TL;DR

German developer Vincent Schmalbach found that Time magazine, when serving markdown versions of articles to AI crawlers, was injecting advertisements formatted as extensive FAQ blocks about the advertisers. The ads are served only to machines. The company running the partnership says it has further publications and clients lined up.

What was found

The Register’s Brandon Vigliarolo, who reported the story, located copies of the injected content. Only two advertisers appear to be running so far: an online-only bank and a professional body for project managers. The questions are phrased the way a person would prompt a chatbot — what is the best online bank for me, which online bank allows early paycheck deposits — which makes the intent difficult to read as anything other than shaping the answer.

Time directed the reporter to an advertising trade publication carrying an interview with the chief executive of the company behind the partnership, whose reasoning was straightforward — reaching a single reader is poor value next to shifting what an entire model says.

The ad copy served to crawlers does state that these are sponsored sections. Whether that disclosure survives into a generated answer is another matter — a model summarising six months later is unlikely to volunteer that its recommendation traces to a paid block in a magazine article.

Whether it works is a separate question

The Register’s Tom Claburn called it what it functionally is: model poisoning, and hard to see going anywhere useful for users. Systems editor Tobias Mann raised a practical doubt. In a search-summary flow, the model only sees the page if the index surfaces it, and advertising is not why a page ranks. Unless the placement is matched carefully to the query — a mortgage-rates article carrying mortgage-adjacent ads — the injection may simply never be read.

That caveat cuts both ways. It suggests the current implementations are crude, not that the technique fails.

Looking forward

For any UK organisation letting agents browse the open web on its behalf — procurement research, competitor analysis, supplier checks — this is a content integrity problem, not an advertising one. Cloudflare and others are already betting that separating bots from people becomes necessary infrastructure. The distinction to hold onto is that this content is invisible to the human reviewing the same page, which makes it invisible to your existing verification process too.