What happened
On August 11, 2026, Digiday reported that Perplexity had blocked Time's ads for AI agents from influencing Perplexity's search index. Digiday said Perplexity described the markdown ad format as deceptive advertising and warned that publishers using it could take a reputation hit inside Perplexity's proprietary index.
Business Insider followed on August 12 with the same core development and additional comment. The report said brands including Ally Bank and the Project Management Institute had recently tested ads on Time's site that were designed for AI bots to read, and that Perplexity objected to the practice as cloaking: showing one page to people and a different page to software.
The background matters. On July 30, Digiday reported that Time had begun converting webpages into stripped-down markdown copies intended to be easier for AI systems and agents to read. Time then worked with Mobian to place sponsored FAQ-style brand information in those markdown pages. Digiday reported that those ads were labeled as sponsored content, and that Time saw the product as a way to monetize growing bot traffic.
Perplexity's own help material says PerplexityBot indexes pages in a search-engine-like way, respects robots.txt directives, and is not used to train foundation models. The company also says it may still index a domain, headline, and brief factual summary if a page is blocked. That does not settle the ad dispute, but it confirms that Perplexity treats retrieval, crawling rules, and publisher trust as product issues.
Those are the reported facts. REC's read is about expert publishing practice. If a page has one version for humans and another for agents, and the agent-facing version includes paid claims, the record behind a future answer becomes harder to explain.
Why the argument matters
Brands have always tried to influence discovery. The newer problem is that the audience for a page may be software that later answers a person in a different interface, without showing the original page, the page version, or the surrounding disclosure.
A sponsored FAQ on a publisher's machine-readable page may be clearly marked at the point of publication. The harder question is what survives after retrieval. If an AI answer repeats a brand claim without the sponsored context, the reader may understand it as neutral editorial information. If the system blocks the claim, the advertiser may feel the paid placement never had a fair chance to work.
That is why Perplexity's reaction is useful even if other AI search systems choose a different policy. It draws a practical line around trust in retrieval. AI products need to decide whether they treat machine-targeted advertising as context, contamination, paid source material, or something else. Publishers and brands need to decide whether they are building evidence for readers or messages for crawlers.
Expert-led content sits in the middle of that tension. A founder, researcher, operator, or creator wants to be accurately represented when AI systems answer questions about their work. But accuracy will not come from quietly seeding promotional claims into a bot-facing page. It comes from source material that can stand up when separated from its first format.
What a source record adds
A source record is the internal trail that explains where a public claim came from. For REC-style work, that starts with the research brief, the on-camera answer, the transcript, the supplied sources, and the final review.
This matters because machine-readable publishing can flatten different kinds of language into the same page. A reported fact, a brand claim, an expert opinion, a customer quote, and a sponsored line can all become text blocks. An AI system may parse them as candidate evidence unless the surrounding structure makes the difference clear and the publisher keeps the record behind it.
A recorded interview gives teams a stronger starting point. The person said the thing. The transcript can be checked. The editor can see the question that produced the answer. The public source links can be attached to the claim. AI can still help with research preparation, transcript cleanup, summaries, titles, captions, and format changes, but the team can point back to the human source.
That record is useful even if no AI crawler ever sees it. It helps the team avoid publishing a line that sounds good but has no owner. It helps an editor decide whether a short clip preserved the meaning of the fuller answer. It helps a sales or comms team answer a partner who asks whether a claim is editorial, sponsored, customer-supplied, or generated.
The Perplexity-Time dispute makes the same discipline more urgent. When distribution systems may read a different layer of the web than human visitors read, teams need a stable way to prove what each claim is.
How to publish for agents without writing for agents
The practical response is not to ignore AI discovery. Teams should make their useful material easy to find, parse, cite, and check. The problem starts when the page is written mainly to persuade a crawler while the human evidence stays thin.
First, put the human source upstream. Record the expert answer before turning it into a page, post, clip, or FAQ. If the claim cannot be traced to a person, a document, a product event, or a public source, treat it as draft language rather than evidence.
Second, label claim types internally before writing for the public. Separate observed facts, reported facts, analysis, opinion, recommendations, customer language, and paid placement. This does not require a heavy system. A short note beside each publishable claim is enough if the team actually uses it.
Third, keep public pages consistent. If a machine-readable version exists, it should not carry a materially different claim record from the human page. Format can change. Structure can become cleaner. The source and disclosure context should not quietly change.
Fourth, make AI assistance specific. A disclosure that says AI was used is less helpful than a record that says AI drafted interview questions from supplied links, cleaned a transcript, suggested clip titles, or adapted a reviewed answer into a short post.
Fifth, review the final asset against the source, not against the desired positioning. The strongest sentence in a piece should still be supported by the interview, source document, or public record after the format changes.
What teams should avoid
Do not treat machine-readable pages as a side channel for claims the team would not put in front of readers. If the claim is true, useful, and properly sourced, it should be safe to show. If it needs to hide in a bot-facing version, the trust problem is already visible.
Do not assume a sponsored label will travel with the claim. Labels can be stripped, summarized away, ignored by a crawler, or left out of the final answer. Keep the advertising status in the source record and make the public presentation clear enough that a person can see it too.
Do not confuse brand accuracy with editorial evidence. A brand-approved sentence may be accurate in the sense that the brand stands behind it. That is different from a sentence grounded in reporting, customer proof, expert experience, or a recorded explanation. Both can belong in marketing, but they should not be mixed without context.
Do not let AI visibility work replace the interview. Visibility work can tell a team how it appears in answer systems. The interview tells the team what it can responsibly say. REC's bias is to start with the human answer, then adapt it for distribution.
The practical takeaway
Perplexity's block on Time's agent ads is timely because it turns AI discovery from a measurement problem into a provenance problem. Answer systems need to know what kind of claim they found, not only whether a page is easy to retrieve.
For expert teams, the useful response is simple. Record the answer. Keep the transcript. Attach the sources. Mark what AI changed. Separate analysis from reported fact and paid placement. Keep the machine-readable version aligned with the human-readable version.
AI search and agent traffic will keep pushing publishers to make pages easier for software to read. That is fine. The source record is what keeps the work honest when the software reads it out of context.