiv

Auctioning attention on social networks

The paper proposes building a social feed as an auction — users bid for one another's attention — with a mechanism that weighs producers, consumers, the platform operator and social welfare, in place of a recommender that ranks for engagement alone. On real and synthetic networks it reports higher "producer welfare" (36.3% and 31.4%) and, more to the point, "consistently more equitable distributions of attention" than the algorithmic baselines. Authors Andy Lee and Hari Sundaram state the status quo plainly: producers optimise for the algorithm, consumers absorb the externalities — polarisation, misinformation, overconsumption — and the platform maximises engagement.

Why we logged it. A formal treatment of the thing we keep asserting — attention as a scarce, priced, allocable resource, distributed today by an opaque recommender in the platform's interest, and worth asking who should get to claim it. Caveat: an unrefereed preprint proposing a mechanism, not evidence about any deployed system; "producer welfare" here means content creators.

06 Aug 2026
Regulation·First Judicial District Court, New Mexico

A New Mexico court orders Meta to fund $567 million in youth-harm remedies — and refuses to touch the engagement features

In the remedies phase of State of New Mexico v. Meta Platforms Inc., Chief Judge Bryan Biedscheid found Meta's platforms a "public nuisance" that significantly contributes to a state teen mental health crisis, and ordered Meta to pay $567 million into an abatement fund for treatment, awareness and monitoring — on top of the $375 million a jury awarded in March. But the Attorney General had asked the court to force removal of infinite scroll and autoplay, and the judge declined: "regulation of industry-wide features that contribute to problematic use, such as infinite scroll and autoplay, require legislative or executive branch action," and ordering feature changes could run afoul of the First Amendment and Section 230.

Why we logged it. The most aggressive youth-harm remedy in the US to date funds the treatment of the symptom and leaves the extraction mechanism — the engagement architecture itself — legally untouched. The money is paid; the machine keeps running. Caveat: a first-instance state ruling, and Meta says it will appeal.

06 Aug 2026
Critique·Pluralistic

Doctorow — social media without socializing

Cory Doctorow's argument is that Zuckerberg profits from a fact he resents — "you love your friends more than you hate him," and those friends are what keep you on Facebook and Instagram. The trouble, for Zuck, is that friends "just want to be your friends"; they do not "maximize your engagement" so he can show you more ads. Doctorow traces the workarounds: tuning the algorithm to swap friends for flamewar-baiting trolls, recruiting engagement-farming "theater kids" on an intermittent-reward revshare, and now flooding the feeds with chatbots — "social media without socializing," bots "tuned to keep you scrolling no matter what" that "will never demand anything of Meta."

Why we logged it. A clean statement of the core antagonism — engagement optimisation is structurally opposed to the human connection it feeds on, and the endgame is to replace the people with machines that extract attention without friction. Caveat: a polemical blog essay, not reporting; it links its sources throughout.

05 Aug 2026
Research·arXiv / AIES 2026

The first audit of ads inside ChatGPT finds the tax falls hardest on the poor

The first empirical study of advertising inside a large language model's interface. Using a "sock puppet" audit — 91 automated accounts signalling three racial/ethnic groups and three income terciles through geolocation and prompt cues — Emma Lurie, Ro Encarnación, Sorelle A. Friedler and Danaé Metaxa collected over 3,000 ads from 186 advertisers across 335 prompts on ChatGPT. Two findings stand out: accounts began receiving ads 14 days after creation, and lower-income accounts, regardless of race, were more likely to be shown ads. In this first phase the ads skewed to consumer goods and stayed visually separated from the model's answers — a separation the authors expect to erode as ads are integrated more deeply.

Why we logged it. Independent measurement, in a peer-reviewed venue, of the attention tax arriving inside the interface people now use to think — and of who pays it most. Caveat: a preprint, accepted to AAAI/ACM AIES 2026; it audits one platform in an early test phase with simulated rather than real users, and reports association, not a causal effect of income on ad delivery.

04 Aug 2026
Research·European Journal of Public Health

Digital engagement and adolescent wellbeing — the UK Millennium Cohort Study

Using 24-hour time-use diaries from 3,288 fourteen-year-olds — 144 consecutive ten-minute slots each — the authors ran latent profile analysis to find real daily patterns rather than treating "screen time" as a single number, then linked those patterns to later psychological wellbeing. Three profiles emerged: "frequent digital users" (21%, heavy gaming, more male and lower-income, poorest wellbeing), "physically active, moderate digital users" (59.7%, best wellbeing), and "in-person socialisers, low digital users" (19.3%, intermediate). Their conclusion: balance across the whole day, not digital time in isolation, is what tracks wellbeing — digital use matters largely through what it displaces.

Why we logged it. A careful, peer-reviewed corrective to crude "screen time bad" accounting — the harm sits in the trade-offs the extraction economy engineers people into, not a raw hours-on-glass count. Caveat: observational, so associations rather than causation; a wellbeing study, not a study of extractive design as such. Published in the European Journal of Public Health.

03 Aug 2026
Research·Nature Human Behaviour

A peer-reviewed number for what early social media costs a child's schooling

Drawing on 5,227 students across 28 northern Italian schools, matched to standardised test records, Marco Gui and colleagues find that children who opened a personal social media account at 11-12 scored around 0.22 standard deviations lower in Italian and 0.18 lower in maths by age 13-14 — the maths gap widening to 0.27 by 15-16. The team put 0.2 SD at roughly six months of schooling. Opening an account later narrowed the gap in a clean dose-response pattern, and the association ran through how often pupils checked their phones. There was no effect on English scores. A research briefing accompanies the paper.

Why we logged it. A citable, peer-reviewed figure for a cost we usually argue in the abstract: attention extracted early leaves a measurable, durable deficit in what a child learns. Caveat: the study is observational and cannot prove causation, social media use was self-reported, and the authors themselves ask that it be read cautiously.

