There is a version of Time.com you have never read. No photographs, no design, none of the furniture of a magazine. A plain text document, one twenty-third the size of the page you'd see, built for a readership that has no eyes: the crawlers AI companies send out to gather the world so their chatbots can describe it back to you.

In June, TIME converted its entire site into these machine-facing copies.

In July, it started selling ads inside them.

On July 30, TIME and the ad tech startup Mobian announced Agent Ads, the first advertising product built for machines instead of people: sponsored FAQ blocks planted in the markdown, invisible to every human who will ever visit the page. Ally Bank and the Project Management Institute bought first. Mobian's launch post named two AI systems that fetched the debut ad in real time: ChatGPT and Perplexity.

Twelve days later, one of the two named launch guests walked out of the party and told everyone at the door what they thought of the hors d'oeuvres.

Digiday reported August 11 that Perplexity has blocked all markdown advertising on Time.com from influencing its agents, roughly since the practice was first reported, and that its search and security teams are building broader protections against markdown ads generally. Perplexity's chief communications officer Jesse Dwyer said publishers who deploy "deceptive advertising like markdown ads" risk a reputational downgrade in Perplexity's index, including a hit to their trust score.

Sponsored label or not.

The same day, Mobian CEO Jonah Goodhart published a roughly 4,000-word response declaring the reaction since launch "overwhelmingly positive." It handles the possibility of a block in a section titled "Won't models block this, and what happens if they do?"

Note the tense. Might. Hypothetically. Someday, perhaps, in a galaxy far away.

It had already happened. To Mobian. On its only live publisher. From one of the two agents it name-dropped at launch like a wedding invite list. Goodhart's reaction, when Digiday asked: the block was "a bit perplexing."

For the record: ADOTAT has been seeking answers from Mobian for several weeks, and recently sent ten detailed questions about methodology, billing and conflicts. Mobian's response was a pointer to Goodhart's blog post. We re-asked the eight questions the post doesn't address; those remain open, and this story will be updated with any substantive answers. TIME didn't respond to Digiday on the block.

So we read the post. Slowly. Twice.

It is genuinely well-built, better argued than most of what this industry produces in its own defense, and it answers the easy version of every hard question while skipping the hard version entirely. A document titled Initial Response that does not respond to the actual response is not a FAQ.

It is a press release wearing a lab coat.

"Overwhelmingly positive" is a claim about demand. So check it against the people who control demand. Digiday interviewed working media buyers the week after launch, and the transcript reads less like a standing ovation than a crowd quietly retrieving its coats.

The best review on record, from Jeff Eisenfeld of Media by Mother, is a conditional: worth a test "if you know that the ads you're buying are actually going to show up in the LLMs."

Perplexity answered his "if" four days later.

Danny Weisman of Obsessed Media said markdown ads should be publisher added-value, not a campaign line, which is what this business calls inventory when it's being polite about not paying for it. Stephan Kopp of Mediaplus Performance: you can try to influence LLMs with ad content, but "it will not happen." He predicted AI companies would rewire their systems against the tactic the way search engines did.

He said that August 7. Perplexity did it by the 11th.

The two-week ledger: one live publisher. Two advertisers. One conditional maybe. One flat no. One filing under make-goods. And one of two named launch platforms blocking the product.

What Mobian demonstrably has is press coverage, and this industry has never once been able to tell attention from demand, which is arguably why we have this industry.

The buyer who did buy is no fool, which is what makes her numbers so useful. Goodhart's post quotes Ally CMO Andrea Brimmer telling Business Insider "the signals are impossible to ignore": visitors from AI platforms 3.5 times likelier to open an account, AI traffic up 9x. Real numbers. They measure that AI referral traffic converts. They say nothing about whether a paid FAQ in a markdown file caused any of it.

Ally's own record answers that. Per Fortune, Scrunch data shows Ally was the most-mentioned bank in unbranded banking queries across major AI assistants every month from January 2025 through July 2026. Nineteen straight months of winning AI recommendations.

