[The Engines]
Why Traditional SEO May Not Guarantee AI Citations Anymore
A new Ahrefs analysis shows Google's AI Overviews citing far fewer top 10 pages than they did in 2025. Here is what changed, who it affects, and how to adjust.
Explore this article with AI
Open a source-aware analysis with this article as the primary source.The short answer
Ranking in Google's top 10 no longer predicts whether an AI engine cites you. A large-scale Ahrefs analysis found that only 38% of pages cited in Google's AI Overviews in March 2026 also ranked in the top 10 for the same query, down from 76% in July 2025. Rankings still matter, but they stopped being the gate they used to be.
What changed between July 2025 and March 2026?
Eight months ago, a top 10 ranking was close to a guarantee of an AI citation. Ahrefs analyzed 1.9 million citations pulled from 1 million AI Overviews in July 2025 and found that 76.1% of cited pages also held a top 10 spot for the exact query being answered. Getting cited and getting ranked were nearly the same job. A content team that won the SERP could reasonably expect the citation to follow.
That link has come apart, and it happened fast. A follow-up study covering 863,000 keywords and roughly 4 million AI Overview URLs, reported by Search Engine Journal on March 2, 2026, found that only 38% of cited pages still rank in the top 10 for the query they are cited on. Another 31.2% rank somewhere between position 11 and 100. A full 31% do not appear anywhere in the top 100 results. Three out of ten citations now come from pages a rank tracker would never surface at all.
Ahrefs has an honest caveat: its citation detection improved between the two studies, so some of the swing reflects better measurement rather than a pure behavior change. Even with that caveat, the company treats the gap as too large to explain away, and it lines up with a separate, mechanical change in how Google builds an answer.
What is query fan-out, and why does it break the old rules?
Query fan-out is the likely mechanism behind the drop. Instead of summarizing whatever sits in position one, Google now splits a question into a cluster of related sub-queries and cites whichever pages answer each piece best. A search for 'best project management software for remote teams' can spawn internal lookups for pricing, integrations, security certification, and mobile support, each pulling from a different page and often a different domain.
Google rolled out Gemini 3 as the default model behind AI Overviews worldwide in late January 2026, and the timing matches the shift Ahrefs measured in March. Ranking well for the visible head term stopped being enough once the model started shopping for the best answer to a dozen hidden questions instead of one visible one. A page ranking eighth for the head term but first for one buried sub-query can now out-cite a page sitting in position one.
| Metric | July 2025 | March 2026 | What to do about it |
|---|---|---|---|
| Cited pages that also rank in the top 10 | 76.1% | 38% | Stop treating a top 10 ranking as proof of a citation; track citation share separately |
| Cited pages ranking outside the top 100 | Not reported | 31% | Publish deep, fully sourced pages that answer a sub-query directly, even without a page one ranking |
| Citations with no brand mention (ghost citations) | Not measured | 61.7% | Name the brand and the specific claim in the same sentence a model is likely to quote |
Who does this affect?
The shift hits hardest wherever a team still treats a rank tracker as a citation tracker. B2B SaaS and tech growth teams that assume a page one ranking for 'best [category] tool' will carry over into ChatGPT's or Gemini's answer are flying on an instrument that stopped matching the terrain. Local, multi-location, and service businesses face the same gap: a top map-pack position no longer guarantees a mention when someone asks an AI assistant for the best plumber, dentist, roofer, or accountant nearby.
It also affects how success gets measured internally. A marketing team reporting keyword position to leadership can show a green dashboard while citation share, the percentage of relevant AI answers that actually name the business, quietly falls. One page can rank on page one and never appear in an answer. Another, buried on page four or missing from the top 100 altogether, can get cited because it answers one fan-out sub-query better than anything else indexed. Rank position alone can no longer tell a team which situation it is in.
Why are ghost citations making the problem worse?
Even a citation does not guarantee anyone learns your name. Semrush, working with Kevin Indig and Growth Memo, analyzed 3,981 domain appearances across 115 prompts run in 14 countries and published the results on June 9, 2026. It found that 61.7% of citations placed a link to the source page without ever naming the brand in the answer text itself, a pattern Indig calls a ghost citation.
The rate varies sharply by engine. ChatGPT cites a source link 87% of the time but names the brand in its written answer only 20.7% of the time. Gemini runs close to the opposite pattern: it names a brand in 83.7% of answers but formally cites the source link only 21.4% of the time. Getting pulled into an answer and getting recognized for it are two separate outcomes, and most teams still only track the first one.
