What the current evidence says
In July 2025, Ahrefs reported that 76.1% of the pages cited in its Google AI Overview sample also ranked in Google's top 10.[1] A later Ahrefs analysis using a larger sample found that about 37.9% of cited pages ranked in the top 10.[2] These studies used different snapshots and samples, so the change should not be read as a controlled before-and-after experiment. It does show why a single industry statistic should not become a permanent rule.
A useful preprint, with an important caveat
A 2026 AI+Automation Research preprint compared AI-cited pages with Google ranking positions. It reported that a top-three page was 7.82 times as likely to be cited as a page ranking 11 to 30, and roughly 34 times as likely as a page ranking 31 to 100, on a per-page basis.[3] The same analysis reported an odds ratio of 1.31 for the presence of schema markup.
Those are observational associations. The paper is explicitly labelled a preprint and has not yet been peer-reviewed. It does not prove that adding schema causes a 31% lift, that schema is the most important intervention for every site, or that the work takes one day. Treat it as a useful signal to test, not a guaranteed playbook.
Google's official position is simpler
Google says the same SEO fundamentals used for ordinary Search also apply to AI Overviews and AI Mode. A page must be indexed and eligible to appear with a snippet, but there are no additional technical requirements, special schema types, or AI-specific text files required for these features.[4]
- Make important content accessible to Googlebot and available as readable page text.
- Use accurate titles, internal links, and canonical signals.
- Use structured data only when it matches visible content and the page type.
- Keep factual claims current, sourced, and easy for a reader to verify.
Where GEO adds work beyond SEO
GEO is most useful as a research and measurement discipline. It asks whether AI-generated answers represent the brand accurately, which sources shape those answers, and which commercially important questions omit the brand entirely. That creates a work program without pretending to control the model.
1. Establish a reproducible baseline
Define a fixed prompt set from real buyer questions. Record the engine, product mode, date, language, location, account state, answer, citations, and screenshots. Repeated runs help distinguish a durable pattern from a one-off response.
2. Fix ordinary discoverability first
Resolve crawl, indexing, canonical, rendering, and content-quality problems before adding AI-specific experiments. For ChatGPT Search, distinguish OAI-SearchBot from GPTBot: OpenAI documents OAI-SearchBot for search inclusion and GPTBot for potential model training.
3. Improve the evidence around the brand
Make official facts consistent across the site and legitimate profiles. Publish useful first-party material, correct inaccurate third-party information where possible, and earn independent coverage on merit. Wikidata or Wikipedia should only be pursued when their notability and conflict-of-interest rules are met.
4. Measure outcomes independently
Track brand mentions, owned-domain citations, third-party citation sources, and answer accuracy as separate outcomes. A page being indexed does not mean an AI engine retrieved it, and a brand mention does not mean the brand's website was cited.
A conservative priority order
- First: verify crawlability, indexation, rendering, and factual consistency.
- Second: build a logged baseline for a small, commercially meaningful prompt set.
- Third: improve pages that answer those questions with original evidence and clear sourcing.
- Fourth: strengthen legitimate third-party corroboration and correct inconsistent brand facts.
- Finally: rerun the same panel and report observed changes without assigning causation unless the design supports it.
The bottom line
SEO remains the base layer. GEO is a measured effort to understand and improve how a brand appears in AI-generated answers. The reliable promise is better visibility into the problem, stronger source material, and disciplined testing. A guaranteed citation or recommendation is not a reliable promise.
Sources
- [1] Ahrefs. How Often Do AI Overviews Cite Top-Ranking Pages? ahrefs.com/blog/search-rankings-ai-citations
- [2] Ahrefs. Do AI Overviews Still Cite the Top 10 Results? ahrefs.com/blog/ai-overview-citations-top-10
- [3] AI+Automation Research. The SEO Floor: Measuring Google Rank Distribution of AI-Cited Pages. Preprint, not yet peer-reviewed. aiplusautomation.com/research/the-seo-floor
- [4] Google Search Central. AI features and your website. developers.google.com/search/docs/appearance/ai-features