Is GEO enough to be cited by AIs?
Published on
Not on a question that does not name you. GEO improves what an AI retains from a page it has read: on a question that names your company, this matters. But on a market question, our measurements show that AIs mostly consult third-party sources, and not the website of the suppliers concerned: a page that no one is looking for does not get cited, no matter how well-written it is. Ranking does not bridge the gap either: according to Ahrefs in August 2025, only 8% of the pages cited by ChatGPT appear in Google's top 10 for the same query, and Moz, in the announcement of a study covering approximately 40,000 queries in February 2026, indicates that 88% of citations from Google's AI mode do not correspond to the organic top 10.
For what type of question? This page is about market questions, the ones asked by customers who do not know you yet. For customer service or documentation questions, where the customer names you, your site remains the primary source and GEO produces its effect there.
What GEO actually improves
A page that answers from its very first lines, using the words of the question in expanded text, is better reused than a page that beats around the bush. This is what the foundational GEO paper finds, in a simulated engine where sources are provided to the model: adding citations and statistics increased visibility, while keyword stuffing yielded almost nothing.
This result must be read for what it is: the page was already given to the engine. The study measures the share of a page within a response, not its chances of being found. Yet it is this 'being found' step that is most often missing.
What it cannot do: be read
In our measurements, on a question that does not name anyone, AIs run generic searches and retrieve third-party sources. A supplier's site only appears there exceptionally, and when it does, it is rarely cited. No setting on that site changes this, because the choice is made before reading, at the moment when the model decides where to search.
This is the purpose of distribution across third-party sources: it acts on the step that GEO does not touch. It does not replace GEO, however, since the third-party page must also be well-constructed to be reused.
What the published figures show
Two sets of external figures point in the same direction. Ahrefs, in August 2025, measures a very low overlap between pages cited by ChatGPT and Google's top 10: 8%. Moz, across approximately 40,000 queries in February 2026, announces that 88% of citations from Google's AI mode do not correspond to the organic top 10 of the same query. We were unable to read Moz's study page, as their site rejects our reader; the figure is the one Moz itself published.
These figures do not mean that SEO or GEO are useless. They say that the position in an AI response is played out on a different playing field than the position in a list of results, and that work limited to your domain does not decide that playing field.
What we do not know
Our measurements cover the API of GPT-5, Claude Sonnet 4.6, and Gemini 3.5 Flash, not the consumer interfaces, which may behave differently. They describe what the models search for and cite at a given moment; they do not measure the effect of a publication on citations, and an AI changes its response from one time to another.
Sources
- Ahrefs, Louise Linehan — overlap between AI search and top 10, .
- Moz — announcement of its study on Google AI Mode citations, approximately 40,000 queries from February 2026; the study page is not accessible to our reader, .
- Aggarwal et al. — "GEO: Generative Engine Optimization", KDD 2024, .
- Google Search Central — "Optimizing your website for generative AI features on Google Search", .
- FaireDuBruit measurements, from September 4 to 16, 2026 — see "How we measured".