The short version
- In Google’s AI Overviews, the first organic result was cited in 49.5% of answers and the ninth in 15.5%, in our AI Overview citation study.
- In a simulated test, moving a page to the top of what the AI reads beat every content rewrite tested, Puerto and colleagues found (opens in a new tab).
- Only 8.3% of the pages ChatGPT cited ranked in Google’s top 10 for the question, against 25.7% for Claude, in our rankings study.
- ChatGPT ran a mean of 3.7 of its own searches per answer, so ranking for the buyer’s exact question is not enough, our hidden-searches study found.
Does ranking on Google still matter for AI answers?
Yes, especially for Google’s own AI features. AI engines first search, then choose what to cite from what they found. A page that is not retrieved cannot be cited.
In our AI Overview citation study of 4,051 citations, 28.7% were page-one organic results for the same search. And 78.8% of AI Overviews cited at least one page-one result. Our guide on whether Google rankings get you into AI Overviews breaks this down.
Zhang and colleagues (opens in a new tab) saw the same pull in 2025 across six AI engines. Among domains that both AI and traditional engines returned, the most common was the domain ranked first in traditional search: 23.27% in Bing and 14.53% in Google. Those shared domains were a minority of everything the AI engines cited.
How much does a page’s position matter?
A great deal: higher-ranking pages are cited far more often. In our AI Overview study, the first organic result was cited in 49.5% of AI Overviews, the third in 33.5%, the sixth in 23.2% and the ninth in 15.5%.
A second study of 3,096 top-10 pages confirmed it. In our cited-pages study, 41.7% of pages ranking 1 to 3 were cited, against 20.1% at positions 7 to 10. Position explained more than the 14 page features we tested, combined.
Controlled tests agree. Puerto and colleagues (opens in a new tab) built a simulated AI search engine and tried published content rewrites meant to win citations. Most did little: 61.0% of retail product rankings were unchanged after one rewrite method. Placing a page first among the documents the AI read produced far larger gains than any rewrite.
That test simulated ranking by reordering documents, not by real SEO work. It still shows how much order matters once a page is retrieved.
A small 2024 study by Narayanan Venkit and colleagues (opens in a new tab) saw the same pattern from the user’s side. Across sessions with 21 participants, people using Google often explored beyond the top five results. The answer engines they tested drew mainly on the top two or three.
Why doesn’t a top ranking guarantee a citation?
Because AI engines run their own searches and choose differently from Google’s ranking. In our hidden-searches study, ChatGPT ran a mean of 3.7 searches per answer. None of the 509 searches we saw repeated the user’s question word for word.
Many cited pages are outside Google’s rankings entirely. In our rankings study, 68.5% of ChatGPT’s citations were not in Google’s top 100 for the question or for any of the searches we tracked. For Claude the figure was 36.8%.
Google’s AI Overviews also look past page one. Xu and colleagues (opens in a new tab) tracked 7,583 AI Overviews in spring 2026. Only 31.5% of cited domains overlapped with the top five results, rising to 70.2% across the full first page. The other 29.8% did not appear on the first page at all. We compare the overlap across studies in whether AI engines cite the same sites as Google.
Does SEO matter equally for every AI engine?
No. Google rankings carry over best to Google’s own AI and least to ChatGPT. In our rankings study, 8.3% of pages ChatGPT cited ranked in Google’s top 10 for the question, against 25.7% for Claude.
Even Google’s two AI products differ. In our AI Mode comparison, AI Overviews cited 29.7% of top-10 pages and AI Mode 16.3%, on the same 400 searches.
It varies by industry too. Chen and colleagues (opens in a new tab) compared AI search with Google for local business questions in 2025. The share of sites both returned ranged from 20.6% for home cleaning to 2.5% for auto repair and 0.1% for IT support.
Engines without live search are harder still. Sharma (opens in a new tab) tested 112 Product Hunt startups. A version of ChatGPT without web search surfaced them in only 3.32% of discovery questions. Perplexity, which searches the web, managed 8.29%, and its results tracked SEO signals such as the number of sites linking to a startup.
Can content changes make up for a weak ranking?
