How to Optimize Your Content for AI Overview Citations

What changed in AI Overview citations?

Google's AI Overviews stopped mirroring the organic top 10. A March 2026 analysis by Ahrefs of 863,000 keyword SERPs and 4 million AI Overview URLs found that only 37.9% of cited pages also held a top 10 organic ranking for the same query. In July 2025, that share was about 76%. Pages ranking positions 11 to 100 accounted for 31.2% of citations, and pages ranking outside the top 100 (or not ranking at all) accounted for another 31.0%.

That somewhere else is what Ahrefs calls fan-out: the sub-queries Google's AI runs behind the scenes before it writes a summary, each pulling its own set of candidate pages. Google made Gemini 3 the default model behind AI Overviews on January 27, 2026, and the citation mix shifted hard in the weeks that followed. A page sitting at position 40 for your target keyword can still get quoted, if it directly answers one of the sub-questions the model generates on the way to that summary.

One thing has not moved: Google still cites multiple sources per answer, and content quality still matters. What changed is which pages qualify as a source in the first place. Rank went from a near-guarantee to one signal among several.

The size of the underlying dataset is part of why the finding holds up. Ahrefs analyzed roughly twice the citations it used in its original July 2025 study, the same study that produced the 76% baseline. A shift this large, on a dataset that big, is not noise. It is Google changing how it decides what counts as a good enough source to quote.

Who does this shift affect?

Anyone whose AI visibility plan started and ended at rank tracking. B2B SaaS teams competing for 'best [category] tool' prompts lose the old assumption that the page 1 winner is also the AI Overview winner: a competitor sitting on page 3 can now out-cite them entirely. Local and service businesses chasing '[service] in [city]' prompts face the same gap. Showing up in the map pack no longer guarantees a citation.

It changes how teams should measure success, too. Citation Share, the percentage of relevant AI answers in a category that cite you, is a more honest scoreboard than rank for this exact problem. A brand can hold strong rankings and still watch a competitor pull ahead on citations, because the citation pool now draws from a far wider set of pages than the ten links a person actually sees on the results page.

The platforms doing the citing have shifted too. A separate Ahrefs tracking run from September 2026, covering more than 3 million US queries, put YouTube at 22.9% of all AI Overview citations and Reddit at 18.5%. The top five domains together captured roughly 65% of all mentions. Brands with no video presence and no active discussion footprint are effectively invisible to a meaningful share of citations, no matter how their owned site ranks.

None of this means rank stopped mattering. Pages that rank well are still cited more often than pages that do not, on average. What changed is the size of the gap between ranking well and getting cited. That gap used to be small enough to ignore. That is no longer true.

How do citation sources differ by query type?

Articles dominate informational prompts. Listicles lead commercial prompts, and product pages take the largest share on transactional ones, according to a query-type breakdown of AI Overview citations published by SEOValley.

A comparison-style listicle aimed at a commercial 'best X for Y' prompt competes in a different citation pool than a straight how-to article aimed at an informational prompt. Match the format to the query type before worrying about word count or keyword density. Format is closer to a citation factor than either of those two ever was.

Local and service-business prompts followed similar logic in the same research: product and category pages outperformed generic homepages for '[service] in [city]'-style questions. A homepage listing every service a business has rarely beats a page built around one service in one location, because the fan-out layer is looking for the answer to one specific question, not a directory of everything a company does.

How should you respond?

Stop treating rank as a stand-in for citation and start measuring citations directly. Pull the AI Overview for your 10 to 20 highest-intent prompts every week and log who gets cited and from what position. That habit alone will tell you more about where you stand than a rank tracker ever will.

Then write for the sub-question, not only the head keyword. If your target term is 'best CRM for small teams,' the fan-out queries probably branch into pricing, setup time, integrations, and migration effort. A page that answers each of those directly, in its own clearly headed section, has a real shot at getting pulled into more than one AI Overview from a single URL.

Build presence outside your own domain, too. Given how much citation volume flows through YouTube and Reddit-style discussion now, a comparison video or a genuinely useful forum answer can out-cite a landing page that outranks it on Google itself. That is the logic behind Generative Engine Optimization: coverage and citability across engines, not a single ranking position on one page.

Scale matters more than any single tactic here, because the fan-out pool is large and it keeps changing. LeadHaste's own site moved from zero organic footprint to 1 million impressions and more than 200 AI citations in four months, with Domain Rating climbing from 1 to 24 over the same stretch, by publishing at a pace and level of specificity that gave the fan-out layer something to quote on nearly every sub-question, not only the primary one.

Track this weekly, not quarterly. The citation mix moved by roughly half in eight months once already, and it shows no sign of settling.

What should you actually change in your content?

Two things matter most here: structure and specificity. Structure means a real H2 for every question a buyer might fan out into, not a wall of prose sitting under one heading. Specificity means a dated number or a named, linked source in the first two sentences of a section, not a vague claim a model has no reason to quote.

Standalone answers matter because AI Overviews lift passages, not full pages. Each section should make sense pulled out of context, because that is exactly how it gets used. Add a comparison table wherever two or more options are being weighed: Google's summaries lean on structured data when it exists, and plain text struggles to compete with a table on those prompts.

None of this replaces getting the content right at the ranking level, too. The two goals reinforce each other. A page built to answer one specific question well tends to rank better and to fan-out cite better, because both systems ultimately reward the same thing: an answer a reader, or a model, does not have to look twice at.

Keep a source close to every number you publish, too. Citation-grade content is not about writing more, it is about writing the kind of sentence a model would rather quote than rewrite. A claim with a name, a date, and a link attached to it does that job. A claim without one falls flat.

What does a citation-ready page look like in practice?

It opens with a direct answer in the first two sentences, the same job a Speakable-eligible TL;DR does at the top of a citation-grade article. It carries a question-shaped heading for every sub-topic a fan-out query might touch.

Pages built this way tend to win twice: once in the organic result a person clicks, and again in the AI Overview a person reads instead of clicking anything at all.

1. A one to three sentence answer at the top, before any framing or backstory.
2. A question-shaped H2 for each sub-question a reader, or a fan-out query, is likely to ask.
3. At least one dated stat, table, or original data point that a model can lift and quote directly.
4. A named, linked source next to every specific number, not a vague reference to unnamed research.

Sources

Source: Ahrefs

AI Overview citations from top 10 organic pages fell from about 76% in July 2025 to 37.9% in March 2026, based on 863,000 keyword SERPs and 4 million AI Overview URLs Ahrefs, 2026-03-02

Source: Pew Research Center

18% of Google searches produced an AI summary in March 2025, and click-through to traditional links dropped from 15% to 8% when a summary appeared Pew Research Center, 2025-07-22

Source: Ahrefs

YouTube (22.9%) and Reddit (18.5%) are the two most-cited domains in Google AI Overviews, with the top five domains capturing roughly 65% of all citations, based on tracking of more than 3 million US queries Ahrefs, 2026-09-02

Source: SEOValley

Citation format tracks query intent: articles dominate informational prompts, listicles lead commercial prompts, and product or category pages lead transactional and local service prompts SEOValley, 2026-09-21