Does SEO still work with AI?
Published on
Yes, SEO still works with AI, and it remains the foundation, but it is no longer enough. For its own AI answers, Google writes in July 2026 that SEO best practices remain relevant because these features rely on its ranking and quality systems. With ChatGPT, on the other hand, Google ranking matters little: what matters is the rank in the results of its own searches, which it reformulates with its own words, often in English and often with the name of a source it already knows.
For what type of question? It all depends on the AI and the question. For Google's AI answers, SEO remains the main lever. For ChatGPT and Claude, which write their own searches, it remains a useful condition, but the reputation of the sources carries as much weight as the page on a market question.
What Google says: optimizing for AI is still SEO
Google has settled the question for its own search. In its guide on optimizing for its generative AI features, updated on July 10, 2026, it writes that, from its perspective, optimizing for generative AI search is equivalent to optimizing for the search experience, and therefore that it is still SEO. SEO best practices remain relevant because these features rely on Google Search's ranking and quality systems.
The entry requirement is the same as for classic search: a page must be indexed and able to display with a snippet. However, the link between ranking and citation is weakening in AI Overviews: Ahrefs measured about 76% of citations coming from the top 10, then 37.9% in March 2026, using a different method that calls for caution. Google breaks the question down into sub-searches, and the cited pages increasingly come from those searches.
What research shows: "GEO" recipes do not beat SEO
A benchmark presented at NeurIPS in 2025, C-SEO Bench, evaluated methods that modify a page to get it cited by a conversational engine, based on answering questions and product recommendations. Its authors conclude that most of these methods are largely ineffective, and that they often have a negative effect on the document's ranking, contrary to expectations.
The same study shows that strategies aimed at improving the source's rank among those provided to the model are significantly more effective, and that the gains from optimization recipes decrease as more actors adopt them. In other words, being well-placed remains the main lever, and a recipe that everyone applies ceases to distinguish anyone.
What we measure: rank matters, but not Google's
With ChatGPT, a page's rank in its search results dominates its choice: a page in the top three positions is cited 6.9 to 8.5 times more often than a page beyond the ninth, across all our corpora. But these results are from searches that ChatGPT writes itself—a median of ten per answer for our questions on September 15, 2026—and almost always at least one in English. According to Ahrefs, only 8% of the pages cited by ChatGPT appear in Google's top 10 for the same query.
With Claude, rank matters little across most of our corpora. A page brought back by two different searches for the same answer is cited there in 88% to 100% of cases, and a page that overlaps with the content of the other pages found is cited 1.2 to 1.5 times more. The SEO that serves these AIs is therefore not the ranking on a keyword, but a page's ability to appear in the numerous searches through which an AI explores a topic.
What the AIs did with this question
On the question "Does SEO still work with AI?", ChatGPT went looking for primary sources: research papers published on arXiv, SparkToro, and Google's documentation and blog. Claude and Gemini mostly cited French-language agency and digital marketing media websites.
Claude added the year 2025 to all its searches on this question, even though we asked it in September 2026. The words an AI adds to the user's question determine the pages it finds: a page that only carries the terms of the question may miss a search that the AI has specified in its own way.
What this changes for a company
Do not sacrifice SEO to recipes presented as specific to AI: indexing, page speed, comprehensive content that answers your customers' questions serve Google, its AI answers and, in part, ChatGPT and Claude. On ChatGPT, the slowest pages are cited about half as often as the fastest.
Then add what traditional SEO does not cover: searches that AI reformulates, sometimes in English, and the reputation of the sources they look up by name. Google warns that there is no point in creating a page for each search variant: a comprehensive page, which truly answers, is better than a series of almost identical pages.
What we do not know
We do not measure the Google ranking of pages, nor the traffic that SEO brings. Our rank measurements focus on the search results that models receive through their API, and C-SEO Bench evaluates conversational engines under controlled conditions.
Sources
- Google Search Central — "Optimizing your website for generative AI features on Google Search", .
- Puerto et al. — "C-SEO Bench: Does Conversational SEO Work?", NeurIPS 2025, .
- Ahrefs, Louise Linehan — cross-referencing between AI search and top 10, .
- Ahrefs, Louise Linehan — AI Overviews citations and top 10, .
- FaireDuBruit measurements, from September 4 to 16, 2026 — see "How we measured".