ChatGPT Ads Sit Next to Answers GEO Cannot Buy

OpenAI said on August 11 that ChatGPT ads are live in South Korea, Japan, the United Kingdom, Mexico, and Brazil. The units sit under the reply and are marked sponsored. The company says they do not change the answer.

Generative engine optimization, or GEO, is the practice of getting a brand cited inside that answer. Marketing firms in Seoul and abroad are selling it as the next SEO, while the platforms auction the leftover space beside the text.

ChatGPT Ads Sit Beside Answers They Cannot Buy

The test began in the United States on February 9 for logged-in adults on Free and Go. Plus, Pro, Business, Enterprise, and Education plans stay ad-free. A May 7 note named the five markets that later went live, and the August 11 update confirmed they had.

OpenAI’s public rule is blunt. The company says ads do not influence ChatGPT answers, and that sponsored units are always labeled and kept visually apart from the organic reply. Advertisers see aggregate views and clicks. They do not see chats, memories, or personal details.

Matching, during the test, uses the topic of the conversation, past chats, and past ad taps. A recipe thread can pull a meal-kit ad. Several advertisers on the same topic compete on relevance, not on rewriting the model’s reply. Users can dismiss an ad, turn off personalization, or pay to drop ads in exchange for fewer free messages.

THE RULES ON THE AD UNIT

• Who sees them: Logged-in adults on Free and Go; paid and education tiers do not.
• What they cannot do: Change the answer, read a private chat, or target people OpenAI predicts are under 18.
• Where they stay out: Health, mental health, and politics, with more sensitive topics held back as the test grows.
• What advertisers get: Totals for views and clicks, not a transcript of the session that triggered the ad.

That design is why a GEO retainer and a ChatGPT insertion order are not the same product. One tries to become a source the model quotes. The other buys a labeled box under a quote the buyer is not allowed to write.

THE CHATGPT ADS CALENDAR

1. February 9, 2026: OpenAI starts the U.S. ads test on Free and Go.
2. March 26, 2026: The company says trust scores held and names Canada, Australia, and New Zealand as the next pilots.
3. May 7, 2026: The United Kingdom, Mexico, Brazil, Japan, and South Korea are listed for the following wave.
4. August 11, 2026: OpenAI says ads have launched in those five markets and that more countries will follow this year.
5. September 10, 2026: Amazon Ads opens a U.S. path for select advertisers to buy the same inventory through its own desk.

Health brands already live with a sharper version of the split. OpenAI will not run ads next to medical topics, so a clinic that wants to appear in a ChatGPT reply has no paid unit beside that answer and no paid unit inside it. Citation is the remaining door, which is why GEO pitches land hardest in categories the ad product refuses.

Citation for Sale on the Korean Side

Korean ad-tech shops moved first on the unpaid door. Enliple launched i-GEO in May to score a site’s chance of being cited, reshape pages for generative systems, and write supporting copy. In a pilot with a large domestic duty-free retailer, ordinary search traffic rose 21.9% after the work, AI-driven traffic rose 37.3%, and purchase conversions rose 63%. Enliple also said AI traffic ran 9 percentage points above rival duty-free operators, a comparison against competitors rather than against its own search lift.

Plus Zero shipped Gewriter on May 13 as an on-site engine that rewrites official brand pages into forms search systems and models parse more easily, automating consulting that used to be done by hand. PK Communication sells diagnostics across seven buckets, among them crawler permissions, structured data, and off-site citation signals. KPMG Korea, in a May 28 note on how AI is reshaping content businesses, called GEO tactics a new core advantage and said firms would have to write for machines that reason, not only for people who click.

Akamai, the U.S. CDN and security company, put a similar idea at the edge. AI Brand Presence, announced May 19, serves a machine-readable copy of a site without a CMS rebuild and logs which agents visit. Used on Akamai’s own pages, the company reported an 85% increase in citations, a 364% jump in brand presence on general searches that did not name Akamai, a 133% ChatGPT presence gain against rivals, and machine payloads cut by 99%. It also said AI bot traffic on the open web was up more than 300% year over year. The product is in limited release with selected North American customers.

GEO PRODUCTS NOW ON THE MARKET

Those figures are first-party tests, not a shared industry audit. They still explain the retainer math. Dimension Market Research puts the global GEO market at $1,089.3 million in 2026 and, at a 40.6% compound annual rate, $17,148.6 million by 2034. Samsung Securities wrote in May that keyword ads and SEO were already expanding into AI-reply work. The sales pitch is simple enough to print on a slide: if the answer is not for sale, buy a better chance of being the source it rests on.

What the Princeton Paper Measured

The term GEO did not come from an agency deck. Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, and Ameet Deshpande, working across Princeton, IIT Delhi, and independent labs, posted the paper in November 2023 and presented it at KDD 2024. They treated ChatGPT-class tools as generative engines that pull several sources and write one reply, then asked how a publisher could raise the odds of appearing in that reply.

