How TIME, The Economist and Dow Jones each decided what to expose to AI agents

Hi, I’m media innovation journalist Ulrike Langer and you’re reading my weekly AI in media newsletter. Thank you for being here! If someone has forwarded it to you please subscribe and never miss any future post.

A few weeks ago I wrote about how publishers can get paid by AI agents — the coalitions, the licensing rails, the question of who cuts whom a check. This issue is about the strategic question that comes before that one. Before you decide how an agent pays you, you have to decide what you let it have. And on that question, three of the smartest media companies in the business have taken three genuinely different routes — each one following, it turns out, from what that company already knows it's selling.

TIME's Chief Operating Officer Mark Howard and Josh Muncke, VP Generative AI at The Economist, describe the near future in almost the same words: two versions of the web, one built for humans and one for machines. Humans get the rich experience with the art, the layout and the feature spread. Agents get data points. As Muncke puts it, they want "clear structure, questions and answers, ideally text," not carousels and cover images.

They read the situation the same way. But then they take different actions.

TIME is betting that the thing to fear is being ignored. If agents are going to answer the world's questions by using somebody's journalism, they had better use yours rather than a competitor's — so make yours the easiest in the building to reach.

Under CEO Jessica Sibley, TIME dropped the paywall on its 100-year archive in 2023 and has spent the three years since turning openness into an operating system. In June 2026 it flipped from allowing AI bots by default to blocking them by default, then put roughly 70 of them on a whitelist. The ones it lets in don't get the normal page. They get a stripped-down markdown version, built by the vendor TollBit, that a machine can read in about a quarter of a second instead of a minute, using roughly 90 percent fewer tokens. Humans still see the full site. Machines get the fast lane.

Then comes the move that turns a concession into a business: TIME sells the visibility back. In March it launched a consulting product that shows brands how they appear inside ChatGPT, Claude, Gemini and Perplexity — TIME says it shows up in 30 to 50 million AI citations a month, and that number is the pitch. The more accessible TIME's content is to AI agents, the more often — and more accurately — TIME shows up in AI answers itself. That measurable visibility is also the credential TIME sells its GEO product on: brands are buying the analysis from a company that demonstrably appears in AI answers and owns the infrastructure to prove it.

What’s interesting about this approach: The licensing checks from OpenAI, Perplexity and Amazon are not a major revenue factor. "They are not financially game-changing for our business," Sibley said last year at SXSW London, "but we would rather play offense than defense. We'd rather be at the table." When OpenAI first came calling, she framed the choice as three options — "litigate, negotiate, or do nothing" — and dismissed the third on the spot: "I'm not doing nothing, ever." The checks aren't the point; the room is. The visibility is the product. Expose everything, measure where you land, then sell the map.

Muncke sees the same two-track web and draws the opposite conclusion: The more you feed the machines, the more you risk giving away the thing people pay for.

So The Economist is being, in Muncke's word, "forensic." Its early agent-readable experiments run only on content that already sits outside the paywall — marketing pages and B2B sales copy — rebuilt into the stripped-back Q&A structures agents prefer. B2B buyers increasingly start a purchase by asking ChatGPT or Claude, so those pages have to show up cleanly. But the journalism stays behind the wall, and the question of "what portions of our editorial content should also appear in those surfaces" is one Muncke says the company is still circling carefully.

You can watch the same restraint in its ChatGPT app, launched in May — the first from a major consumer news publisher. It does exactly one thing: pull up charts from the Economist's Trump approval tracker. Not because this limited version is all they could build, but because the tracker already lives in front of the paywall. "We thought we can explore this surface," Muncke said, without "directly exposing the depths of some of our premium written content."

TIME treats exposure as inventory to maximize. The Economist treats it as a leak to manage. 

And then there's the company that seems barely troubled by the question the other two are wrestling with — because it decided the journalism was never the thing worth guarding.

Ask Dow Jones's Chief Growth Officer Scott Havens whether readers might one day reach The Wall Street Journal through a chatbot instead of the front page, and he shrugs it off: You might read the Journal "through OpenAI, Anthropic or whatever." That’s his read on where the money will be. CEO Almar Latour is chasing a billion dollars in EBITDA (operating profit before financing and accounting deductions) by building what he calls the "stack" — news, data, analytics and events stacked around a vertical such as energy or geopolitical risk, each layer feeding the next until customers "have work tools that lead back to Dow Jones, whether they know it or not."

In that model the articles are the top of a funnel, not the treasure. The treasure is a twenty-year database of how the best wealth managers in America invest, or fuel-pricing intelligence bought and wired across the whole company, or the geopolitical-risk analysis that clients build into their decisions. Latour's test for what's safe is whether it's "truly proprietary" — the things his journalists "pick up in the corridors of power" that no bot can scrape. The other half of the test is how deeply a customer's workflow is wired into Dow Jones. Let an agent summarize today's WSJ story. It can't summarize the database underneath, and it can't replace the tool a bank has already built its process around.

Line the three up and the contrast is real. TIME and The Economist genuinely disagree about whether exposure is an inventory to maximize or a leak to manage. But both companies are reading their own situations correctly and arriving at different places, because they're protecting different assets.

That's the pattern underneath all three. TIME is selling reach, and the data about where that reach lands, so it opens the doors and meters the traffic. The Economist is selling judgment you can't get for free, so it guards the judgment and exposes the rest. Dow Jones is selling proprietary information and a place inside the customer's own decisions, so it can hand the journalism to the agents and keep the part that pays. Each move follows from the asset.

And the routes aren't mutually exclusive. A publisher could meter its archive like TIME, ring-fence its premium reporting like The Economist, and build a proprietary-data product like Dow Jones — this is a menu, not a fork. What decides the mix isn't nerve or timing. It's the same prior question in every case: What are people actually paying you for? And would a machine reading your work make that more valuable, or less? Answer that and the decision starts to make itself.

The machines are arriving on a clock — Cloudflare starts blocking mixed-use crawlers by default in September, and the standards for how agents and publishers talk to each other are being written this year, whether publishers help write them or not. Some of these bets won't pan out; maybe none of them survive what the web becomes. But that's how the map gets drawn — by publishers who moved early enough to learn something. TIME, The Economist and Dow Jones are three of them. The useful question isn't which one guessed right. It's what you'd try first, once you're honest about what you're really selling.