All of them before Ally bought a single Agent Ad.

A WARC study by agency Charlie Oscar explains why: 63% of a brand's LLM visibility comes from long-term brand equity. Current marketing: 26%.

Ally spent a decade building a lighthouse. Mobian sold it a flashlight.

The post's central defense is transparency, and Goodhart has receipts. Sponsor named in the first line. Citation on every fact. Fully inspectable.

All true. All beside the point. And there's thirty years of research on why.

Disclosure science runs on the Persuasion Knowledge Model (Friestad and Wright, 1994) and the CARE model (Wojdynski and Evans). The replicated finding: a disclosure only protects a consumer who recognizes the message as advertising. Recognition is the antibody. No recognition, no defense.

A warning label is not a safety feature when it's printed on the inside of the engine block.

Mobian's label is served to a crawler. The crawler feeds a model. The model writes an answer for the human who asked which bank to consider. The label does not survive the trip. It is consumed en route by the one party it cannot protect and never delivered to the one party it exists to warn.

The research has already reached this exact scenario. A study covered in The Conversation this April found that even when chatbots included disclosed, labeled ads directly in their responses, half of participants didn't notice the advertising language at all. Many preferred the ad-infused answers.

Disclosure fails humans even when it's hand-delivered to humans. Mobian's architecture guarantees it's never delivered at all.

That is either an oversight or the entire business model. The post never says which, because it never mentions that person exists.

Goodhart said the quiet part at launch, in the quote that will outlive the company: "When you influence ChatGPT, you're influencing potentially all of ChatGPT. If ChatGPT changes what it says about [a brand], it's massive and it's more than any one campaign could ever do."

Read that against the response post's insistence that Agent Ads are neutral brand facts a model may weigh or ignore..

Influence when he's selling. Information when he's defending.

Pick a lane, Jonah.

The post says this isn't cloaking because nothing is concealed. There's a stronger version of that defense than the one Mobian wrote, so let's argue with that one.

Vincent Schmalbach, the German developer who found the ads before most of the trade press, documented the routing: ClaudeBot, OAI-SearchBot and PerplexityBot get the sponsored markdown. Googlebot gets the identical HTML a human gets, confirmed byte for byte by developer Umesh Malik on August 6. GPTBot, OpenAI's training crawler, gets an HTTP 406 and nothing at all. Malik's conclusion is serious: because Googlebot and humans see the same page, this isn't cloaking in the sense Google's spam policy means, and the parity is deliberate.

He's right.

Now sit with what that routing table says, because a routing table is a confession written in configuration. The one crawler that could penalize cloaking gets the clean page. The crawlers that answer consumer questions get the paid one. The training crawler, the traffic nobody pays for, gets turned away at the door like it forgot its wristband. Every fork stays millimeters inside the letter of a spam policy written for a different machine in a different decade.

That's not honesty. That's a driver who memorized every speed camera on the highway.

Perplexity just moved the cameras.

As for "everybody's gamed crawlers since crawlers existed": sure. And? That's a confession with a shrug attached. Agent Ads is not a decade-old hack rotting in a legacy CMS. It's a brand-new product, built from scratch in 2026, by the former CEO of a company whose entire reason for existing was independent measurement you could trust. Mobian had the rarest gift in ad tech, a blank page, and the founding decision on it was to fork the response by User-Agent header.

The sins didn't have to be inherited. They were chosen off the shelf, in the original packaging.

Rob Derow of BCG X told Digiday on July 30 that the biggest risk was the absence of any rule on how LLMs treat markdown ads, and that models could come to read the practice as cloaking. Twelve days from caution note to enforcement action.

That wasn't a hedge. That was a forecast.

The post leans hard on provenance. Facts with citations are trustworthy; facts without them are the mud of the open web.

Fine. Apply the standard to the man holding the ruler.

