Publishers have started selling advertising designed to be consumed by AI systems rather than human readers. The idea is simple enough to attract attention: if brands are worried about whether ChatGPT, Gemini, and other answer engines mention them, perhaps paid placement inside machine readable content can influence what those systems retrieve.
An August 7 Digiday report examines that emerging market and the skepticism around it. Time has begun inserting sponsored FAQ style ads into markdown versions of its pages, with the inventory designed for AI systems and agents to read.
Some media buyers are willing to test the idea. Others see it as speculative because there is no established evidence that buying these placements will reliably change future AI answers.
For marketers, this is a useful test case for a broader AI search visibility question: which parts of AI discoverability can be influenced through paid media, and which still depend on broader brand and information ecosystems?
Markdown ads are trying to create a new media surface
Time's experiment follows a broader move toward machine readable versions of webpages. Markdown strips away much of the design and interface around an article and presents the underlying text in a simpler structure that agents can process.
Time is working with Mobian to place sponsored material into those machine readable pages. According to Digiday, the ads are formatted as FAQs containing brand information and messaging and are labeled as sponsored content.
The commercial logic is understandable. If AI agents increasingly consume publisher content, publishers want a way to monetize that consumption. Advertisers, meanwhile, want more influence over the information AI systems use when users research products and brands.
Traditional advertising buys access to people. Markdown agent ads are testing whether marketers can also buy access to the information environment surrounding an AI system.
That is a very different proposition, and the evidence standard should be higher than simply showing that a crawler fetched the sponsored text.
Being crawled does not prove an ad changed an answer
The central measurement problem is causality.
A publisher can show that an AI agent accessed a page containing an ad. That does not establish that the sponsored material was used in an answer, that the brand was mentioned because of the placement, or that any resulting mention affected a buying decision.
Digiday's reporting reflects that uncertainty. Several buyers described the channel as worth testing only if results can be demonstrated. Others argued that the underlying idea resembles magical thinking because AI developers could ignore or discount these placements at any time.
This is similar to the evidence problem surrounding many new GEO tactics. The Mustard Seed guide to best practices for GEO emphasizes content clarity, authority, evidence, and useful information because those are durable inputs even when specific platform behavior changes.
A paid placement may create another input. It should not be assumed to override those fundamentals.
Brand equity may matter more than a single machine readable placement
Digiday cites WARC research conducted by Charlie Oscar estimating that 63% of a brand's visibility came from long term brand equity, while 26% came from current marketing activity. One media buyer used that broader point to argue that marketers might be better served investing in brand building than trying to engineer a shortcut into LLM outputs.
That argument is worth taking seriously.
AI systems synthesize information from multiple sources. A brand with years of coverage, customer discussion, reviews, product documentation, search demand, trusted citations, and strong category recognition provides far more signals than one sponsored block inside one publisher page.
This does not make experimental inventory useless. It changes what success should mean. A test should ask whether the placement adds measurable incremental visibility relative to the brand's existing baseline.
Without that baseline, any later ChatGPT mention can be incorrectly credited to the media buy.
The wider GEO marketing challenge is therefore closer to brand distribution than to classic paid search. Marketers need owned content, earned references, useful third party information, and clear positioning working together.
Media buyers are treating this as an experiment, not a channel
One of the clearest conclusions from the Digiday article is that buyers are not yet treating markdown ads as a mature media channel.
Some said they would test the inventory, especially for high consideration categories where AI systems may play a larger role in product research. Potential budget could come from programmatic display or experimental funds rather than established search allocations.
That distinction matters. A mature channel has understood inventory, standardized measurement, repeatable buying practices, stable policies, and enough historical data to set expectations. Agent focused ads currently have very little of that infrastructure.
There are also unresolved platform questions. AI developers could decide to downweight sponsored material, ignore machine only variants, or create policies around how promotional information is handled. Publishers could change disclosure formats. Measurement vendors could disagree about whether a mention was influenced by the placement.
A sensible test should therefore be small enough to learn from without pretending the economics are already known.
The right test measures lift against a stable prompt set
A marketer considering markdown ads should establish the measurement framework before the campaign begins.
Choose a defined set of commercially relevant prompts. Record baseline brand mentions, recommendations, citations, and competitor presence across the AI platforms that matter. Keep the prompt wording and sampling method as stable as possible. Then run the media test and look for a directional change relative to the baseline and competitors.
The team should also examine whether the sponsored page itself is retrieved or cited. A brand mention without any relationship to the placement is weak evidence of campaign impact.
This kind of testing fits better inside an omnichannel marketing model than inside a last click acquisition model. AI visibility may influence research and later brand search without producing an immediate referral from the sponsored content.
AI visibility can be influenced, but reliable buying is another question
The most useful line in Digiday's report comes from a media buyer who says brands can influence AI visibility but cannot reliably buy it yet.
That is the right distinction for the current market.
Paid machine readable placements may become a meaningful advertising surface. Publishers have an incentive to build them, marketers have an incentive to test them, and agentic browsing creates a plausible future in which commercial information needs machine readable formats.
But the existence of the inventory is not evidence of predictable influence. Marketers should demand the same thing they would from any emerging channel: a clear mechanism, a measurable outcome, a comparison against baseline behavior, and enough repeatability to justify more budget.
Until then, markdown ads belong in the experiment column, not the guaranteed GEO column.
Originally published on the Mustard Seed blog.
Top comments (0)