July 2026

20 entries
30 Jul 2026
Critique·Pluralistic

Doctorow — The stupidest imaginable excuses for surveillance pricing

Cory Doctorow's account of surveillance pricing as AI's one indisputable commercial success: feed a model behavioural data, let it cluster your customers, and let it run continuous experiments to find the maximum markup each group will stomach — "it will simply find the most vulnerable people and rip them off the most." He cites the FTC's finding that profiled new parents are shown higher-priced thermometers at night, Instacart's markups of up to a third, and Delta's abandoned per-seat scheme. The goal, he argues, is to shift all consumer surplus to the platform: you are "gouged according to your ability (to pay) and charged according to the desperation of your need."

Why we logged it. The extraction thesis extended from attention to price — the same dossiers that target what you see now reprice what you need. Caveat: a polemic in support of California's AB 2564; it pairs with the EFF letter logged below.

29 Jul 2026
Regulation·US Federal Trade Commission

The FTC and two states sue Hims & Hers over health data shared with ad platforms

The FTC, joined by Utah and California, filed a federal complaint alleging the telehealth provider shared consumers' sensitive health information — conditions disclosed in medical intake forms — with advertising platforms including Meta and Snap, via customer lists and tracking technologies that relayed visitors' on-site actions, while promising patient privacy. Alongside sit classic dark-pattern claims: charging almost immediately after intake despite promising a provider consultation first, auto-enrolment into recurring subscriptions, and a cancel button hidden behind an "add/remove items from order" flow.

Why we logged it. Extraction on both ends of the funnel — the most intimate category of personal data monetised into ad-targeting infrastructure, while interface design stops the customer leaving. Caveat: allegations in a filed complaint, not findings; the case will be decided by the court.

28 Jul 2026
Regulation·New York Office of the Attorney General

New York finalises the SAFE for Kids Act — addictive feeds off by default for under-18s

New York released the final rules implementing the Stop Addictive Feeds Exploitation (SAFE) for Kids Act, which requires social-media companies to switch off algorithmically personalised feeds and overnight notifications (12am–6am) for users under 18 unless a parent consents. The rules define the target precisely: an "addictive feed" is an endless, personalised stream built on data the platform gathers, "designed to encourage a user to continue to use and return to a platform," and the law bites on platforms where users spend at least 20 percent of their time on such feeds. Under-18s who decline the feed are not cut off — they see the same content from accounts they follow, in a set order such as chronological. The rules take effect 25 January 2027.

Why we logged it. A state legislating against the feed mechanism itself — the engineered capture of attention and return-visits — rather than the content inside it; Governor Hochul's line, "our kids' mental health isn't for sale," names it as an extraction. Caveat: a child-protection measure, not an attention-market instrument; it applies only to minors, and its age-assurance requirements raise their own privacy questions.

28 Jul 2026
Regulation·Electronic Frontier Foundation

San Francisco stalls its backing of California's surveillance-pricing ban

California's AB 2564 (Ward) would ban "a customized price for a good for a specific consumer or group of consumers based, in whole or in part, on personally identifiable information collected through electronic surveillance" — with carve-outs for cost-based differences, retention discounts, and uniformly posted discounts. The San Francisco Board of Supervisors was set to pass a resolution of support, then stalled the vote after a single Chamber of Commerce letter recycling objections the bill's text already answers. EFF's rebuttal letter restates the principle: privacy is a right, not a currency, and a fair price should not depend on submitting to surveillance.

Why we logged it. A working ban on converting harvested data into higher prices, met by organised lobbying at the municipal level — a live study in how extraction defends itself. Caveat: advocacy around a pending bill, not a regulator action; the resolution fight is procedural, and the bill's fate sits with the state legislature.

24 Jul 2026
Regulation·European Commission

The Commission tells TikTok to stop feeding minors into the For You Feed

The European Commission sent TikTok preliminary findings that its minors' account settings breach the Digital Services Act. On TikTok, minors can set an account to "public," and the default lets content from 16–17 year-olds be recommended to any user — including people with no TikTok account — through the For You Feed. The Commission's view is that a minor's content should by default be visible only to contacts the minor has accepted, and that TikTok "should refrain from recommending minors' content to other TikTok users through the For You Feed."

Why we logged it. A regulator naming the recommender itself as the mechanism at fault — treating an engagement-maximising default as a harm to be switched off, the same move made against Meta on addictive design. Caveat: a preliminary finding, not a decision, and framed around child safety and privacy rather than attention monetisation; TikTok's separate advertising-transparency issues were closed by binding commitments in December 2025.

23 Jul 2026
Regulation·European Commission

The Commission issues its first DMA fines against Google — €890 million

The European Commission took two decisions finding Google in breach of the Digital Markets Act, its first fines under the regime: €460 million for self-preferencing its own shopping, hotel, transport and sports results above third parties in Search, and €430 million for anti-steering restrictions that stopped Google Play developers directing users to cheaper channels off-platform. Google has been ordered to bring both practices to an end, with the DMA's escalating penalties available for repeat non-compliance.

Why we logged it. A regulator naming and pricing the mechanics of platform self-dealing — the gatekeeper taking a cut of the attention and the transactions it did not earn. Caveat: the case concerns search-ranking prominence and app-store steering, not advertising revenue as such, and a one-off fine at this scale functions closer to a price than a deterrent.

23 Jul 2026
Critique·Pluralistic

Doctorow — California's privacy obstacle course

Cory Doctorow walks through the sign-up flow for California's new Delete Request and Opt-out Platform (DROP): a Login.gov identity, redundant driver's-licence photos, hunting down your mobile advertising identifier, and repeated re-verification of a phone and email the system has already confirmed. His argument is that the process assumes everyone wants to be surveilled and makes leaving laborious — while a data broker's cost to open a file on you is "Step one: Exist. Step two: There is no step two."

Why we logged it. Consent theatre described precisely — the burden of not being surveilled dumped on the surveilled, when it belongs on the extractor. Caveat: a polemical blog post; Doctorow still urges readers to complete the process, because the underlying harm is real.

