ChatGPT Ads Have Arrived, But Paying for the Slot Won't Earn You the Citation

ChatGPT Ads Now Show Up On a Quarter of Commercial Prompts#

Every marketer chasing the sharp scent of a shifting SERP, half a step into the brush after any signal that moves, has asked some version of the same question this year: does chatgpt have ads, and if they're real, how often do they actually show up. SE Ranking answered it with a patient, prowling dataset: 50,006 commercial prompts spread across 20 niches at roughly 2,500 apiece, collected on 23 July 2026 and judged for relevance with embeddings, cosine similarity, and a manual review of every flagged edge case (SE Ranking, 2026). Published on 10 August and corroborated the same day by Search Engine Land, 2026, the headline number is unambiguous: ChatGPT ads appeared on 25.94% of commercial prompts, roughly one in every four.

That figure alone would justify a folkfox client meeting. What makes it sharper is the comparison. SE Ranking ran an almost identical methodology against Google's AI Mode eight weeks earlier: 50,032 ad-eligible commercial keywords across the same 20 niches, collected 30 June 2026, and found text ads on 29.45% of queries, close to one in three (SE Ranking, 2026). Search Engine Land, 2026 covered that study too, and the takeaway holds across both write-ups: two different answer engines, two studies eight weeks apart, and the ad-presence rates land within four points of each other. ChatGPT advertisements have arrived at genuine, unmissable scale, faster than most paid media teams have built a proper playbook for them.

Scale is only half the story though. The other half, the one that should worry any brand mid-brief on an ad campaign in this channel, is what happens once the ad actually lands.

One in Seven ChatGPT Ads Miss the Mark Completely#

SE Ranking didn't stop at presence, it also tested relevance, and the results explain why so many users report chatgpt advertisements that feel oddly beside the point. Across the full dataset, 14.35% of ads carried no meaningful connection to the prompt that triggered them, a rate that swings wildly by niche. Pets sat lowest at just 2.6% irrelevant; News & Politics sat highest at 54.2%, with Relationships close behind, both comfortably above 50% (SE Ranking, 2026). Search Engine Land, 2026 put a face on the mismatch: a dating app prompt paired with a clothing retailer's ad, a newspaper subscription query triggering an electricity provider.

Why the mismatch happens#

There's a mechanical reason for this undergrowth of nearly-right signals, and OpenAI documents it plainly. Advertisers don't buy exact-match keywords in ChatGPT ads the way they might in a search campaign. Instead, at the ad-group level, they submit context hints, natural-language descriptions of the conversations where their product might fit, and OpenAI is explicit that "these hints help guide ad matching, but they are not exact-match keywords and do not guarantee delivery in specific conversations" (OpenAI, 2026). Selection then runs on expected relevance and outcomes, blended with a relevance-weighted, second-price auction, a system built to be probabilistic rather than precise. Loosen the leash on exact matching and some ads will wander off the trail. Fourteen percent of the time, in this dataset, they wandered a long way off it.

None of that irrelevance is the real headline though. The real headline sits one layer deeper, in what an ad slot does and doesn't buy once it's live.

The Decoupling: Why ChatGPT Ads Don't Buy You the Citation#

Here is the finding that should reorder every GEO and paid-media conversation folkfox has about chatgpt ads this quarter. Across the whole SE Ranking dataset, advertiser domains appeared among ChatGPT's own cited sources in just 3.63% of ad placements. Push the bar higher, to an exact URL match rather than just the domain, and the figure collapses to 0.09%, twelve cases out of the entire study. Even a soft measure, the advertiser's brand simply being named somewhere in the answer text, only hit 4.44% (SE Ranking, 2026; Search Engine Land, 2026). Paying for the ad slot and earning the citation are, for practical purposes, two separate campaigns hunting on two separate trails. Almost nobody is managing to work both at once.

This isn't a ChatGPT quirk either. SE Ranking found much the same split when it ran Google's AI Mode through the same test: advertiser domains showed up among AI Mode's cited sources 11.53% of the time, and the exact URL matched just 1.95% of the time (SE Ranking, 2026). Google's overlap runs roughly three times higher than ChatGPT's, but it's still a small minority of ad placements: only 2% of paid URLs also ranked organically for the same keyword, so 98% of AI Mode's ad-triggering keywords saw no organic ranking from the advertiser at all (Search Engine Journal, 2026). A second, independent tracking effort landed in almost the same place on ChatGPT specifically: Seer Interactive's own running audit of ChatGPT ads found that only 3.3% of the ads it tracked cited the advertiser as a source, and aggregator-style advertisers controlled roughly 11.6% of the volume it observed, a scale not far off SE Ranking's 13.56% for the single advertiser BestMoney (Seer Interactive, 2026). Two firms, two different sampling methods, and the citation-overlap number keeps landing under 4%.

Academic researchers building generative-engine-optimisation frameworks have been mapping the same territory from the other direction: what does earn a citation, if payment doesn't. A citation-behaviour study spanning 602 controlled prompts across ChatGPT, Google's AI Overview and Gemini, and Perplexity, built on more than 21,000 valid search-layer citations, found that ChatGPT specifically cites fewer sources than its rivals but grants the sources it does cite substantially higher influence over the final answer (arXiv, 2026). A separate study of citation behaviour across Brave Summary, Google AI Overviews and Perplexity, auditing 1,100 cited URLs, found that overall page quality, not payment, was the strongest predictor of citation, with pages scoring well on metadata freshness, semantic HTML and structured data clearing the bar most reliably (arXiv, 2026). Neither paper set out to study advertising specifically, but both describe, from the content side, the same mechanism SE Ranking's ad data describes from the media side: the quiet quarry worth chasing, the citation, is won on quality and structure, not bought with a budget.

