The syndicated Stacker briefing that landed on KESQ on September 11, 2026 tells brands to treat ChatGPT ads and generative engine optimization as two separate paths. That is the 2010 search playbook pasted onto a 2026 surface. Paid ChatGPT ads buy a labeled card under the answer. GEO tries to get a brand named inside the answer. Those are not two routes to the same slot.

  • ChatGPT Ads are a conversation-matched unit that sits below the response and, by OpenAI's own rule, does not change the answer.
  • SE Ranking's August 2026 study of more than 50,000 commercial prompts found the advertiser in that slot was absent from the answer copy in 96.37% of placements.
  • McKinsey's AI-search work says half of surveyed consumers now seek AI-powered search on purpose, and puts $750 billion of US revenue through that surface by 2028.
  • The open-web habit of buying the paid unit and optimizing the organic listing does not transfer, because the organic listing is now a citation drawn from publisher pages the brand does not own.
  • Brands that only fund the card under the answer still lose the recommendation. Publishers that supply the pages behind the recommendation still are not in the buy.

What conversation-native advertising actually is

Conversation-native advertising is a sponsored message matched to the intent of a live chat, not to a keyword list. David Dugan, OpenAI's head of global ads solutions, has been explicit: it is not a keyword-based buy, and it is different from traditional search. OpenAI's February 9, 2026 test post, updated through the year, repeats the same product rule — ads stay labeled, sit after the response, and are not supposed to alter what ChatGPT says.

That rule is the product. It is also the gap. The commercially valuable event on this surface is the recommendation inside the answer, not the card beneath it. Buying the card does not purchase the sentence users came to read.

The SERP's two slots and the answer engine's two inventories

In the open-web search economy, paid search and organic search shared a page. The advertiser bought a rank. The publisher earned a click. Attribution ran through the same billing event. Both sides of the market could see the money move.

In the agentic-web economy the page splits. One inventory is the labeled unit under the answer, now sold by OpenAI and, as of September 10, 2026, also reachable through Amazon DSP. The other inventory is the citation inside the answer, assembled from publisher pages the engine ingested. That second inventory is what GEO chases. It is not priced, not sold, and not paid back to the source.

What the Stacker briefing told brands, and what we verified

The KESQ piece, produced by elk Marketing and distributed by Stacker on September 11, 2026, argues that brands should start small on ChatGPT ads, keep product data accurate, and run GEO as the unpaid path. It cites McKinsey on shoppers seeking AI-powered search, Viant CEO Tim Vanderhook on the funnel collapsing into one conversation, and eMarketer analyst Jeremy Goldman on users leaving if ads feel clumsy or opportunistic.

The McKinsey claim holds, with a tighter citation: in "New front door to the internet," McKinsey reports that half of consumers polled now intentionally seek out AI-powered search engines, that 44% of those users call it their primary source of insight against 31% for traditional search, and that $750 billion in US revenue will flow through AI-powered search by 2028. Unprepared brands, the same paper warns, face a 20 to 50 percent drop in classic search traffic. Stacker's "start of 2026" test date is looser than OpenAI's own record: the company announced the test on January 16 and began serving ads to US Free and Go users on February 9. By August 31, OpenAI said ChatGPT Ads had reached a $1 billion annualized run rate in under 200 days.

Why keeping ads out of the answer is not a complete media plan

The trust case for a labeled unit under the answer

The strongest version of Stacker's advice is the trust case. OpenAI has spent 2026 telling advertisers the answer will not be for sale. McKinsey's August 27, 2026 consumer work found generative-AI trust still below 40 percent, even as people keep using the tools. Goldman's warning is rational: a clumsy ad on a help surface is easier to flee than a clumsy ad on a SERP. If the product promise is "we will not sell the sentence," then paid and organic really do look like two paths.

Why the labeled unit does not purchase the recommendation

The data does not support treating those paths as substitutes. SE Ranking's August 2026 analysis found ads on 25.94% of commercial prompts — close to the 29.45% it measured on Google AI Mode — and found that in 96.37% of those placements the advertiser was not cited in the answer. About one in seven ads was off-topic. Paying for the card bought presence under the response. It almost never bought the mention that does the recommending.

AI commerce collapses the search → click → checkout funnel into a single, intelligent conversation.

— Tim Vanderhook, CEO, Viant Technology

If the funnel has collapsed into the conversation, the sentence inside the conversation is the media. A card that is structurally forbidden from entering that sentence is a remnant of the old SERP, not a replacement for it. GEO does not close the gap either. It is unpaid work on pages the brand does not own, performed so an engine can keep quoting publishers for free.

What this means for brands and for publishers

For CMOs and media buyers: buy the card, measure the citation

Treat ChatGPT Ads as a testable demand channel, not as AI Search visibility. Ringfence a line for the labeled unit, then instrument whether the brand is named in the answer on the same prompts. Last-click from the card will flatter the buy and hide the miss. The KPI that matters is citation inside the response, not only clicks beneath it. Agencies that sell "paid plus GEO" as a complete plan are selling the 2010 bundle on a surface that no longer shares a page.

For publishers: GEO advice is a bid on your pages with no price

Every brand brief that says "keep product data accurate so the model describes us correctly" is a request to keep using publisher reporting, reviews, and explainers as unpaid raw material. Traffic recovery will not price that role. Citation will — but only if the answer itself becomes inventory with a buyer, a seller, and a split. Until then, GEO is demand for your authority with no insertion order attached.

Three signals the two inventories will not stay separate

Watch three things over the next 18 months. First, whether in-chat checkout and product feeds pull commerce into the answer while the ad unit stays parked underneath it. Second, whether buyers start paying for citation share rather than only for the labeled card — SE Ranking's 96% disconnect will not survive as a secret. Third, whether any platform other than the engine itself builds a rail that pays the publishers whose pages already sit inside the response. The category forms at that third signal, not at another DSP plugging into the slot below.

Conclusion

Hold on to this: ChatGPT ads bought the slot under the answer, and GEO still rents the answer from publishers who are not on the insertion order. The rational category is generative engine advertising — native ads for AI agents that place brands inside the response and open a revenue stream for the media sources those agents already cite. What to watch next is whether the first priced citation is sold by the engine alone, or by a network that finally puts publishers in the buy.