Media Trendlines — October 20, 2026
📰 Key Themes
- Reuters’ president argues AI is a production shock, not a distribution shock, and that only journalism nobody else has will keep its price.
- Top U.S. publishers block Anthropic’s training crawler more often than OpenAI’s, and they block AI training far more than AI search, so robots.txt now works like a price list.
- Meta’s new shopping agent used one review site’s testing to pick a product, then sent the sale to stores that site earns nothing from.
- Slate made its editor-in-chief the CEO, which shows that subscription publishers now treat the product and the P&L as one job.
- Versant’s CEO ruled out buying the cable networks Skydance just took on, a sign that linear TV assets have run out of buyers.
Jump to: 📺 Big Media Moves · 💡 Business Model Innovation · 🎙️ From the Pods · 📎 Also Noted · 🧭 Takeaways
📺 Big Media Moves
Slate Making Its Editor the CEO Shows Subscription Publishers Can’t Split the Product From the P&L
Source: Axios Media Trends — Sara Fischer
Slate named editor-in-chief Hillary Frey CEO, effective immediately. She keeps running the newsroom. Frey succeeds Dan Check, who moved to another role inside parent company Graham Holdings. Her brief is to make Slate’s editorial products better so they sell more subscriptions, which are now its largest revenue stream. The timing is hard. Graham Holdings said in a July filing that Slate reported lower revenue and ran at a loss in the first half of 2026. The magazine had just over 100 employees at the end of 2025.
Axios places the move in a line of editors who became chief executives, including Nick Thompson at The Atlantic, Stephanie Mehta at Mansueto and Edward Felsenthal at Time. The logic is simpler than the trend story makes it sound. In an ad-funded business, the editor makes the content and someone else turns traffic into money. In a subscription business, the content is what customers pay for, so every editorial call is also a revenue call. Putting one person in charge of both removes the handoff where most subscription strategies fall apart. The risk is just as clear. An editor-CEO at a loss-making outlet will eventually have to cut the newsroom they built, and Slate’s readers will learn quickly whether a journalist makes those cuts differently from an operator.
Versant Passing on Skydance’s Cable Networks Shows Linear TV Has Run Out of Buyers
Source: Axios — Sara Fischer and Christine Wang
Interviewed on stage at Axios Media Trends Live, Versant CEO Mark Lazarus said his company is unlikely to buy any of the cable networks Paramount Skydance picked up from Warner Bros. Discovery. Versant, NBCUniversal’s cable spinoff, was the obvious buyer if Skydance wanted to sell. If the company built to consolidate cable won’t bid, there’s no natural market for these channels. That means Skydance either runs them for cash while they decline or sells them cheap to a financial buyer. Neither option helps with the debt load the market is already marking down.
💡 Business Model Innovation
Reuters’ Production-Shock Warning Shows AI Will Pay for Unique Journalism and Nothing Else
Source: The Rebooting — Brian Morrissey (newsletter and podcast)
Paul Bascobert, president of Reuters, says the industry has misread the threat. “The internet and cable were distribution innovations,” he told Brian Morrissey. “AI is a production disruption, which is going to have a different effect on the industry.” Reuters has about 2,600 journalists, grows 6–7% a year at roughly 21% margins, and gets around 80% of revenue from multi-year contracts. It sells the same reporting through what Bascobert calls windows: financial terminals, media clients, its own paywalled site, and now AI companies, with disclosed deals with Meta and Microsoft. “We think of it as build it once and sell it four times,” he said.
Inside the newsroom, the practice matches the pitch. Reuters’ human-assisted economic data service cut the time from announcement to publication from about 90 seconds to roughly 34. A World Cup tool picks the five to seven minutes that matter from a 30-minute press conference and hands them to an editor. Journalists own whatever the tools produce. Bascobert doesn’t worry much about financial news. His worry is that media clients move too slowly and churn. “If there’s something that does give me some cold sweat, it is a fear that the industry isn’t gonna move fast enough,” he said.