Models get brand facts wrong 34% of the time: Mobian's internal data. 17% of AI-cited sources conflict with brand positioning: a Mobian study, unpublished. 98% of Newsweek's content graded brand-safe versus 63% for keyword blocklists: a Mobian trial, no published methodology, run on a publisher whose own advertising page states "Newsweek collaborates with Mobian" in its ad operations.

The relationship predates the study. An August 2024 Newsweek-side piece in AdExchanger was already praising Mobian and previewing the exact punchline, that inventory marked unsafe is actually safe.

The advocate became the sample. The post discloses none of it.

And note who's standing just outside the frame. The 63% comparator is never named, but Newsweek's advertising page names its verification partner on the same page that names Mobian: Integral Ad Science. Whoever the 63% is, follow the money through the triangle. Safe content is sellable content, so a study finding 35 more points of Newsweek's inventory brand-safe is a rate card expansion. Mobian sells the fix. Newsweek sells the blessed inventory. The only loser is the incumbent category the study indicts, which was not named, not consulted, and not handed a methodology to check. Moat's clients scattered to IAS and DoubleVerify when Oracle shut its ad business; Mobian is Goodhart's re-entry into the market those two now run.

Perhaps the trial was pristine. But an undisclosed commercial relationship between the tester, the tested, and the conclusion needs to be stated, not discovered.

There's a literature on this. A Cochrane review of 75 papers found industry-sponsored studies significantly more likely to favor the sponsor, and names the mechanism that matters here: selective reporting of favorable results. Mobian's honest test, Model Impact, queried models cold across a pre-registered family of questions and reports significant movement on exactly one, described as the question "with the most room to move."

One survivor out of a family. Self-run. Unaudited. Explained after the fact.

Here's the thing about that standard: nobody articulated it better than Goodhart. On The ADOTAT Show in 2024, before Mobian launched, he told us: "Brand safety, we've not gotten right so far. I think we've given brands a false sense of security. We've told them they have insurance, when in fact we've not been doing a good job as an industry on getting that right." His prescription, same conversation: companies should "show their data in a transparent way."

He meant it. We believed him then and we believe him now.

The 34%, the 17%, and the Newsweek 98% are the data. They are not shown.

The kindest thing anyone can say to Jonah Goodhart right now is that 2024 Jonah was right, and his numbers should meet him there.

The impression has the same referee problem. Real impressions are verified by rDNS "depending on the CDN and agent operator."

Depending on. That's not a standard. That's a weather report.

Meanwhile the served pages carry a Mobian header minting a fresh impression ID on every request. Serve the page. Insert the ad. Count the exposure. Define the impression. Score the outcome.

Five hats. One referee. Paid by the home team.

Among our open questions: whether blocked Perplexity traffic is still being counted, and billed, on the Ally and PMI campaigns.

The post cites real research, and the sophistication works against it. Kandpal. Allen-Zhu and Li. Legitimate papers about how many times a fact must appear in a pretraining corpus before a model retains it. All true.

All from a different room in the building.

Agent Ads are a retrieval product: a page a model reads while answering never touches its weights, and the model forgets it the moment the conversation ends, the way you forget a phone number after dialing it. That amnesia is exactly why Mobian's other pillar, real-time updatable brand facts, works at all. The post half-concedes this in a subordinate clause and keeps the conclusion anyway, which is the intellectual equivalent of footnoting your own alibi. And notice which conclusion survives: repetition builds retention, therefore buy more impressions.

The citations aren't scholarship. They're load-bearing for the media plan.

The strange part: the right literature exists and flatters him. The GEO paper, Aggarwal et al., KDD 2024, found that adding citations, statistics and authoritative structure boosted visibility in generative answers by up to 40%. That's the Agent Ads spec sheet, peer-reviewed. Mobian built the top recipe from the actual literature and cited pretraining papers instead, possibly because the literature comes with an asterisk: a 2026 follow-up found the tactics failing to replicate, and Rand Fishkin's 600-volunteer experiment concluded AI visibility tracking is "inherently unreliable" at the query level.

