TIME serves hidden ads to AI crawlers, not readers

Your website already has two audiences. One of them is a bot, and somebody just figured out how to sell ads to it. A German developer found that TIME.com serves a completely different version of itself to AI crawlers — a stripped markdown document carrying sponsored content that no human visitor ever sees. If you sell anything that an AI assistant might recommend, this is the shape of the channel you're about to compete in.

What actually happened

On August 5, developer Vincent Schmalbach documented what TIME returns depending on who's asking. Request a page as Chrome and you get 303,235 bytes of HTML. Request the same URL as ClaudeBot and you get 13,409 bytes of markdown. PerplexityBot and OAI-SearchBot get the markdown too. GPTBot and ChatGPT-User get a 406.

The markdown isn't just a lighter render. It carries sponsored blocks written as FAQ sections — Ally Bank answering questions about banking, the Project Management Institute on the business pages — none of which appear in the HTML humans receive. The response headers name the vendor: x-mobian-impression with a fresh UUID per request, and x-mobian-tokens: 3323. The impression is metered in tokens, not pageviews. That is a billing model built for readers that don't have eyes.

TIME isn't hiding it. The Register reported that Mobian CEO Jonah Goodhart pitches the reach directly: "When you influence ChatGPT, you're influencing potentially all of ChatGPT." TIME COO Mark Howard confirmed the sales team sells agent ads to brands. TIME converted its pages to markdown in June to court AI crawlers in the first place.

Why AI crawler ad inventory matters for your business

Two things just became true at once, and they pull in opposite directions.

First, the retrieval layer is now a paid channel. When a buyer asks an assistant "best online bank for early direct deposit," the answer is assembled from documents that a publisher was paid to shape. You are not competing on product pages anymore. You are competing on whether your facts are in the context window when the question gets asked — against parties who bought their way in.

Second, and more useful: the technique itself is a plain HTTP trick. Content negotiation by user agent, markdown for machines, HTML for humans. You can serve a clean machine-readable version of your catalog — specs, availability, return policy, real pricing — without shipping 300KB of layout to a crawler that will throw it away. That's not a growth hack, it's just structured product data with a delivery format attached.

The line we'd hold: serve machines a truer version of your data, not a different one. Cloaking that contradicts your human pages is a search-penalty problem and, eventually, a trust problem. And when your assistant recommends a vendor, keep in mind you may be reading a placement. Verify the claim against the vendor's own docs before it enters a contract.

Sources: Vincent Schmalbach, The Register.