The production framing is the most useful idea in AI-and-news commentary this month, because it explains why the usual newsroom defense doesn’t work. Most publishers plan to limit AI to transcripts and workflow. That does nothing about a market flooded with near-free content competing for the same attention and ad dollars. Morrissey calls it a K-shaped split. At the top are wholesalers and specialists with data nobody else has, which AI companies will pay for. At the bottom is the recap economy built to catch Google traffic, now competing with machines that make the same thing for free. Bascobert predicts “a golden age of journalism” that “may not be a golden age of massive media companies.” He’s right, and media executives should notice he was talking about companies like theirs.
Publishers Blocking Claude More Than ChatGPT Shows Robots.txt Has Become a Price List
Source: Axios Media Trends — Sara Fischer and Kerry Flynn · The Media Copilot — Pete Pachal
Axios read the robots.txt files of 102 top U.S. publisher domains. 46% fully block Anthropic’s ClaudeBot, compared with 38% that block OpenAI’s GPTBot. More than 40% block AI training crawlers, but only 25% bar AI search crawlers. OpenAI’s training crawler is more than three times as likely to be fully blocked as its search crawler. Axios points out that OpenAI has struck dozens of licensing deals with publishers, while Anthropic hasn’t announced any.
The traffic numbers explain why publishers are drawing lines. Chartbeat’s annual playbook, cited by The Media Copilot, found that Google referrals to hundreds of publishers fell 40% between July 2025 and July 2026, twice the previous year’s decline. A Pew study of more than 68,000 searches found that only 1% ended with a click on a source cited in an AI Overview. Separate data from Ceramic.ai shows AI crawlers hitting sites at about 10 times the rate of search crawlers.
Read together, the data shows publishers aren’t blocking AI. They’re blocking AI that hasn’t paid. They let in the search crawlers that might send a visitor and the bots from companies that have signed deals, and they shut out the training crawlers and the company with no deals. That’s a pricing strategy expressed in a text file. It’s also fragile. Robots.txt is a request, not a lock, which is why Ziff Davis is suing OpenAI for scraping despite it. Publishers that want this position to hold need enforcement at the edge and a licensing offer ready, so that a block starts a negotiation instead of ending one.
The WordPress angle: the publishers furthest ahead aren’t just blocking. They serve two versions of the same story. Time shows human readers full storytelling and gives agents structured data, and People Inc., Forbes and the Daily Mail are converting pages to Markdown to cut crawl load and token costs. That work happens in the CMS. Publishers whose content models already separate structured fields from presentation can do it with a template. Those with years of unstructured HTML face a migration project before they can sell anything to a machine.
Meta’s Muse Shows AI Shopping Agents Will Take the Reviewer’s Work and Send the Sale Elsewhere
Source: The Media Copilot — Pete Pachal
Pachal asked Muse, Meta’s personal AI agent, to pick a robot vacuum. When pressed, it said it had relied most on Vacuum Wars, a site that does real hands-on testing. The other “reviews” it cited were a thin repackaging of Vacuum Wars data and an AI-written roundup on an aromatherapy blog. Wirecutter, PCMag and Consumer Reports didn’t appear at all. Pachal flags it as inference, but the missing outlets are the ones that block crawlers, sue AI companies or paywall their ratings. Vacuum Wars pays for its testing with affiliate links, mostly to Amazon, and Amazon blocked Muse from transacting. So Muse sent Pachal to Best Buy, Target and manufacturer storefronts, none of which pay Vacuum Wars. It also recommended a model that wasn’t in its own review summary, and every product it suggested had a working checkout. Mark Zuckerberg has said transactions in Muse will be a revenue source.
This is the affiliate business model breaking down in real time. Commerce journalism worked because one link connected the test to the sale. In an agent world, a reviewer would need a separate deal for every combination of assistant and store, and the agent picks the store based on where it can check out, not who did the testing. DataDome figures cited in the same issue show the scale. Muse made 19.5 million requests in its first nine days, and 8.9% of a sampled slice went to login pages, payment systems and forms. Blocking doesn’t solve it either, because the clone sites scraping Vacuum Wars got into the answer anyway. Commerce teams should stop assuming they can win on test quality and start asking which agents can see their recommendations and which stores those agents can actually buy from.