21 Jul 2026
Critique·Pluralistic

Doctorow — the cookie banner as a rigged market stall

Cory Doctorow takes John Gruber's coinage "dickover" — the modal popover that deliberately obscures a page to force an unwanted interaction (cookie consent, app nag, terms of service) — and reads Europe's cookie-banner epidemic as malicious compliance with the GDPR. Surveillance firms stretch "legitimate interest" to cover "spying on you because our terms of service say we will," he writes, yet insist it cannot possibly cover remembering that you already told them not to. He ends on the market metaphor directly: this is "a very curious sort of market" where "as soon as you stop to browse someone's wares, the stallholder gets to reach into your pocket and clean out your wallet."

Why we logged it. Consent theatre read as a rigged exchange — prices "set unilaterally and extracted from anyone unwise enough to cross the threshold," with ad-blocking recast as the user bargaining back. Caveat: a polemical post whose back half is a practical how-to (Firefox, Reader Mode, Adblock Plus); the quotable framing is in the opening and the close.

17 Jul 2026
Research·JAMA Network Open

Problematic social-media use comes before the attention problems, not after

A prospective cohort study across five annual waves of the Adolescent Brain Cognitive Development study (11,286 US youths) finds that, within individuals, years of higher-than-usual problematic social-media use were followed by elevated parent-reported ADHD symptoms (β ≈ 0.08), with the effect stronger in boys. The reverse path — ADHD symptoms driving later use — was small and sporadic.

Why we logged it. Large-sample, within-person evidence that compulsive platform use appears to precede attentional difficulty rather than merely coincide with it. Caveat: the authors stress the population-level effect sizes are small, and problematic use is self-reported.

15 Jul 2026
Regulation·Ofcom

Ofcom counts the age checks — 69 million of them — and finds the harm largely intact

Ofcom's first statutory Use of Age Assurance Report finds that in the first six months of the Online Safety Act's child-protection duties (July–December 2025), 69 million age checks were completed across the 32 services examined — a 23-fold increase — with facial age estimation and photo-ID matching the dominant methods. The judgment is mixed: checks deterred access on pornography sites that implemented them, but Ofcom says it has "not seen a material reduction in the volume of harmful content being surfaced to children" and that age assurance "is not delivering the intended outcome." No single method eliminates circumvention; the regulator calls for layered protections.

Why we logged it. A regulator putting a number on a compulsory friction imposed at national scale, then honestly assessing whether the extraction it was meant to interrupt actually fell — the discipline attention markets have never been subjected to. Caveat: this is child-safety regulation, not attention-market regulation; the relevance is by analogy of method, not subject.

15 Jul 2026
Research·The BMJ

The BMJ — treat social media like tobacco, and expect it to adapt like tobacco

Writing in The BMJ, Amrit Kaur Purba and colleagues argue that adolescent social media restrictions — including outright bans — should be treated as complex-systems interventions rather than isolated behavioural policies, because they sit inside a system of families, schools, governments and commercial actors. They frame social media as a commercial determinant of health and draw on the tobacco and alcohol playbooks to predict how platforms will adapt after regulation — politically, scientifically, technologically and economically — including redefining what counts as "social media" so as to fall outside the new rules.

Why we logged it. The extraction case in a public-health register — the industry modelled as a rational actor defending a revenue stream, not a neutral service, and certain to route around any rule that leaves the revenue intact. Caveat: an analysis article rather than new empirical data, and its subject is the regulation of minors rather than the attention transfer itself.

14 Jul 2026
Critique·Verfassungsblog

Morgan — Press-release governance: the legality of the Commission's addictive-design findings

Julian Morgan (Weizenbaum Institute) argues the Commission's preliminary findings against Meta rest on a contestable legal basis — Articles 34–35 DSA were drafted deliberately vague, no guidelines on "addictive design" exist, and the leap from "mitigate systemic risks" to "autoplay must be off by default" may not survive the legality principle in court. He reads the findings instead as strategy (leverage toward negotiated design commitments) and symbolism (performing digital sovereignty on a news-cycle timescale, with any court defeat arriving years after attention has moved on).

Why we logged it. The sharpest sentence is the one aimed at the frame itself: the psychologically framed addiction harm does "useful representational work" while leaving "comparatively invisible the underlying business model of advertisement-funded data extraction." Even flagship enforcement, in this reading, regulates the symptom and shields the revenue model. Caveat: a legal-scholarship blog post, sympathetic to acting against addictive design but critical of the mode.

10 Jul 2026
Regulation·European Commission

The Commission preliminarily finds Instagram and Facebook's addictive design in breach of the DSA

The European Commission has preliminarily found Meta in breach of the Digital Services Act over the addictive design of Instagram and Facebook — the investigation focused on infinite scroll, autoplay, push notifications, and the platforms' highly personalised recommender systems. The finding is twofold: Meta did not adequately assess the risks its design posed to the physical and mental wellbeing of users, including minors and vulnerable adults, and the mitigations it deployed failed to address them. It is the second DSA addictive-design case after TikTok (February 2026); non-compliance can carry fines of up to 6% of global annual turnover.

Why we logged it. The biggest regulatory landing on this site's core argument since it launched — a regulator asserting, for the second time in five months, that the engagement architecture is itself the violation, not the content it carries. Caveat: preliminary findings, not a decision. Meta has the right of defence, no fine has been imposed, and Meta says the findings do not recognise its Teen Accounts work.

08 Jul 2026
Regulation·New York City

New York City writes click-to-cancel into law — and moves on junk fees

New York City finalised a Click-to-Cancel rule making it, by the administration's account, the first municipality in the nation to require that cancelling a subscription be as easy as signing up — subscribe online, cancel online — paired with a separately proposed all-in pricing rule banning hidden "junk fees." The city projects the cancel rule alone will save New Yorkers up to $162.5 million a year, with penalties starting at $350 per violation; it takes effect 1 October 2026.