The same study found that short, conversational prompts produced far more brand mentions than long, structured ones, and that comparative and how-to queries named a brand roughly twice as often as narrow informational ones. A business optimizing only for the head keyword, and ignoring the comparative and how-to questions around it, is optimizing for the exact query type least likely to say its name out loud.
How should you respond?
Traditional SEO still matters. Rankings remain one input into citations, and 38% is not zero. But treating a page one position as the finish line stops working once six in ten of your citations can now originate from outside the top 100, and six in ten of those citations may not mention your brand at all even when they land.
The fix is not a trick played on the model. It is coverage, depth, and freshness at a scale most in house teams cannot sustain alone. Answer the sub-questions buried inside your category, not only the visible head term. State the product name and the specific claim in the same sentence a model is likely to lift, since a citation without a brand mention does little for anyone. Publish comparison and how-to content deliberately, given how much more often those formats earn a named mention over a narrow informational page.
Track citation share, the percentage of relevant AI answers that name your business, across ChatGPT, Perplexity, Gemini, and Google's AI Overviews, instead of watching a single rank tracker dashboard and assuming it tells the whole story. Add answer presence, how broadly you show up across the full question universe in your category, so a strong showing on one popular prompt does not hide a blank spot everywhere else. None of this involves gaming a model. It means publishing more of the specific, sourced, current content the fan-out process is already hunting for, before a competitor fills the gap first.
Key takeaways
- Only 38% of pages cited in Google's AI Overviews in March 2026 also ranked in the top 10, down from 76.1% in July 2025 (Ahrefs).
- Query fan-out, Google splitting a question into hidden sub-queries, is the likely driver, and it followed the January 2026 rollout of Gemini 3 as the default AI Overviews model.
- 31% of citations now come from pages outside the top 100 results entirely.
- 61.7% of AI citations name no brand at all in the answer text, a pattern Semrush and Kevin Indig call a ghost citation.
- ChatGPT cites sources far more often than it names brands (87% versus 20.7%); Gemini does the reverse (83.7% versus 21.4%).
- Track citation share and answer presence across every engine instead of relying on a single rank tracker dashboard.
Omnicite Editorial. "AI Citations Need More Than SEO Rankings Now" The Citation Report, Omnicite. https://omnicite.co/blog/why-traditional-seo-may-not-guarantee-ai-citatio/
Sources
Source: Search Engine Journal
Only 38% of pages cited in Google's AI Overviews in March 2026 also ranked in the top 10 for the same query, down from 76% in July 2025 Search Engine Journal, 2026-03-02
Source: Ahrefs
76.1% of AI Overview-cited pages ranked in the top 10, based on 1.9 million citations across 1 million AI Overviews Ahrefs, 2025-07-21
Source: Semrush
61.7% of AI citations occur without the brand being named in the answer text, a ghost citation, based on 3,981 domain appearances across 115 prompts in 14 countries Semrush, 2026-06-09
Frequently asked questions
Does ranking on page one still help with AI citations?
Yes, but it is no longer close to sufficient. Ahrefs found that only 38% of pages cited in Google's AI Overviews in March 2026 also ranked in the top 10, down from 76.1% in July 2025. A top ranking still helps, it just stopped being close to a guarantee.
What is query fan-out?
Query fan-out is Google's practice of splitting one visible search into a cluster of related sub-queries, then citing whichever pages answer each piece best. A single answer can pull citations from several different pages and domains instead of just the top-ranked result.
What is a ghost citation?
A ghost citation is when an AI engine links to a page as a source but never names the brand behind it in the answer text. Semrush and Kevin Indig found this happens in 61.7% of citations across the engines and prompts they studied.
Should I stop investing in traditional SEO?
No. Rankings remain one input into citations, and 38% is not zero. The change is that rankings alone no longer explain most citations, so a team needs to track citation share and answer presence alongside rank position, not instead of it.
Which AI engines does this affect?
The studies behind this shift cover Google's AI Overviews, ChatGPT, and Gemini. Citation and brand mention behavior differs sharply by engine, so a strategy built only around one of them misses the others.
How can a business track whether AI engines are citing it?
It means monitoring citation share (the percentage of relevant answers that cite you), answer presence (how broadly you show up across the question universe), and share of voice against named competitors, checked across ChatGPT, Perplexity, Gemini, and AI Overviews rather than a single rank tracker.