Cited is not recommended: when AI uses your page but names a rival
The dashboard looks fine. Citations are up. Someone in the room says the brand is “visible in AI.” Then a buyer opens ChatGPT, asks for a recommendation in the category, and a rival gets named. Your URL sits quietly in the source list. Their name sits in the answer. That gap is not a glitch. It is a different outcome wearing the same score.
The short answer
Being cited and being recommended are two different outcomes. An AI answer can use your page as evidence and still name a competitor as the pick. Studies from September and October 2026 show the gap is common, especially when a brand ranks itself first. Report “named” and “cited” as separate columns for each engine. Then get a claim about you confirmed by someone else.
Cited is not the same as recommended
Most visibility reports collapse those two results into one number. That is how a brand celebrates a citation while a competitor takes the shortlist.
Self-ranked listicles often help a competitor more than the publisher. In September 2026, Lily Ray tracked 100 “best B2B software” queries in Google AI Overviews. When an Overview cited a listicle where a brand ranked itself first, that brand was left out of the recommendation 83% of the time, up from 69% in June, and those listicles were cited 38% less often. In 70% of cases, the listicle helped competitors get recommended instead (Lily Ray, 4 Oct 2026). One analyst, one hundred queries: still hard to ignore.
Rank helps discovery. It does not decide the pick. Discoverable’s UK shortlist study, across 1,000 commercial questions and six AI services in early September, found the same split. On Perplexity, the business ranked first on Google was not named in 74.9% of eligible answers; counting citations as presence drops that to 55.9%. ChatGPT did not name Google’s number-one business in 72.5% of eligible answers (Discoverable, 16 Sep 2026).
Engines pick their own evidence. Page-one rank is not a ticket into the answer. WashU’s audit of 55,393 trending Google searches put that in numbers: 29.8% of domains cited in AI Overviews were not on page one. An average Overview cited about eight sources; up to about 11% of claims lacked support in, or contradicted, the cited pages (WashU, 29 Sep 2026).
None of these studies focused on Indonesian prompts. In our own monitoring we also see the reverse gap: a brand can be named while its own pages are rarely the cited source, and on recommendation prompts, listing platforms out-cite the brand’s own site. A high visibility score is not the same as content being cited.
Four outcomes that look the same on a dashboard
Before anyone celebrates “AI visibility,” ask which of these four rows you are actually in.
| Outcome | What the buyer sees | What it means | First fix |
|---|---|---|---|
| Named and cited | Your brand in the answer, your URL in sources | The engine trusts you and your page | Protect it: keep facts stable, re-measure |
| Named, not cited | Your brand, but sources are listings or third parties | You are known, your page is not the evidence | Make your own page extractable (see Key Sources) |
| Cited, not named | Your URL in sources, a rival in the answer | Your page is evidence for the category, not for you | Rewrite the page around a claim about you; add corroboration |
| Neither | Nothing | No usable trace | Start with a prompt baseline |
Row three feels like winning until you read the answer aloud. The engine borrowed your facts and handed the recommendation to someone else.
Why an engine cites you and still picks someone else
Your page can be excellent evidence about a category and still be a poor case for you. A “top 10” page that names nine competitors hands the engine nine names to repeat. It can trust your list and still judge the author not neutral. The listicle becomes a gift to the shortlist, not a vote for the publisher.
Self-praise is not corroboration. In White Wood’s Narrative Architecture, your own page supplies Authority: a clear claim. Before naming a brand, engines also weigh whether the same claim is confirmed elsewhere. A sentence that only appears on your site is thinner than the same sentence echoed by an independent source.
Your page may also answer the process rather than the shortlist. Discoverable notes that some answers explain a process or ask a follow-up instead of naming suppliers. A how-to page can be cited for the steps while any recommendation goes to a brand independent sources confirm. Being useful for the method is not the same as being the pick.
And engines move even when your pages do not. In early September, AI Mode running Gemini 3.8 Flash showed barely any links. Google’s Robby Stein said “This isn’t working as intended,” and links were back the next morning (Search Engine Roundtable, 3 to 4 Sep 2026). Citation behaviour can shift overnight. Last week’s citation is a snapshot, not a strategy.
Google’s October guidance points the same way
Google’s helpful content guidance, revised in early October 2026, names four attributes raters evaluate in main content: effort, originality, talent or skill, and accuracy. It also calls fabricating creator profiles “a form of deception” (Google Search Central). Its generative AI guidance, updated 1 October, says it is “critical to manually factcheck and review all AI-generated content” before publishing (Google Search Central).
Paid placement is no shortcut. Announcing a new visual ad format for ChatGPT on 5 October, OpenAI repeated that “advertising does not influence the answers ChatGPT provides” (OpenAI). Ads buy a labelled slot, not a recommendation.
What to change
- Split the columns. For each tracked prompt and engine, log: named (yes/no), your URL cited (yes/no), who was recommended. See reading AI visibility scores by engine.
- Retire self-ranked lists. Rewrite them as honest comparisons with stated criteria, including where you are not the best fit.
- Write one sentence engines can repeat about you. Brand, category, proof, near the top. Not “leading,” not “trusted.”
- Get that same sentence confirmed once, somewhere independent. Coverage, an industry list, an interview. See third-party corroboration.
- Re-measure the same prompts per engine 7 to 14 days after changes go live, before claiming a win.
FAQ
Is being cited without being named still worth anything?
Yes. It shows the engine can retrieve and trust your page. That is a foundation, not a recommendation; report it as such.
Should we delete our “best X” listicle?
Not automatically. Rewrite it to help a buyer choose, with criteria and honest trade-offs. If it only exists to put you at number one, it is more likely to help competitors.
If we rank first on Google, will ChatGPT recommend us?
Not reliably. In Discoverable’s study, ChatGPT did not name Google’s number one business in 72.5% of eligible answers. Rank helps discovery; it does not decide the pick.
Can ChatGPT ads fix this?
No. OpenAI states that advertising does not influence ChatGPT’s answers. Ads buy a labelled slot, not a recommendation.
What to bring to your next review
- The prompts where your URL is cited but a rival is named.
- The pages behind those citations, and the sentence about you each should carry.
- One independent source per claim to earn before the next re-measure.