Why we logged it. A government treating designed friction — the hard-to-find cancel button, the fee revealed only at checkout — as a compulsory extraction of money and time, and pricing it. The same move this site makes about attention, applied where the transfer is denominated in dollars. Caveat: the junk-fee rule is a proposal in its comment period; both are municipal rules stepping into the gap left when a US appeals court vacated the FTC's federal click-to-cancel rule in 2025.

02 Jul 2026
Regulation·Court of Justice of the EU

Google's €4.125bn Android fine is final — and the Court says defaults extract

The Court of Justice dismissed Google and Alphabet's appeal against the 2022 Android judgment, fixing the fine at €4.125 billion and closing the case permanently. The reasoning matters more than the number: the Court confirmed the finding of "a status quo bias in favour of pre-installed apps," and that Google "had not demonstrated that user preferences or the alleged quality of their services alone accounted for the behaviours observed." No counterfactual analysis was required.

Why we logged it. A court accepting, as settled EU law, that default placement produces user behaviour that preference does not explain — the legal shadow of the claim this site makes about attention, stated in the language of Article 102. Caveat: the conduct concerns search distribution between 2011 and 2018, not advertising or attention markets, and the remedy flows to competitors, not to the people whose defaults were set.

01 Jul 2026
Critique·Pluralistic

Doctorow — Technocarcinization: every platform is evolving into Facebook

Cory Doctorow borrows "carcinisation" — the way unrelated species keep evolving into crabs — to describe how Apple, Google, and Meta are all drifting into the same corner of his old grid: maximally surveilling and maximally controlling. Apple, once the "we don't spy" walled garden, now runs on-device ad targeting; Google, once the company that sent you out to the open web, walls searchers inside AI summaries and nerfs Chrome's ad-blocking. His claim is that the convergence is driven not by the people in charge but by a policy environment where surveillance and lock-in are the traits that survive.

Why we logged it. The extraction case stated as evolutionary pressure — "if you're not paying for the product, you're the product" is only half right, Doctorow writes; "even if you pay for the product, you're the product." Caveat: a polemical blog post, and the grid is a rhetorical device, not a measurement.

01 Jul 2026
Regulation·Connecticut General Assembly

Connecticut bans engagement-extension design — and targeted ads — for minors

The 2025 amendments to the Connecticut Data Privacy Act (SB 1295) take effect on 1 July 2026. They impose a categorical bar on processing a minor's personal data for targeted advertising or any sale, regardless of consent — and, separately, prohibit using "any system design feature to significantly increase, sustain or extend any minor's use" of a service. Signed by Gov. Lamont in June 2025; the new impact-assessment duties attach to processing created on or after 1 August 2026.

Why we logged it. A US state writing the engagement-hold mechanism into statute and switching it off for minors — the design itself named as the harm, not merely the data taken or the hours spent. Caveat: a state law, limited to minors, and consent survives as a route for some essential-service profiling.

— Jul 2026
Regulation·California Privacy Protection Agency

California's one-request data-broker deletion becomes enforceable

Under the 2023 Delete Act, California's Delete Request and Opt-out Platform (DROP) lets a verified resident file a single request that requires every registered data broker to delete their personal information and stop selling or sharing it — replacing the old broker-by-broker opt-out. The platform has been live since 1 January 2026; from 1 August 2026 brokers are legally obliged to honour these centralised requests, with a 90-day window to delete and confirm.

Why we logged it. The first working, government-run mechanism to claw back extracted data at scale — the extractor, not the extracted, made to act. Caveat: it is opt-out rather than opt-in, so it still treats surveillance as the default state, and the sign-up burden is heavy (see Doctorow, same week).

— Jul 2026
Research·Humanities and Social Sciences Communications

Excessive screen time, mental health, and the mediators in between

Now peer-reviewed in Nature Portfolio's Humanities and Social Sciences Communications, an analysis of 50,231 US children and adolescents (National Survey of Children's Health, 2020–2021) finds four or more hours of daily screen time associated with higher risks of anxiety, depression, conduct problems, and ADHD. Structural-equation modelling locates most of the association in what screen time displaces: physical activity is the strongest mediator (roughly 31–39% of the effect), followed by irregular bedtime and short sleep.

Why we logged it. The harm is real and large-scale, but it runs mostly through displaced sleep and movement — a more defensible claim than screens acting directly on the mind. Caveat: cross-sectional and parent-reported, and it does not separate advertising-driven feeds from other use.

— Jul 2026
Regulation·UK Information Commissioner's Office

The counter-current: the ICO loosens consent for "low-risk" advertising

While the regulatory current of 2026 mostly runs toward constraining platforms, the UK's data regulator has been moving the other way on consent. The Data (Use and Access) Act 2025 introduced new categories of cookies exempt from PECR's consent requirement (in force 5 February), and in May the ICO advised the government on creating a "consent-free route" for low-risk online advertising — a posture that earlier saw it accept Meta's "consent-or-pay" model as capable of lawfulness.

Why we logged it. The near-photographic negative of the Conversion Dividend — rather than declaring a share of ad revenue owed back, a regulator is widening the circumstances in which attention can be taken without a consent step at all. Read against Ofcom's report of the same fortnight, the UK state is pulling in two directions at once. Caveat: UK-specific, framed as reducing compliance friction for "low-risk" processing, and a composite of several 2026 developments rather than a single dated action.

June 2026

9 entries
29 Jun 2026
Regulation·noyb / US Supreme Court

The Supreme Court strips the FTC of independence — and the EU–US data deal loses its foundation

In Trump v. Slaughter (29 June), a 6–3 Supreme Court upheld the President's power to dismiss an FTC commissioner and held the agency's statutory independence unconstitutional. Max Schrems' noyb argues the consequence is structural: the EU's adequacy decision for US data transfers leans on the "independent" FTC 259 times, and EU treaty law requires independent oversight — so the legal basis of the EU–US Data Privacy Framework has collapsed. noyb has asked the Commission to withdraw the deal in an orderly way and says it will file suit.