The abstract is the line the market repeats. Their methods, they wrote, can boost visibility by up to 40% in generative-engine responses, and the gains change by domain. They also released GEO-bench, a large set of queries and source pages, because the engines themselves are black boxes. Content authors, the paper argues, otherwise have little control over when or how their work is shown.

That last point is the part vendor demos skip. A 40% lift is a relative move in a lab-style pipeline, not a promise that ChatGPT will send a shopper. Summaries of the paper’s tests keep circling the same content changes, adding citations, quotations, and hard numbers, and they also flag keyword stuffing as a tactic that can hurt. None of those edits buys the ad unit OpenAI is rolling out. They try to make a page easier to quote, which is a different job from ranking ten blue links.

Eight Percent Click Through When the Summary Appears

The commercial case for GEO still assumes a visit. Independent click data on Google’s AI summaries, the closest large study of what people do after an AI answer, points the other way. Pew Research Center tracked 900 U.S. adults in March 2025, rebuilt 68,879 Google searches, and found an AI summary on 12,593 of them.

Users who saw a summary clicked a result link in 8% of visits. Users who did not see a summary clicked in 15% of visits, nearly twice as often. A click on a link inside the summary itself happened in 1% of those visits. Sessions ended on 26% of pages with a summary, against 16% of pages that had only classic results. About 18% of all searches in the set produced a summary, 88% of those summaries cited three or more sources, and the median summary ran 67 words. Some 58% of the adults hit at least one AI summary that month.

HOW THE CLICKS FELL IN PEW’S GOOGLE PANEL

• Result click with a summary: 8% of visits, against 15% when no summary appeared.
• Click inside the summary: 1% of visits, the citation footer almost unused.
• Session over: 26% of summary pages, against 16% of classic-result pages.
• How common: 18% of the 68,879 searches carried a summary; 58% of adults saw at least one.

Enliple’s duty-free pilot can still be true on its own site. A retailer that feeds models cleaner product copy can pick up AI-referred sessions that convert. Pew is measuring a different surface, Google’s summary at the top of a results page, where the summary is often enough and the cited link is decoration. GEO vendors are selling visibility on a page people increasingly do not leave.

Naver Now Pays Creators for AI Citations

South Korea’s own search company turned that visibility into a paycheck. Kim Kwang-hyun, Naver’s chief data and contents officer, said on May 28 that the firm will put 1 trillion won ($663 million) into its creator base over five years, because competition among AI platforms now runs through data quality. Naver Mate, which followed in June, names about 3,000 creators a month from Blog, Cafe, Knowledge iN, and Premium Contents, ranked largely on how often AI Briefing quotes them. Cash runs from 300,000 won to 10 million won a person, inside a 20 billion won yearly support budget.

The AI Tab, a conversational layer on Naver search, passed 3 million users within a month of its April beta and was set for a full rollout in June. AI Briefing, the summary that sits above classic results, is the surface Mate pays against. Writers who used to chase page views now chase a citation counter on the profile. First-hand detail, prices paid, meals eaten, trips taken, is what the program rewards, because a model can already fake a generic explainer.

That is GEO with a wage. Naver is not selling a dashboard to brands so much as paying its own UGC bench to become the corpus AI Briefing trusts. OpenAI, by contrast, sells a sponsored unit under a reply it refuses to alter, and tells advertisers they cannot buy the sentence itself. Two platforms, two prices for two different objects, both called AI search.

The DSP Seat Under the Chat Reply

Amazon Ads announced on September 10 a partnership with OpenAI to support advertising in ChatGPT. Select U.S. advertisers are already in a pilot, extending campaigns they already run on Amazon into the chat. Chris Conetta, director of omnichannel supply at Amazon DSP, called conversational ads the fastest-growing way for brands to reach people who are already comparing options. Katrin Koenig, president of Delta Vacations, said the travel brand is among the first testers and wants the ads to appear when someone is planning a trip.

Buying through Amazon does not let a brand rewrite the reply. OpenAI still decides which ad sits under which answer. The deal simply puts ChatGPT inventory on a desk that already buys the rest of the open web. GEO agencies have spent the year selling the unpaid half of that screen. The paid half now has a DSP.

The assumptions underpinning 20 years of digital strategy are breaking down in real time. We’re witnessing one of the most significant shifts in digital history, a move from human-driven search to machine-mediated discovery. Businesses face the risk of digital invisibility. If you’re not the primary source of truth for the AI models your customers trust, you effectively don’t exist.

Patrick Sullivan, CTO of Security Strategy, Akamai, May 19, 2026 press release

Google’s public line runs the other way. Brendon Kraham, vice president of search and commerce for global ads solutions, wrote in June on Think with Google that new generative features rest on the same ranking systems as classic Search, pull from the same index, and do not need a new playbook. Good SEO is good GEO, he argued, and brands should stop chasing fake mentions and copy written for bots. That is an incumbent telling marketers the acronym is a rename.

The GEO desks will keep selling citation scores. OpenAI will keep the answer independent and the sponsored unit separate. Amazon will keep adding buyers under that unit. Naver will keep paying writers to be the quote. The thing none of those checks can purchase is the sentence ChatGPT actually says.