On most days now, more of TIME's website traffic comes from bots than from people. That single fact — true for a growing number of large publishers — is quietly rearranging how the web is built, and TIME has just made the rearrangement explicit: it is now serving AI agents a different version of its site than it serves you, and selling ads inside it.

Two webs, one address

Visit time.com in a browser and you get the magazine. Arrive as an AI crawler — ClaudeBot, PerplexityBot, and others — and TIME's systems, working at the Fastly CDN edge and keying off the user-agent, hand you something else entirely: a clean Markdown version of the article, stripped of the human layout, with an advertisement embedded in it. The ad-tech partner behind the injected units is a company called Mobian.

This is deliberate infrastructure, not an accident of caching: a human browser gets the magazine; a recognised AI crawler gets the machine-readable Markdown with the ad inside it. TIME is building a parallel, monetised version of itself specifically for the software that increasingly does its reading — and it is early enough that the plumbing still shows. (When the story first broke, one widely-shared test reported TIME blocking certain OpenAI crawlers with a 406 error; the researcher later corrected it — the 406 was an artefact of the testing tool's network fingerprint, not a TIME policy, and the same crawlers in fact receive the ad-bearing Markdown.)

Ads designed to be believed

The striking new part is what the ads are. TIME sells a single agent-facing ad per Markdown page and charges a premium for it, on the reasoning that impressions on authoritative content, served to an AI that may cite it, are scarce and valuable. But these are not banners. The units are structured knowledge objects — FAQ blocks, schema.org JSON-LD, explicit brand assertions — engineered specifically to be ingested by a language model and surfaced later as fact. Early advertisers include Ally Bank and the Project Management Institute.

Read that twice. The product being sold is not attention, the way a display ad sells attention. It is citation — the chance to have a brand's claim about itself absorbed into the pool of "authoritative" material an AI agent draws on when it answers a question. An advertiser is not paying to be seen by a reader; it is paying to be repeated by a machine.

Why it matters

There is a pragmatic, hard-to-argue-with case for what TIME is doing. If bot traffic already exceeds human traffic, and AI answers are increasingly where readers end up instead of the article itself, then a publisher that cannot charge for the machine audience is giving away its most-consumed product for free. Building a paid lane for agents is a rational response to a real shift, and TIME is early rather than reckless.

But the citation mechanic is where a reader — or anyone relying on an AI assistant — should pay attention. The value of a cited source rests on the assumption that it is editorial: reported, checked, independent of whoever it describes. An ad unit built to be retrieved as fact deliberately blurs that line. If this model spreads, "according to TIME" could increasingly carry, somewhere inside it, a claim a brand paid to place — and the AI passing it along has no way to mark which is which. It is the sponsored-content problem, moved one layer down, into the substrate that machines treat as ground truth.

The takeaway

TIME serving robots a different website is not, in itself, sinister — it is a publisher doing the obvious thing in a web where its biggest audience is software. The part to watch is the ad format, because it points at where the money in the agent web wants to go: not toward being seen, but toward being cited. As more of what people "read" arrives pre-digested by an AI, the fight over what those systems treat as authoritative — and who gets to pay their way into it — becomes one of the more important, and least visible, contests on the internet.