Why we logged it. The accountability layer meant to police cross-border surveillance dismantled in a single ruling — the oversight the whole data pipeline was supposed to answer to, gone. Caveat: no immediate effect. The Commission's decision stands until it repeals it or the CJEU annuls it, and a court challenge could take two to three years.

23 Jun 2026
Critique·Pluralistic

Doctorow — "Age verification" is not verification, it is mass surveillance

Cory Doctorow argues there is no such thing as "age verification" for the internet — what is sold under that name is mass surveillance, "so invasive and pervasive that it makes the ad-tech industry's commercial surveillance look like some kind of cypherpunk darknet pirate utopia." An age-check mandate would require everyone online to submit to fine-grained tracking of all their activity, handing the surveillance-advertising industry a world where avoiding tracking is illegal. The harms to children people fear, he writes, "all start with surveillance" — kids cannot be targeted or funnelled toward harmful content without the commercial spying that primes the funnel.

Why we logged it. The child-safety frame turned inside out — the surveillance sold as protection is the mechanism of extraction, and Doctorow ties it straight to the wallet: the data "is being used to raise the prices you pay and lower the wages you're offered." Caveat: a polemical post, and its sharpest claims — VPN bans, ICE misuse — are where the argument leads, not events that have happened.

15 Jun 2026
Regulation·United Kingdom

Britain moves to ban under-16s from social media

Prime Minister Keir Starmer announced that the UK will bar under-16s from a range of social platforms — Snapchat, TikTok, YouTube, Instagram, Facebook, and X — while exempting messaging services such as WhatsApp and Signal, and will block livestreaming and stranger-contact for under-16s. The government, calling it a "landmark" move, will bring measures to Parliament before Christmas, with protections expected in force next spring, putting the onus on platforms to keep children off and threatening large fines. A national consultation drew more than 116,000 responses; 90% backed a minimum age of 16.

Why we logged it. A national government treating access to the engagement-maximising feed itself as the harm — not the content within it, nor the hours spent. Caveat: announced, not yet law, and enforcement is the open question — Australia's parallel under-16 ban, in force since December 2025, is already struggling, and age verification carries its own privacy and free-expression costs.

13 Jun 2026
Regulation·European Commission

The Digital Omnibus keeps moving — and the fight is over who may override your "no"

The Commission's Digital Omnibus Regulation (proposed 19 November 2025) would mandate a single-click reject with equal prominence to "accept," bar re-prompting for the same purpose for at least six months after a refusal, and require sites to honour browser- or device-level privacy preference signals. One carve-out has drawn fire: media providers whose revenue depends primarily on advertising would be exempt from honouring those signals — a carve-out France's Alliance Digitale is fighting, calling (21 May 2026) to delete the centralised cookie mechanism and the six-month re-consent ban.

Why we logged it. The mechanism by which refusal is registered — and who is permitted to override it — is the live battleground. A durable, machine-readable "no" is the precondition for any honest accounting of extracted attention. Caveat: a proposal, not law; GDPR and ePrivacy remain in force, and if adopted it phases in over up to four years.

06 Jun 2026
Research·arXiv preprint

Luo, Yao & Zhang — a platform's "sleep reminder" campaign increased late-night use

A large-scale field experiment on a short-video platform tested a "sleep reminder" campaign meant to reduce late-night use. It did the opposite: late-night engagement rose 14.75%, overall usage rose 2.18%, and the effects persisted for weeks after the campaign ended. The mechanism is the finding — the intervention acted as forced exploration, revealing latent demand the recommender would never have surfaced on its own, and the algorithm updated its policy on that data, "reinforcing the very engagement loops the campaign aimed to mitigate."

Why we logged it. Close to a controlled demonstration of this site's core claim: an engagement-optimising system metabolises even wellbeing interventions into more extraction. Any "digital wellbeing" feature evaluated as a static nudge, while the algorithm underneath keeps learning, is measured wrong by construction. Caveat: an unrefereed preprint, and one unnamed platform.

— Jun 2026
Critique·New Media & Society

Lee — Brain rot: cognitive decomposition as a structural externality of attention assetization

Hera Hyeonseo Lee (Binghamton University) theorises "brain rot" not as cultural decline but as a structural condition: in the post-2008 turn from material expansion toward cognitive extraction, platforms "assetize" attention to stabilise their speculative valuations, inducing a rewiring that degrades the capacity for sustained thought. The argument is built from SEC filings and earnings transcripts of Meta, Alphabet, Microsoft, and Amazon — tying the architecture of corporate finance directly to systemic cognitive shifts.

Why we logged it. The extraction tied explicitly to share-price logic, and the resulting cognitive damage named as an externality — a cost the public absorbs so the firms can hold their valuations. The Conversion Dividend's moral premise, in political-economy terms. Caveat: a single-author theoretical article reasoning from financial documents, not a measurement of individuals' cognition.

— Jun 2026
Research·PNAS

Differentiation drives the erosion of positivity on social media

Analysing 2.05 billion comments across 2,150 Reddit communities, Hongkai Mao and colleagues find that discourse becomes measurably more negative over time — within individual threads and across a community's history — and that the trend is mediated by the semantic uniqueness of comments. The mechanism: users compete to say something new, negative information is more heterogeneous and counter-normative than positive information, so as a conversation fills up, the cheapest remaining way to be distinctive is to be negative. The negativity emerges without anyone intending it.

Why we logged it. The extraction argument in its purest form — a structural property of a scarce-attention contest producing a harm that no individual participant chose. The nastiness is not the users; it is the contest they are made to compete in. Caveat: observational analysis of one platform, establishing a robust pattern and plausible mechanism rather than a controlled causal test — and it concerns user-generated negativity, not advertising or platform design directly.