Not on their own, according to the evidence so far. Zhu and Chang (opens in a new tab) audited 14 Tokyo hotel websites against Gemini’s citations. Cited hotels scored 8.6 out of 15 on content depth, against 3.4 for hotels that were not cited.
The authors describe a two-stage process. A hotel site first needs content deep enough to rank in Google. Only then does it compete on how well it answers the question. The audit is small and exploratory, and the authors work at an AI company.
Another controlled test points the same way. Vishwakarma and colleagues (opens in a new tab) ran 252,000 trials across six AI models. Topical relevance and list position were the biggest drivers of being cited first; formatting-only edits had little impact.
What should you do about it?
Keep investing in SEO, then add the work that rankings alone do not cover.
- Protect your rankings for the questions buyers ask. Higher positions are cited far more often.
- Rank for the related searches AI engines run, not just the headline question, such as reviews, prices and comparisons.
- Build links and mentions from other sites. They help ranking, and one study tied them to Perplexity visibility.
- Check Google’s AI features and ChatGPT separately, since rankings carry over very differently.
- Make each page answer its question clearly once it is retrieved.
For a combined approach, see our generative engine optimization service.
What does the research not tell us yet?
No study we reviewed tested whether improving a real page’s ranking changes its AI citations. The reverse question, whether ChatGPT citations lift Google rankings, is also unproven.
- The strongest ranking evidence comes from simulations that reorder documents, not live SEO campaigns.
- Most live studies are observational snapshots on one or a few days.
- We cannot see the indexes ChatGPT and Perplexity use, so Google rankings are only a stand-in.
- Small samples, such as 14 hotels or 112 startups, limit how far some findings travel.
- Much remains unexplained. In our cited-pages study, 91.4% of the variation in which top-10 pages were cited was left unexplained by anything we measured.
Frequently asked questions
Do I still need SEO if I am optimizing for AI search?
Yes. Pages ranking 1 to 3 were cited in 41.7% of cases in our study of 3,096 top-10 pages, about twice the rate of positions 7 to 10.
Does ranking first on Google get me into AI Overviews?
Often, not always. The first organic result was cited in 49.5% of AI Overviews in our study.
Does ChatGPT use Google rankings?
Only loosely. Just 8.3% of ChatGPT’s cited pages ranked in Google’s top 10 for the question in our study of 80 buyer questions.
Is GEO replacing SEO?
No. In a simulated test, a page’s position among the documents the AI read mattered more than any content rewrite tested.
Sources
- Puerto et al. (2025), C-SEO Bench: Does Conversational SEO Work? (opens in a new tab), arXiv:2506.11097.
- Zhang et al. (2025), Source Coverage and Citation Bias in LLM-based vs. Traditional Search Engines (opens in a new tab), arXiv:2512.09483.
- Xu et al. (2026), Measuring Google AI Overviews: Activation, Source Quality, Claim Fidelity, and Publisher Impact (opens in a new tab), arXiv:2605.14021.
- Zhu and Chang (2026), The End of Rented Discovery: How AI Search Redistributes Power Between Hotels and Intermediaries (opens in a new tab), arXiv:2603.20062.
- Sharma (2026), The Discovery Gap: How Product Hunt Startups Vanish in LLM Organic Discovery Queries (opens in a new tab), arXiv:2601.00912.
- Narayanan Venkit et al. (2024), Search Engines in an AI Era: The False Promise of Factual and Verifiable Source-Cited Responses (opens in a new tab), arXiv:2410.22349.
- Chen et al. (2025), Generative Engine Optimization: How to Dominate AI Search (opens in a new tab), arXiv:2509.08919.
- Vishwakarma et al. (2026), What Gets Cited: Competitive GEO in AI Answer Engines (opens in a new tab), arXiv:2605.25517.
- Underneath (2026), Do AI Overviews cite the pages that rank? 4,051 citations
- Underneath (2026), What pages cited by AI Overviews have in common: 3,096 pages
- Underneath (2026), Do ChatGPT, Gemini, Perplexity and Claude cite pages that rank?
- Underneath (2026), The hidden searches AI assistants run before they answer
- Underneath (2026), AI Mode vs AI Overviews: how different are the sources?