🎙️ From the Pods
Digiday’s Post-Google Playbook Shows Publishers Are Selling AI Visibility Before Anyone Can Measure It
Source: The Digiday Podcast — Kimeko McCoy and Tim Peterson, with Jessica Davis and Sarah Guaglione
Digiday’s reporters, back from the publisher summits in the U.S. and London, say the mood has moved from dread to “recalibration.” They see three moves. The first is agentic subscription products, such as Axios Direct, an enterprise feed that will connect to Claude, Gemini or ChatGPT so the money comes from paying subscribers rather than chatbot referrals. The second is AI visibility services sold to brands. Future reportedly has about 30 clients for its GEO offering, and the Daily Mail said tracked-prompt visibility for test clients rose from about 50% to nearly 70%. The third is machine-readable content. Reorganizations follow the same pattern. The New York Times’ licensing team doubled to about 10 people, and People Inc. remains the example everyone cites after cutting its reliance on Google referrals from about 80% to 20–25%.
The weak point is the GEO business. The hosts concede that AI visibility measurement is panel-based and opaque, hard to tie to conversions, and nobody knows how long a lift lasts. Publishers are now selling brands the same unproven metric that Google sells them. That can work as an extension of branded content for a year or two. It won’t survive the first renewal where the client asks what the visibility actually sold.
The Grill Room’s Self-Audit Shows Sports Talent Now Has More Power Than the Networks That Pay It
Source: The Grill Room (Puck) — Dylan Byers, with Bryan Curtis and Peter Kafka
Byers brought on two fellow media reporters to ask whether the beat is oversaturated. Kafka answered “Yes,” and added that plenty of media executives enjoy reading about themselves. The more useful part came later. Curtis described an “announcer empowerment era” at ESPN, with one half of its top NFL booth signed and the other not in a Super Bowl year, Pat McAfee taking airtime across the network, and Kirk Herbstreit working for a competitor at the same time. Sports announcers used to want money and little else. Now the biggest ones bring their own audiences, so networks have to bargain with them like partners. Sports rights may be the last mass-audience asset, but the people who talk over them now take a bigger share of the value.
📎 Also Noted
- 🔹 Netflix cut its full-year ad revenue forecast by about A$430 million, to A$3.9 billion, blaming a slower-than-expected rollout of new ad products, even though subscription revenue beat expectations. — Mumbrella
- 🔹 Google made five changes to AI Mode and AI Overviews that add more and higher-placed links, and said Preferred Sources usage has grown more than sevenfold since January. It released no click-through data. — The Media Copilot
- 🔹 Reach signed a paid deal giving Google enhanced content rights across its UK and Ireland titles as part of Google’s news AI pilot. Neither side disclosed the features, the term or the money. — The Media Copilot
- 🔹 Joel Kaplan, Meta’s chief global affairs officer, will leave next year after more than 15 years as Mark Zuckerberg’s top Washington adviser. — Axios
- 🔹 CBS News president Tom Cibrowski flew to London to calm the foreign desk after new foreign chief Shayndi Raice warned staff at an all-hands that they could be fired for insubordination. ⚠️ Paywalled — Status
- 🔹 Ad requests tied to low-quality AI content rose 354% from January to September, passing 3.2 billion, according to DoubleVerify. — Axios
- 🔹 At an AE Ventures trade show, 71 of 126 guests completed an AI pre-event interview, but only 9 used an AI meeting recorder. AI products that replace an existing habit get adopted, and ones that ask for a new habit don’t. — A Media Operator
🧭 Takeaways
- The recap economy is finished. If a machine can produce your story for free, it has no market value. Publishers should look at their output by originality, not by page views, and move budget toward the reporting nobody else can replicate.
- A block is only worth something if a deal comes after it. Robots.txt rules that separate training from search and paying companies from non-paying ones are a sound start. Publishers should back them with edge enforcement and a licensing offer ready to send.
- Affiliate revenue is now an agent-access problem. Commerce teams need to know which assistants can see their recommendations and where those assistants check out. Test quality no longer guarantees the commission.
- Don’t sell a metric you can’t defend. GEO packages are an easy add-on to branded content. Publishers should tie them to outcomes the client can verify before the first renewal conversation.
- The editor and the business now answer to the same customer. When subscriptions pay the bills, Slate’s structure is the logical end state. Publishers that keep editorial and revenue in separate silos should expect to lose time at every handoff.