04 Jun 2026
Critique·Pluralistic

Delusion as a service

Cory Doctorow likens the internet — and now sycophantic chatbots — to Disney's "Mission: Space" centrifuge: benign for almost everyone, but a machine for surfacing rare, dormant vulnerabilities in the few it harms. Applied to "AI psychosis," he argues chatbots supply "delusional reinforcement as a service, on tap, 24/7," and that even if they only amplify rather than cause delusion, mass deployment guarantees mass harm absent mitigation.

Why we logged it. Extends the extraction case from attention to psychological fault lines — engagement-maximising systems, run at scale, inevitably reach the people least able to withstand them. Caveat: its subject is AI-driven delusion, not advertising or the attention-tax framing specifically.

03 Jun 2026
Regulation·Illinois General Assembly

Illinois passes a bill banning algorithmic feeds for minors

The General Assembly passed the Children's Online Social Media Safety Act (HB 5511) unanimously — 57-0 in the Senate, 113-0 in the House — and Gov. JB Pritzker has said he will sign it. The law bars platforms from using a minor's viewing history or on-device data to rank their feed: feeds for under-18s may show only what the user searched for, requested, or follows, and notifications are prohibited between 10pm and 7am. Enforcement runs through the Attorney General, with fines of $2,500–$7,500 per child.

Why we logged it. The cleanest legislative statement yet that the engagement-ranked feed itself is the harm — not merely the content inside it. For minors, the extraction mechanism is simply switched off. Caveat: it takes effect in 2028, and tech lobbyists have flagged First Amendment challenges.

May 2026

10 entries
27 May 2026
Research·Nature

Brady et al. — the less extractive feed that users liked just as much

A large-scale field experiment altered the feed-ranking algorithm on Bluesky during the 2024 US presidential election. A "diversified extremity" algorithm — making feeds more representative of the actual population of users — reduced exposure to polarising content and produced more accurate perceptions of social norms, without reducing users' enjoyment of the platform. Both the experimental feed and a plain reverse-chronological feed surfaced less uncivil content than standard engagement-based ranking.

Why we logged it. The counterfactual the industry insists does not exist. The trade-off — engagement optimisation or a worse product — did not appear: if a less extractive feed costs nothing in satisfaction, then the extraction is a revenue choice, not a design necessity. Caveat: run on Bluesky — small, self-selecting, technically literate, and with no advertising business — so it does not establish the same result at Meta or TikTok scale. The full paper is paywalled; the Research Briefing is open.

25 May 2026
Critique·Pluralistic

No honor among (ad-tech) thieves

Cory Doctorow uses the Cox Media settlement as the spine for a clean restatement of the case — ad-tech firms lie to users about what they take, and to advertisers about what they deliver. He cites the Procter and Gamble experiment, in which a $200m annual surveillance-advertising spend was zeroed out with zero effect on sales.

Why we logged it. The clearest May statement of the argument from the other side — if the extraction is largely waste, the user is owed the share the advertiser is being scammed out of.

21 May 2026
Regulation·FTC

FTC settles with Cox Media Group over "Active Listening" ad targeting

Cox Media Group and two co-respondents will pay just under $1m to settle FTC charges that they deceived buyers about an AI-powered service that claimed to record and transcribe consumer conversations via smart devices for ad targeting. The action treats inflated ad-tech surveillance claims as actionable deception against the buyer, not only the user.

Why we logged it. A regulator finally punishing one ad-tech firm for lying to another about how much it surveils — proof the extraction is real, and the boasts about it are inflated.

18 May 2026
Research·arXiv preprint

Generative AI advertising as a problem of trustworthy commercial intervention

Qiu and Mei (University of Michigan) argue generative AI changes advertising in kind, not degree. Rather than dropping a product into a slot, it intervenes on the model's output itself — shaping product mentions, information framing, behavioural redirection, and long-term preference through channels users cannot easily detect. Deployed systems and regulators govern only the most visible tier, while the influence most consequential for user autonomy goes unmeasured and undisclosed.

Why we logged it. Extraction moving from your attention to your judgement — the ask hidden inside the answer. A preprint, not yet peer-reviewed.

14 May 2026
Critique·The Ideas Letter

Slobodian — Digital Bandung: stress-testing the language of extraction

Quinn Slobodian interrogates the now-ubiquitous register of "digital empire" and "data colonialism" — Zuboff, Couldry and Mejias, Hao, Crawford — and argues it too often "mistakes polemic for analysis." Subjecting the empire metaphor to the actual history of empires, he proposes three sharper concepts (the comprador, the creole, counter-colonisation) and insists that Global North users are not simply expropriated "digital natives" but "digital creoles," whose pension funds and portfolios are bound up in the very firms extracting from them.

Why we logged it. The most serious engagement with this site's own vocabulary in months — and it partly cuts against us. Its warning that "extraction" used loosely becomes an emotional rather than analytical category is a useful discipline for a movement built on a metaphor: naming a tax is powerful, but the name has to survive contact with how value actually moves. Caveat: a critical essay, not evidence — cite it as a sharpening counterpoint, not an ally.

12 May 2026
Regulation·House Energy & Commerce

Pallone opens a surveillance-pricing inquiry into 25 retailers

Energy and Commerce Ranking Member Frank Pallone wrote to 25 major retailers and grocers — Amazon, Walmart, Kroger, Target, and Costco among them — demanding to know whether they feed individual shoppers' personal data into algorithms to charge different people different prices for the same goods, with answers due 26 May. The letters cite a Consumer Reports investigation into Instacart's AI pricing tests, and argue the absence of a federal privacy law is the gap that lets the practice spread.

Why we logged it. The extraction logic turned on price itself — your data harvested to set what you pay. A congressional record of the mechanism, before any finding. Caveat: a minority-party information request, not legislation or enforcement.

12 May 2026
Regulation·European Commission

Von der Leyen commits to a Digital Fairness Act naming addictive design

The Commission President publicly committed to bringing forward the Digital Fairness Act, naming infinite scrolling, autoplay, and push notifications as "addictive and harmful design practices" engineered to maximise engagement. The Parliament's Legislative Train confirms the DFA as a Q4 2026 initiative targeting dark patterns, addictive design, and unfair personalisation.

