Perplexity has blocked all markdown advertising on Time.com from influencing its agents and its user-facing search results, and has called the format deceptive. Digiday reported the block on 11 August 2026, less than two weeks after the same publication first reported that Time had begun selling ads aimed at AI agents inside the markdown copies of its pages.
Relevant Audience covered that first story when Time started selling the format. The new part is not that markdown ads exist. The new part is that an AI answer engine has killed one, publicly, and attached a consequence to the publisher that ran it.
What Perplexity blocked, and what it said
Digiday reported on 11 August 2026 that Perplexity has stopped all markdown advertising on Time.com from influencing its agents or the results its users see. Jesse Dwyer, Perplexity's chief communications officer, said the company works "continuously" to protect users from deceptive practices, and warned publishers that deploying "deceptive advertising like markdown ads" risks reputation downgrades and hits to their trust score. The full account is in Digiday's report.
The warning is the part with teeth. A block on one publisher's ad inventory is a commercial dispute. A stated trust score penalty is a site-level visibility consequence, applied by the engine, to a publisher, for the ad format it chose to run. Perplexity's search and security teams are building further defences against markdown ads generally, which means the position is not specific to Time.
How the format works
The ads are built by the ad firm Mobian. Mobian generates FAQ-formatted content from a brand brief, embeds it into the markdown version of a publisher page, and then tracks how often and how favourably AI engines surface it. Ally Bank and the Project Management Institute were early adopters. The markdown copy is the version an AI crawler or agent reads, which is what makes the format interesting to advertisers and what makes it contentious.
Mobian co-founder and chief executive Jonah Goodhart defended it, arguing that "a model forming an answer gets current, sourced, brand-verified information", and that "when models get facts wrong about brands, the people asking the questions get worse answers". That is a real argument and worth stating fairly: brand-supplied facts inside a page can be more accurate than a model's recollection of them.
The counter-argument came from Robert Webster, the former WPP executive who founded TAU, who warned that once sponsorship labels disappear during agent retrieval, a "promotional claim can end up cited as a neutral fact". That is the mechanism at issue. Not the presence of paid content, but the loss of the label between the page and the answer.
The timeline in one place
Everything below is from Digiday's reporting on 11 August 2026 and its earlier coverage of the format.
| Element | What Digiday reported |
|---|---|
| Less than two weeks before 11 August 2026 | Time begins selling ads aimed at AI agents inside the markdown copies of its pages |
| 11 August 2026 | Perplexity blocks all markdown advertising on Time.com from influencing its agents and user-facing results |
| Stated consequence for publishers | Reputation downgrades and a hit to the publisher's trust score |
| Who builds the format | Mobian, generating FAQ-formatted content from brand briefs into the markdown version of a page |
| Named early advertisers | Ally Bank and the Project Management Institute |
Why this one matters more than the first story did
Selling ads to machines was a novelty story. An engine refusing them, naming the practice deceptive, and warning that the publisher's standing in its index will suffer is a governance story, and it is the first public case of it. It establishes that the agent-readable layer of a page is territory the engines intend to police, and that the cost of getting it wrong lands on the host site rather than on the advertiser or the ad firm.
That draws a line through the middle of generative engine optimisation work. On one side sits making a page legible to a model: clean structure, answerable questions, facts stated plainly, machine-readable formats that match the human page. On the other side sits inserting paid claims into the copy the machine reads and the human does not. The first is optimisation. The second is now, at least in one engine's stated policy, a reason to be trusted less.
Anyone doing serious GEO work should treat this as the first data point in a category that had no enforcement precedent last month. The engines had said very little about what they would do with the machine-readable layer. One of them has now acted, and published a rationale.
What Digiday did not report
- How Perplexity detects markdown ads. No detection method was described.
- What the trust score actually is. It is named as a consequence, not defined, and no scale or threshold is given.
- Any response from Time to the block itself.
- Whether Google, OpenAI or Anthropic will take the same line. None of them has said anything on the record in this report.
- Whether the block is permanent, appealable, or reversible if the ads are removed.
Those gaps matter for anyone tempted to generalise. One engine acting is not an industry standard, and a consequence with no published definition is hard to plan around. What is knowable is the direction, and the direction is not friendly to paid insertions in the machine layer.
What to check on your own site
The practical test is simple and does not require any tooling. Fetch your own page the way a crawler does and compare it against what a human sees at the same URL. If a claim, an FAQ block or a mention exists in one version and not the other, you are running the risk this story is about, whether or not money changed hands for it.
The same check applies to llms.txt files, markdown mirrors, and any AI-specific variant of a page. These are usually built with good intentions, to give models a cleaner read of the same information. They become a liability the moment the content diverges from the human page. Keeping the two in sync is a maintenance task that most teams have not assigned to anyone, and it is the kind of thing that quietly drifts once the first version ships.
For content teams, the useful reframing is that the machine-readable copy is published content with the same standards attached. If a claim would need a sponsorship label on the human page, it needs one in the markdown copy, and the fact that a label may not survive retrieval is an argument against the placement rather than an argument for hiding it. That principle belongs in the content marketing brief, not in a technical backlog.
What this means for Thai marketers
This section is analysis. Digiday reported nothing about Thailand or any Thai publisher, and nothing below is attributed to the source.
Thai publishers and brands experimenting with agent-readable page variants and llms.txt style files now have a live example of an engine penalising paid insertions into that layer. The experiments themselves are reasonable, and structuring content so a model can read it accurately is ordinary work covered by AI SEO practice. The line sits at paid placement inside the machine-readable copy.
The rule to take away is short. Content served to AI crawlers should match what a human sees on the same URL. Paid placement inside the machine-readable copy is a visibility risk rather than a shortcut, and the risk sits with the site that hosts it. Local publishers considering the format as a revenue line should price in the possibility that an engine removes their citations entirely, which is a worse outcome than the ad revenue is likely to be worth.
FAQ on the Perplexity block
What exactly did Perplexity block?
All markdown advertising on Time.com, from influencing both its agents and its user-facing search results, as reported by Digiday on 11 August 2026. Perplexity also said its search and security teams are building further defences against markdown ads generally, so the position is not limited to Time.
Are markdown ads against any rule?
No published rule was cited in the report, which is part of why the case is notable. Perplexity called the format deceptive and warned of trust score consequences, but Digiday reported no policy document, no detection method and no definition of the score.
Will Google, OpenAI or Anthropic do the same?
The source did not say, and none of them commented in the report. Treat Perplexity's position as one engine's stated policy rather than an industry standard, while noting that the argument it made is not specific to Perplexity.
Does this affect my site if I publish an llms.txt file?
Not by itself, since the issue in this story is paid content inserted into the machine-readable copy, not the existence of a machine-readable copy. The risk starts when what a crawler reads stops matching what a person reads at the same URL.
Has anything like this happened in Thailand?
Nothing in the source relates to Thailand, and no Thai publisher was named or studied. The Thai section of this post is labelled analysis of what the same rule would mean locally.
If you want a review of what your pages look like to an AI crawler, and whether the machine-readable version says the same thing as the page your customers read, Relevant Audience can take a look.
