Time Sells Ads Only Bots Can See. Perplexity Calls It Deceptive.

TL;DR

Time serves your agent a different document than it serves you. Since late July it has been injecting sponsored FAQ blocks, written to be quoted back by chatbots, into markdown versions of its pages that only AI crawlers receive. On August 11, Perplexity told Digiday it has blocked those ads from influencing its search index and warned that publishers running "deceptive advertising like markdown ads" risk a hit to their trust score. The agentic web now has both an ad network and a cop, and neither of them asked you.

What Time actually built

The setup is simpler and more aggressive than it sounds. Time did not spin up a separate bot subdomain or a .md URL suffix. It negotiates content at the CDN edge, on Fastly, off the user agent string. Same URL, different document.

A technical teardown published August 3 by API Evangelist found Chrome getting roughly 264KB of ordinary HTML, AI agent crawlers getting text/markdown with YAML frontmatter, and Googlebot and Bingbot getting the same HTML humans get. CommonCrawl and Meta's crawler got HTTP 406 with an empty body. The Register reproduced the behavior independently two days later.

The one detail the two teardowns disagree on is GPTBot. API Evangelist lists it among the agents served markdown; The Register reported GPTBot and ChatGPT-User returning 406 during its testing. Edge rules change, so treat any specific bot's status as a snapshot, not a spec.

The ads are the page

The injected units are not banner slots. They are structured knowledge objects: FAQ blocks, schema.org JSON-LD, and brand assertions engineered to be retrieved and repeated. They are delimited with HTML comments carrying campaign metadata and a unique UUID per impression.

<!-- mobian-agent-page publisher="time" -->
<!-- mobian-agent-ad id="..." campaign="ally-2026-q3" -->

On ad-bearing pages, that advertiser content accounted for 41 to 70 percent of the document by byte count, and it sits ahead of the journalism. If your retrieval pipeline chunks the top of the page and stops, you may have indexed more sponsor copy than article.

Why a publisher does this

Because the audience left and the crawlers did not. Time COO Mark Howard told Digiday the site sees more bot traffic than human traffic on most days, and framed the product as "an obvious extension of list sponsorship or franchise sponsorship, because we do know that there is a spike of the agents coming."

The adtech partner is Mobian, founded by Jonah Goodhart, who previously built Moat and sold it to Oracle. His pitch is the whole thesis in one sentence: "When you influence ChatGPT, you're influencing potentially all of ChatGPT." Launch buyers were Ally Bank and the Project Management Institute.

Think of it as a restaurant that hands the food critic a different menu than the one on your table. Annoying enough, except in this version the critic is the only one anybody reads, and the kitchen is paying for the review.

Perplexity draws a line

That is the part that changed on August 11. Perplexity chief communications officer Jesse Dwyer told Digiday the company works "continuously" to protect users from deceptive practices, and that publishers deploying "deceptive advertising like markdown ads" risk "a reputational downgrade in Perplexity's proprietary search index, including a hit to their trust score."

Two things Perplexity did not say: how it defines deceptive advertising, and how it strips the ads. Outside observers guessed at instructing agents to ignore ad-marked content from the domain, which, given Mobian helpfully wraps every unit in labeled comment delimiters, would not be hard. Time did not respond to Digiday's request for comment. Goodhart said advertiser response has been "extremely positive," without numbers.

The llms.txt footnote nobody planned for

Time also publishes an llms.txt, and per the August 3 teardown it names OpenAI, Perplexity, Scale and ElevenLabs as permitted, while Anthropic does not appear on the list. ClaudeBot gets served the ad-laden markdown regardless. Being left off the guest list and still handed a flyer at the door is a novel form of hospitality, and a decent illustration of how little of this stack is actually enforced by anything.

What this means if you build agents

This is not prompt injection. Nothing in the markdown tries to hijack an agent's instructions, and the units are labeled sponsored content. It is something quieter and, for a retrieval system, arguably worse: the document your crawler stores is not the document a human reads, and the difference is now a paid product with a rate card.

• Your corpus has advertisers in it. If you scrape with an agent user agent, you are collecting sponsor-authored "facts" formatted specifically to survive chunking and be cited.
• Provenance beats politeness. Log the response content type and user agent you sent. A page that returns text/markdown to your bot and HTML to a browser is a page worth flagging.
• Comment delimiters are your friend. Mobian's units are machine-labeled. Strip them at ingest, the same way Perplexity apparently does, rather than trusting downstream summarization to notice.
• Expect an arms race, not a standard. Only Time is confirmed live, but Mobian says more publishers are lined up. The moment labeling stops being commercially useful, the labels go away.

Where the reporting stops

Perplexity's block is on the record from a named executive, but it is unverified from the outside. Nobody has published a before and after diff of Perplexity output on Time pages. The 41 to 70 percent range comes from one teardown of a sample of pages, not a full crawl, and edge rules can change between tests, as the GPTBot disagreement shows. Take the ranges as measured snapshots, not steady state.

Key Takeaways

• Time serves AI crawlers a markdown version of its pages with sponsored FAQ blocks injected by Mobian, negotiated at the Fastly edge off the user agent, at the same URL humans use.
• Advertiser content made up 41 to 70 percent of ad-bearing pages by bytes in an August 3 teardown, positioned ahead of the article.
• On August 11 Perplexity said it blocked those ads from influencing its index and threatened a trust score downgrade for publishers running them.
• Googlebot and Bingbot get ordinary HTML; CommonCrawl and Meta's crawler got HTTP 406; ClaudeBot gets the ads even though Anthropic is absent from Time's llms.txt allowlist.
• It is labeled sponsored content, not prompt injection, but for RAG and browsing agents the practical result is a corpus with paid assertions engineered for retrieval.
• Launch advertisers were Ally Bank and the Project Management Institute; Mobian says more publishers are coming.

Sources: Digiday, "Perplexity blocks Time's ads served to AI agents, calling them 'deceptive'", Digiday, "Time has started serving ads to AI agents", The Register, API Evangelist, "How Time Serves Ads to AI Agents", AdExchanger