Why we logged it. An EU institution naming engagement-maximising design as a harm in its own right — the mechanics of extraction, not only a privacy or competition problem. Announced, not yet tabled, and framed so far around minors.

08 May 2026
Regulation·European Commission

EU draft guidelines on AI Act Article 50 transparency

The European Commission published draft guidelines on the AI Act's Article 50 transparency obligations, which become enforceable on 2 August 2026 and cover interactive-AI disclosure, deepfake labelling, and biometric categorisation. For advertising specifically, the draft confirms commercial speech does not get the artistic-expression exception — AI-generated influencer and celebrity ads must carry the disclosure. Consultation runs to 3 June.

Why we logged it. A forward dated mechanism for labelling synthetic ad content at scale, with penalties up to 3% of global turnover. The first real test of whether transparency can be enforced on the surfaces themselves.

04 May 2026
Regulation·FTC

FTC v Kochava — the first ban on a US data broker selling sensitive location data

A stipulated final order prohibits Kochava and its subsidiary Collective Data Solutions from selling, sharing, or disclosing sensitive location data without affirmative express consent, ending a case the Commission filed in August 2022. The settlement requires a sensitive-location-data programme, supplier consent assessments, and a public-facing withdrawal mechanism.

Why we logged it. The first concrete enforcement on the data-broker layer of the attention economy — the layer that converts location, behaviour, and intent into priced inventory.

04 May 2026
Regulation·Electronic Frontier Foundation

EFF tells the EU to regulate surveillance advertising, not "addictive design"

Ahead of the EU's Q4 2026 Digital Fairness Act, the EFF argues the law should target the root causes — surveillance-based business models and deceptive design — rather than pile more control onto users through age verification. It calls for an explicit ban on dark patterns, a direct attack on surveillance advertising and unfair profiling, a ban on pay-for-privacy schemes ("users should not have to trade their data or pay extra to avoid being tracked"), and mandatory recognition of browser and operating-system privacy signals.

Why we logged it. A leading digital-rights body framing surveillance advertising itself, not its symptoms, as the thing to regulate — the Time Tax thesis in policy language. An advocacy submission, not law; the DFA is not expected until Q4 2026.

April 2026

4 entries
28 Apr 2026
Regulation·Maryland General Assembly

Maryland becomes the first state to ban surveillance-data price hikes

Signed 28 April 2026, Maryland's Protection from Predatory Pricing Act makes it the first US state to bar large food retailers and third-party delivery apps from using surveillance data or dynamic pricing to set a higher price. It lands on top of New York's Algorithmic Pricing Disclosure Act — in force since November 2025, which forces a blunt on-screen label: "THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA" — and the FTC's own study, which found intermediaries using signals as fine as mouse movements, browser history, and abandoned-cart contents to set individualised prices across at least 250 retail clients.

Why we logged it. The extraction argument made literal — the data taken from you converted directly into the price you pay — and the policy response moving from disclosure toward prohibition. Caveat: narrow and sector-specific (Maryland covers only certain food retail; New York is disclosure, not a ban), and there is still no federal statute.

24 Apr 2026
Research·SSRN working paper

Yuan — The attention economy as extractive regime

Yongming Yuan argues that "attention economy" misleads if it implies attention only recently became scarce — attention scarcity is an old human condition. What is new is the declining threshold of extractability: digital infrastructure can now capture, measure, and monetise "sub-threshold attention fragments" — glances, pauses, scroll hesitations, interruption windows — that earlier media could not profitably organise. He proposes a minimal-threshold model in which an attentional unit becomes extractable once its expected return exceeds the marginal cost of capturing it, and closes with a class account of sellers, buyers, and protectors of attention.

Why we logged it. The closest academic articulation yet of the thing this site names — value extracted from attention so small the person never notices the transfer. Caveat: a self-posted working paper by an independent author — not peer-reviewed, fourteen pages — a conceptual contribution and a useful citation for the framing, not empirical evidence.

15 Apr 2026
Research·Pew Research Center

Pew — Teens' Experiences on TikTok, Instagram, and Snapchat

Pew's first cross-platform teen comparison (n=1,458, fielded Sep–Oct 2025) finds that roughly 30% of teen TikTok users say they spend too much time on the app — and 44% of parents say so of their teen. TikTok users are also markedly more likely than Instagram or Snapchat users to report harm to sleep and productivity.

Why we logged it. The users say it themselves. Survey data Time Tax can cite when the argument needs voice, not theory.

01 Apr 2026
Research·SSRN working paper

Catovic — Attention scarcity and builder saturation in digital markets

Armin Catovic models digital-market growth under a finite attention constraint and shows that the equilibrium attention available to each builder converges to a ratio of entry cost to monetisation rate. The paper formalises attention as a scarce, contested resource at platform scale, not as an unbounded externality.

Why we logged it. A formal economic statement of the premise the framework rests on — attention is scarce, monetisation rate is the lever, distribution follows from policy.

March 2026

3 entries
13 Mar 2026
Research·arXiv preprint

Daily affect fluctuations in phone screen content predict anxiety and depression

An intensive year-long study captured a smartphone screenshot every five seconds from 145 adults — roughly 111 million images — and scored each for emotional valence and arousal with a deep-learning model, alongside biweekly mental-health assessments. Day-to-day shifts in the emotional tone of what people saw on screen predicted subsequent changes in depression and anxiety; stable between-person differences did not.

Why we logged it. Moves the case past "screen time" to the content of attention as the active ingredient — what platforms feed you, not merely how long. A preprint, not yet peer-reviewed.

02 Mar 2026
Research·SSRN working paper

Chen, Li & Preuss — Algorithmic attention and content creation on social media platforms

The authors study the revenue-maximising allocation of attention on ad-funded social media, modelling the platform's choice to direct attention either to ads (immediate monetisation) or to creators (to grow inventory). The algorithmic feed emerges as a deliberate allocation against advertisers and creators — not a neutral surfacing.

