How to Rank in AI Overviews: What 54 Studies Show About Getting Cited

Most content written about AI Overviews loops through the same advice: add structured data, answer questions clearly, build authority. It sounds right, but none of it says what it rests on. The meta-analysis Cyrus Shepard published in May 2026 tries to close that gap: it reviews 54 experiments, patents and case studies and scores 23 factors by strength of evidence. It is the first time advice in the GEO field has been ranked by weight of proof.

How Were the Factors Scored?

Scoring rests on three criteria: whether the finding repeats across multiple studies, the strength of the evidence (study size and quality), and whether official documentation or patents support it. The scope is not just Google — ChatGPT, Gemini and Perplexity are included too.

That method differs from ordinary "do this" checklists in one important way: a low score does not mean a factor is wrong, it means it is not yet sufficiently proven. When you are deciding where to spend limited resource, that distinction matters.

The Best-Evidenced Factors

The top ten factors and their scores:

Taken together the list is reassuring: four of the top five factors are already the business of classic SEO. The meta-analysis says as much itself — winning SEO usually means winning AI citations, with a few extra steps.

Three Places Where It Departs From Classic SEO

Doing everything the old way is still not quite enough. Three behaviours need to change:

• Put the answer at the top of the page. The amount of text a model can pull from your content is limited. An answer that arrives after a long preamble can fall outside the retrieval window. Writing the answer into the first paragraph and laying the detail out underneath is a stylistic preference in classic SEO — here it is a technical requirement.
• Account for fan-out queries. AI engines do not settle for one query; they split the user's question into sub-questions and run additional searches. Ranking first for the head keyword may not be enough — you also need to show up for the questions around the topic. That forces a move from single-page optimisation to topic-cluster thinking.
• Replace vague phrasing with concrete data. "Significantly increases" does not get cited; "a 15% drop across a 700,000-keyword analysis" does. Numbers, dates, sources — every verifiable element raises the odds of being cited.

What About the Traffic Impact? (Read Carefully)

Honesty is required here: the figures circulating on this subject do not agree with each other, and trusting any single number would be a mistake.

In Seer Interactive's study, organic click-through rate on queries showing an AI Overview fell from 1.76% to 0.61% — a drop of roughly 61%. But the size of the measured effect changes dramatically with method: Amsive's analysis of 700,000 keywords put the drop at around 15%, while DMG Media's study of a narrow set of navigational queries reported up to 89%.

Zero-click rates are climbing too: from about 50% in 2019 to roughly 65% in 2026, and considerably higher inside Google's AI Mode. There is another side to the coin, though — studies also show brands that are cited in AI Overviews earning more organic clicks, although those uplift figures vary just as widely between sources (from 35% to 120%).

The lesson: rather than budgeting against industry averages, read your own data. Google now provides a report for exactly that, and we covered how to read it in our article on the GSC Generative AI report.

Backlinks or Brand Mentions?

One striking finding sits outside the meta-analysis: a brand's mentions across the web correlate with AI visibility roughly three times more strongly than backlinks do. In other words, even an unlinked brand mention carries value on the AI side.

Another finding concerns freshness: across an analysis of some 17 million citations, cited content was on average 25.7% more recent than the organic top 10. That turns regular refreshing of older but well-ranking content into a concrete line of work for AI visibility.

One caveat is essential: these are correlations, not causation. It is perfectly possible that heavily mentioned brands are simply large brands, and that is why they get cited. Even so, both findings offer a reasonable case for shifting investment from pure link building towards brand visibility and content maintenance.

An Actionable Roadmap for AI Search Optimization

The data above translated into practice, in order of importance:

• Secure the technical floor first. Verify that the pages you want cited are crawlable and not restricted by directives like nosnippet. This is the highest-scoring item on the list and the most frequently skipped.
• Focus on queries where you already rank in the top 10. Since 38% of citations come from the top 10, the fastest gains lie in making already well-ranking pages AI-ready — not in producing content from scratch.
• Restructure those pages. Move the answer to the top, write headings that map directly onto the question, and make each section understandable on its own.
• Complete the topic cluster. Cover the questions surrounding the main query with separate pages or clear subheadings; showing up in fan-out searches depends on it.
• Set up measurement from the start. Establish your baseline in Search Console's Generative AI report and date your changes so you can compare before and after.

Conclusion

AI Overviews are not a break that ends SEO; they are a continuation that tightens the criteria. The evidence-based picture says you do not need to learn an entirely new discipline — crawlability, ranking and topical authority are still decisive. What has changed is how easily a machine can extract your content and how much verifiable information it carries.

For the conceptual framing of AIO and how it differs from AEO and GEO, see our article on what AIO is and how to do it; to measure your own AI visibility with Google's data, see our article on the GSC Generative AI report.