The honest state of the science: nobody knows if this works, and nobody can reliably measure whether it worked.

Including the people selling visibility scores to Ally as deliverables.

Underneath all of this is a real problem, and Goodhart is right about it: TIME sees more bot traffic than human traffic most days, that traffic earns nothing, and wanting to monetize it is rational. The business models have to evolve. We're not arguing otherwise.

But TIME deserves one more sentence than anyone has given it. TIME is a named member of the Cloudflare coalition that flipped the default to block AI crawlers unless they pay for content. That same publisher then built a second website specifically to hand those crawlers sponsored content. Together the two moves say the principle was never permission.

The principle was the invoice.

A toll booth on the front door. A gift shop at the loading dock. Both staffed by the same accounting department.

And the platforms Mobian needs were on the record months before launch. Perplexity killed its own ad program in February, executives telling the FT that even labeled ads make users "suspicious of everything." Anthropic committed, in writing and in a Super Bowl ad, that Claude's responses "will not be influenced by advertisers"; ClaudeBot is one of the three crawlers receiving the sponsored markdown. Every major model provider spent early 2026 staking its brand, at Super Bowl rates, on the integrity of its answers.

Agent Ads asks them to un-stake it. For someone else's revenue.

Which is the structural problem no blog post writes around. Platform economics says a two-sided business survives launch by subsidizing the reluctant side. Mobian can discount publishers. It can discount advertisers. It cannot pay OpenAI, Anthropic or Perplexity anything. The side holding veto power isn't a customer.

Mobian is running a marketplace where the landlord isn't on the lease. And the landlord just changed one of the locks.

What the business looks like in two years is among the questions Mobian hasn't answered.

Strip away the citations and the post makes one final argument: brand facts are good for models, so models that block them will suffer for it. That's a vendor telling platforms that refusing his product will be to their detriment, two sentences after insisting the model can always choose freely.

One platform already chose.

Goodhart identified a real problem, moved first, disclosed more than any policy required, and wrote the most substantive defense this category has. Those things count, and we've printed them.

But there's a conversation from 2024 we keep coming back to. Mobian is named after Mobian Nights, a book by Goodhart's father, and when Jonah came on The ADOTAT Show he lit up describing its central idea: "we cannot sidestep context... there is no meaning to words on a page. It only has meaning through the person that's actually reading them."

It's a beautiful idea. It's also true.

And it is the most precise description anyone has offered of what's wrong with the product carrying the book's name. Somewhere right now a person is asking a chatbot which bank to consider, and the answer may have been shaped by a document on a page they'll never see, carrying a disclosure they'll never receive.

Words on a page. No person reading them.

By the philosophy Mobian is named for, that disclosure doesn't mean anything yet. It's waiting for its reader.

Our questions remain open. The offer to come on the show stands, and we'll print Mobian's answers at length whenever they arrive. The post says anyone can open the file and read exactly what's inside.

We did.

The file was easy. The company is still locked.

How we reported this: Based on Mobian's launch and response posts, Digiday's July 30, August 7 and August 11 reporting, The Register's August 5 story, Vincent Schmalbach's and Umesh Malik's technical documentation of TIME's user-agent routing, Newsweek's published advertising partner page and an August 2024 AdExchanger piece on the Newsweek-Mobian relationship, Fortune's Brimmer profile, and academic literature cited in the text.

ADOTAT has sought comment from Mobian for several weeks and recently sent ten detailed questions; Mobian referred us to its public blog post, and the eight questions it doesn't address remain open. This story will be updated with any substantive response. Figures attributed to Mobian (34%, 17%, 98%) are internal, unpublished, and could not be independently verified. Jonah Goodhart appeared on The ADOTAT Show in 2024, before Mobian's launch; quotes from that interview are used with the context in which they were given. ADOTAT has no financial relationship with any company named in this piece. No subject received advance review of this article.