Why we logged it. Useful citation for the Conversion Dividend argument — the feed is an allocation decision, and allocations are governable.

01 Mar 2026
Critique·Politics & Society (Sage)

Mariani — Regulating the attention economy: the possibilities and limits of antitrust

A legal-scholarly argument that competition-focused antitrust is the wrong tool — more competition can intensify the race to capture attention and deepen the harm. Isabella Luisa Mariani calls for abandoning the price-based consumer-welfare standard, conceptualising markets for attention, and imposing ex ante limits on data-mining and behavioural advertising, with user autonomy as the governing standard.

Why we logged it. An academic case that the harm is the extraction itself, not market concentration — adjacent to the premise that the underlying relationship is non-consensual.

February 2026

2 entries
— Feb 2026
Research·Stanford Law Review

Newman — Attention Capitalism: the law and political economy of attention markets

John M. Newman (78 Stan. L. Rev. 415) argues that a wide range of legal regimes — antitrust, privacy, contract, and tax — have together funnelled human activity into "attention markets," where people trade attention for access to search, social media, and the like, and concentrated power among a few firms. The article proposes a broad overhaul of US law to deconcentrate those markets.

Why we logged it. The nearest formal legal scaffolding yet for treating attention as a market with measurable value — a scholarly cousin to the Conversion Dividend's premise that captured attention is value owed back. Caveat: its remedies are structural and doctrinal, not a revenue-return mechanism.

20 Feb 2026
Research·SSRN working paper

Thomas — Taxing Attention

Kathleen DeLaney Thomas gives the attention economy a tax-law treatment, taking up the Pigouvian idea — taxing attention capture to internalise its social cost — and working through how such a levy would sit inside real tax policy, across social media and the broader market for attention.

Why we logged it. The closest formal cousin of the Conversion Dividend yet. Where the Dividend returns a share of ad revenue to the people whose attention produced it, this prices the externality of capturing that attention at all. A working paper — scope to be confirmed from the full text.

December 2025

2 entries
— Dec 2025
Regulation·US House of Representatives

Auchincloss introduces the Education Not Endless Scrolling Act — a digital ad tax

Rep. Jake Auchincloss (MA-04), with co-sponsor Rep. Maggie Goodlander, introduced a bill to levy a 50% excise tax on digital advertising revenue above $2.5bn and route the proceeds to one-on-one tutoring, vocational-technical schools, and local journalism. Auchincloss frames platforms as "attention fracking" children and treating youth "like products, not people," pitching the measure as part of an "UnAnxious Generation" package.

Why we logged it. The closest thing in live US legislation to the Conversion Dividend — it accepts that advertising revenue is built on extracted attention and proposes returning a declared share of it to the public. Caveat: it routes the money to public goods rather than to the people whose attention produced it, and remains a single bill at introduction stage with no clear path to passage.

— Dec 2025
Research·Psychological Bulletin

Feeds, Feelings, and Focus — a meta-analysis of short-form video use

Lan Nguyen and colleagues (Griffith University) pool 71 studies and 98,299 participants and find heavier short-form-video use moderately associated with poorer cognition — the strongest effects for attention and inhibitory control — and weakly with poorer mental health, including anxiety, stress, and sleep disturbance. Notably, addiction-scale measures of compulsive use predicted harm more strongly than time-on-app alone.

Why we logged it. The largest synthesis yet behind the claim that the engagement-optimised feed degrades the very faculty it harvests — attention. Caveat: the associations are correlational, and most of the underlying studies are cross-sectional, so causation is unproven.

In press 2026

3 entries
— In press 2026
Research·American Journal of Preventive Medicine

Problematic social media use trajectories are linked with lower cognitive performance in adolescents

Drawing on the Adolescent Brain Cognitive Development cohort, this analysis links trajectories of problematic social media use — the compulsive, conflict-generating pattern, as distinct from raw hours — with lower cognitive performance in adolescents: impaired attention, reduced working memory, diminished executive functioning. It extends the JAMA trajectory work from October 2025 by shifting the exposure from time spent to the compulsion itself.

Why we logged it. The harm tracks the grip, not the minutes — the variable that matters is the one engagement design optimises for. Caveat: logged from the abstract; the publisher page would not render at logging time, so details await the full text.

— In press 2026
Research·Information, Communication & Society

Problematic social media use and adolescent mental well-being: socioeconomic inequalities across 35 countries

Roughly 145,000 adolescents across 35 countries in the WHO-linked Health Behaviour in School-aged Children study: problematic social media use is consistently associated with more psychological complaints and lower life satisfaction — and the association is stronger for low-SES adolescents than high-SES ones, especially on life satisfaction. A companion analysis appears in the World Happiness Report 2026.

Why we logged it. The distributional claim made empirical: the attention tax is regressive. The people with the least material buffer pay the most in wellbeing terms for the same extraction. Caveat: cross-sectional survey waves (2018 and 2022) and self-reported use.

— In press 2026
Research·Computers in Human Behavior, Vol. 179

Fournier et al. — Attention hijacked: how social media notifications disrupt cognitive processing

A Lyon team ran a Stroop task on 180 university students while delivering notifications, and measured a transient cognitive slowdown of roughly seven seconds per ping, driven by perceptual salience, learned association, and inferred social relevance. Crucially, the magnitude was predicted by interaction frequency — notification volume and checking behaviour — not by total screen time.

Why we logged it. A measurement of the tax in milliseconds, and a pivot from "time on device" to "interruptions per device" as the design lever that matters.

No entries match — try a different term or clear the filter.

The chronicle is curated by The Editors and updated on Mondays and Thursdays. Submissions of regulatory rulings, peer-reviewed work, and primary documents are read on a rolling basis — write in via Contact & correspondence. Selection criteria — first-hand sources where possible, links that work, claims described as narrowly as the source warrants.

The longer-form annotated library lives at Evidence. The framework lives at Glossary. The policy proposal lives at Dividend.