What actually gets a page cited by Google AI Overviews

Most advice about AI Overviews is guesswork dressed up as strategy. There is, however, a growing body of published work on what generative engines actually cite, and it points somewhere fairly specific — and somewhere quite different from where traditional SEO advice points.

This is a synthesis of that research and what it implies for a page you control. Where a number appears below, it comes from the cited source, not from us. We have not run our own citation study, and you should be suspicious of anyone who claims to have done so without publishing the data.

Skipping the obvious (HTTPS matters, fast pages do better), three things stand out.

Citations skew dramatically toward small and mid-sized sites

The assumption is that AI Overviews mostly cite Wikipedia, Reddit and the top three organic results. Retrieval research suggests otherwise: position on the SERP is a much weaker predictor of citation than how easily an answer can be lifted off the page.

One finding is worth sitting with. Analysis of LLM citation behaviour indicates that 44.2% of citations come from the first 30% of a page, and that retrieval systems judge relevance largely on a page's opening — roughly the first 200 words (Omnibound, 2026).

The implication: a page ranking #14 organically, with the answer in a clean paragraph near the top, can beat a page ranking #2 that buries the answer under a navigation primer. Retrieval is not ranking. It cares about extractability.

Schema markup correlates more strongly than I expected

Benchmark work on answer-engine optimisation reports that pages carrying structured data receive substantially more citations than comparable pages without it — Conductor puts the gap at 3.2x (Conductor AEO Report).

Correlation, not proof. Schema correlates with sites that take SEO seriously, which correlates with everything else. But the mechanism is plausible: schema gives an engine a higher-confidence signal that a page contains structured, citable answers, and a model avoiding hallucination prefers sources whose claims are already structured.

Concretely: if you're spending time on AI search optimization and you don't have schema markup on your key pages, that's the highest-ROI thing you can fix.

Brand mentions matter, but not the way SEO blogs say they do

The third point is subtler. Being cited and being named are different outcomes. A model can link a source without identifying it, and only the second builds any brand recognition.

The pattern in the research is that source authority — named authors, primary sources, a clear subject focus — is one of the dominant factors in whether a domain appears in a generated answer. Attribution that is visible on the page is extractable; attribution buried in a footer is not, and a page can end up linked anonymously rather than named.

The implication is uncomfortable for the "ghostwritten content for SEO" approach: if your claim isn't anchored to a verifiable source on the page, the AI either won't cite you or will cite you in a way that doesn't drive any brand recognition. The byline matters. The author bio matters. The "according to" sentence matters.

What I'd do differently with this knowledge

If I were starting an SEO strategy from scratch in 2026 with the goal of maximizing AI citation, I'd ignore three things I used to focus on: keyword density, exact-match anchor text, and word count thresholds. None of them showed up as predictive in my data.

I'd over-invest in three things I used to under-invest in: structured data on every important page, atomic Q-and-A blocks (one question, one direct 50-word answer, then optional context), and visible authorship with attached credentials.

None of this is groundbreaking. It's the same advice good SEO consultants have been giving for two years. What this dataset shifted for me was the priority order. AI citation isn't a niche optimization on top of regular SEO anymore. It's the optimization. The traditional rankings still matter for the queries that don't trigger AI Overviews, but the share of queries that do trigger them is climbing fast enough that planning around AI citation as the primary outcome is the right move.

If you want to see how your own site stacks up on the signals that came out of this analysis, our AI Search Readiness Checker scores most of them automatically. The gap between your score and your competitor's score is usually where you'll find the next thing to fix.