Which raises the obvious follow-up question for any brand budgeting into 2027: if citation is the real prize, why are some categories buying so much of this inventory anyway.

Healthcare Tops the ChatGPT Ads Density Table, and That's No Accident#

Ad density, how often a category's prompts carry an ad at all, tells a different story from relevance. Healthcare sits highest of any vertical SE Ranking tested, at 28.69% of prompts carrying ChatGPT ads, ahead of Finance at 15.21%, Travel at 14.96% and Education at 14.18% (SE Ranking, 2026). That's a striking result for a category OpenAI itself treats as sensitive by default. It only makes sense once you read the fine print OpenAI has been quietly rewriting all year.

OpenAI's ad policy launched in March 2026 with medical, legal and financial advice blocked wholesale. One month later that blanket rule softened: the April 2026 changelog states plainly that "medical, legal, and financial advice contexts are no longer categorically blocked from ads by default," with sensitive personal conversations still ineligible but the wider category opened to case-by-case, manually reviewed approval (OpenAI, 2026). By the policy's own current wording, updated 10 August 2026, OpenAI "may approve ads from approved advertisers within the financial services, healthcare & medicine, and legal services categories," rolled out gradually. Healthcare didn't top the density table by accident, it topped it because the thicket of restrictions around it thinned months before this study ran, and healthcare advertisers moved fastest into the clearing.

Finance shows the same pattern from a different angle, this time through concentration rather than density. One advertiser, BestMoney, accounted for 13.56% of every ad SE Ranking logged across the whole study, comfortably the single largest advertiser presence recorded. Within Finance specifically, Robinhood took 52.24% of that niche's ad volume; within Travel, Booking.com took 62.57%; within Food & Beverage, Hungryroot took a startling 79.92% (SE Ranking, 2026). Whichever measure you trust, a handful of well-capitalised advertisers are currently buying an outsized share of the den, and smaller regulated brands, the kind folkfox works with across healthcare and fintech accounts, are competing against that concentration, not against an open field.

How ChatGPT Ads Pricing and Targeting Actually Work#

Strip away the study for a moment and answer the practical question folkfox clients ask first: what does a chatgpt ads example actually look like, and what does it cost. Every ad OpenAI serves today appears below the model's own answer, never inside it, and carries six fixed elements: advertiser name, favicon, headline, description copy, landing page and a single image asset (OpenAI, 2026). Plus, Pro, Business and Enterprise subscribers never see one; ads currently only reach logged-in adults on the free and Go tiers, and OpenAI excludes any account it identifies, or predicts, as belonging to someone under 18.

The pricing mechanics#

ChatGPT ads pricing runs on the two buying models most paid-media teams already know: cost-per-thousand-impressions for a Reach objective, or cost-per-click for a Clicks objective, with advertisers setting a maximum bid at ad-group level and OpenAI recommending a starting range of $3 to $5 per click. Winners are chosen through what OpenAI calls a relevance-weighted, second-price auction (OpenAI, 2026). SE Ranking didn't just study other brands' campaigns from a distance either, it ran a live test of its own: a Reach campaign that pulled more than 97,000 impressions, produced 1,263 clicks, and landed a 1.30% click-through rate, alongside minimal sign-ups and real attribution headaches tied to how narrow the ad-eligible audience currently is (SE Ranking, 2026).

None of that reads as disastrous for a young ad product, and 1.30% sits in a defensible band next to typical search click-through rates. What it does confirm is scale without maturity: a real auction, real pricing, real measurement inside Ads Manager Beta, running on an audience that's still a fraction of ChatGPT's total user base and a targeting system built on soft context hints rather than hard keyword matching.

What the ChatGPT Ads Data Means For Your Paid Media Budget#

None of this argues against testing ChatGPT ads, folkfox runs them for clients already. It argues against treating an ad slot as a shortcut to the thing GEO and AEO work is actually meant to earn: being the source the model reaches for unprompted. Spend and citation are different budgets, chasing different quarry, and this study is the clearest evidence yet that paid media and AI-visibility work need to be reported on separately rather than one standing in for the other's success metric.

That's the case for running paid ChatGPT campaigns and content-side GEO work as two coordinated but distinct efforts, not one channel doing double duty. It's a case folkfox has made before, most recently when a fortnight of new data showed the same model that cites a brand constantly can still decline to ever recommend it by name, as covered in The model cites you everywhere and recommends you nowhere, and again when tracking-instability data made the point that one snapshot of any answer engine, paid or organic, should never set a strategy, the argument behind A single reading is a rumour, not a measurement. ChatGPT ads deserve a properly budgeted test, run through folkfox's own paid media practice and read on their own terms, alongside a separate, patient programme of the structured, source-worthy content that earns citation rather than buying a slot beside it.

Whichever way a client leans, the decision deserves more than a single headline metric. The steps below are the ones folkfox actually runs through with a client before recommending a paid ChatGPT budget, sensitive category or not.

Frequently asked questions#