The most expensive assumption in generative engine optimization right now is that paid placement inside an AI answer will eventually pull citation along with it. It will not, and there is now a large enough dataset to say so with a number attached rather than a theory.
SE Ranking ran 50,006 commercial prompts across 20 US niches and published the results on August 10, 2026, authored by Yulia Deda. The finding that matters most to anyone building AI visibility: 96.37% of the advertisers appearing in those results were not cited as a source in the answer they appeared beside. Only 3.63% managed both.
What the study measured
Start with the scope, because the number only means something if the shape of the sample is clear. The study covered commercial-intent prompts, which is the population where ads exist at all, across twenty US niches. Ads surfaced on 25.94% of those commercial queries, so about three quarters of commercial prompts returned no ad in the first place.
| MEASURE | FINDING | READING |
|---|---|---|
| Prompts analyzed | 50,006 commercial prompts, 20 US niches | Large enough to treat the headline ratios as stable |
| Ad presence on commercial queries | 25.94% | Monetization is still selective, not universal |
| Advertisers also cited as a source | 3.63% | Paid and earned qualification are effectively independent |
| Ads with no topical relevance | 14.35% | Targeting is immature, not just conservative |
| Off-topic ads, News and Politics | 54.2% | Worst category measured in the study |
| Off-topic ads, Relationships | 51.1% | Second worst, and both are over half |
Two caveats before anyone puts this in a deck. This is a snapshot of a young ad product that is changing month to month, so the ratios describe August 2026 and not a stable equilibrium. And a study of prompts run by a research team is not a study of prompts run by your buyers, so treat the niche-level splits as indicative rather than as a targeting plan.
Why ChatGPT ads and citations are separate systems
The independence is structural, and understanding why is more useful than memorizing the percentage. An ad slot is sold. It qualifies on bid, targeting configuration, and policy compliance. A citation is retrieved. It qualifies on whether a document was reachable, relevant to the specific question, and trustworthy enough to name. Those are two different pipelines with two different gatekeepers, and there is no mechanism by which spending money in one moves you through the other.
This should be familiar, because it is the same lesson search advertising taught for twenty years. Buying AdWords never moved an organic ranking, and everyone eventually stopped asking. The reason the question keeps returning in AI search is that the ad and the answer share one visual surface, so they feel like one system to a buyer looking at the screen. They are not.
“The ad and the answer occupy the same rectangle. They do not share a qualification path, and nothing you spend on one moves you through the other.”
There is a second-order effect worth naming, and it is the uncomfortable one. If your ad appears next to an answer that cites three competitors and not you, the ad is now working against you. The reader gets a paid message from a company the system apparently did not consider a credible source on the question, next to a substantive answer built from companies it did. We flagged the beginnings of this in ChatGPT ads and the citation strategy problem, and this dataset gives it a magnitude.
The relevance problem sitting underneath
The other headline number deserves attention on its own terms. In 14.35% of cases the ad had no topical relevance to the prompt it appeared against. In News and Politics that figure was 54.2%, and in Relationships 51.1%. In two categories, more than half the ads shown were unrelated to what the person asked.
Share of ads showing no topical relevance to the paired prompt, by category, from SE Ranking's August 10, 2026 study of 50,006 commercial prompts.
For advertisers this is a wasted-spend story. For anyone thinking about earned visibility it is something more useful: evidence that the monetization layer and the retrieval layer are not just separate but operating at very different levels of maturity. Retrieval has been trained on relevance for years. Ad matching is months old. Betting your visibility on the newer, less accurate system when the older one is the thing your buyer reads is a strange allocation.
What actually earns the source slot
If the slot cannot be bought, the question becomes what qualifies a document to be named. Across our citation tracking the answer has been consistent and slightly annoying: engines cite sources they can reach, parse, and corroborate elsewhere. Reachability and parsing are technical work and largely solved by a competent team. Corroboration is the hard part, and it is not a content problem.
That list is a link building and digital PR brief, not a content calendar. It is why we moved the AI visibility work for most clients into the link building and digital PR practice rather than treating it as a subset of content. The deliverable is presence on other people's domains, and that has always been a different craft from publishing on your own.
The categories where this matters most are the ones being defined right now, where an engine has few trusted sources to choose from and the first credible ones become the default answer. That is the argument we make to AI and machine learning companies whose category did not exist three years ago, and it is exactly the sequence we ran for Arnica while its segment was still being named.
Where the ChatGPT ads citations data changes the budget
None of this is an argument against advertising in AI answers. It is an argument against counting it as AI visibility. Those are different line items with different success metrics, and merging them produces a plan where the paid spend quietly absorbs the budget that was supposed to build durable presence.
| OBJECTIVE | CORRECT INSTRUMENT | WRONG INSTRUMENT | METRIC |
|---|---|---|---|
| Immediate reach on commercial prompts | Paid placement in the answer surface | Waiting for citations to accrue | Impressions and cost per qualified click |
| Being named as a source | Third-party corroboration and original data | Higher ad spend on the same prompts | Mention rate and citation rank per engine |
| Category definition in a new market | Reference, review, and community presence | Brand campaigns on your own domain | Share of category answers naming you |
| Defending an existing position | Comparison content plus maintained entity signals | Assuming rankings carry over | Per-engine coverage of your top prompts |
The practical split we recommend is to fund paid placement out of the demand capture budget where it belongs, and fund corroboration out of the brand or earned media budget, then report them against different numbers. When both come from the same pot the paid line always wins the quarterly review, because it reports faster, and the corroboration work never gets the eighteen months it needs to compound.
There is also a sequencing argument. Corroboration takes quarters. Paid placement takes an afternoon. A team that starts the slow work now and layers paid on top later ends up with both. A team that starts with paid and plans to add corroboration once budget frees up usually never starts, because budget does not free up. The order is the strategy.
The corroboration work order
Run the diagnostic first, because most teams have never looked. Take the fifteen prompts that matter most to your pipeline, run each one in the engines your buyers use, and record two things: who gets cited, and whether the sources cited are your own domain or someone else's. If the sources naming your competitors are third-party and the sources naming you are your own site, you have found the gap and it is a corroboration gap.
Then pick the three third-party surfaces that appear most often in those answers and get properly present on them. Not a thin profile. A complete, accurate, maintained presence with the details an engine can lift. We wrote the full sequence in the third-party citation playbook, and the reason it takes a playbook rather than a checklist is that each surface has its own qualification bar.
The 3.63% figure is the most honest thing published about AI visibility this month. It says the answer surface is not for sale, and everything that follows from that is good news for anyone willing to do slow work. The full study is at SE Ranking's August 10 report.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Josh leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.