How Google AI Overviews pick citations, and how to become one

Google AI Overviews cite pages that are indexed, eligible to show a snippet, and that contain a passage answering one of the many related searches Google runs behind the scenes for a query. Google says there are "no additional requirements" and no special markup or files needed. The independent data adds the part Google does not spell out: in a March 2026 Ahrefs study, only 37.9 percent of cited URLs ranked in the top 10 for the query itself, because Google fans the question out into sub-questions and cites the best answer to each. To be cited, be the clearest answer to the sub-questions, not just the head term.

What Google documents

Google's page on AI features is short and worth reading in full. Four points from it shape everything else.

• Eligibility: "a page must be indexed and eligible to be shown in Google Search with a snippet." Nothing more is required.
• No special optimization: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." Google adds that you do not need "new machine readable files, AI text files, or markup", and that there is no special Schema.org type for these features.
• Query fan-out: both AI Overviews and AI Mode may issue "multiple related searches across subtopics and data sources" to build a response, which surfaces "a wider and more diverse set of helpful links."
• Controls and reporting: snippet controls such as nosnippet, data-nosnippet, and max-snippet apply to AI features, and AI feature traffic is reported inside the normal Web search type in Search Console, not separately.

Two myths fall out of that list. An llms.txt file does not make you eligible or more likely to be cited; Google says no AI text files are needed. And FAQ, HowTo, or any other schema type is not an AI Overview signal in itself. Structured data still helps Google understand the page and the entity behind it, which is useful, but there is no markup that requests a citation.

What the independent data adds

The most important number for strategy is the overlap between ranking and citation. Ahrefs analyzed 863,000 keyword results and 4 million AI Overview citations and found that 37.9 percent of cited URLs appeared in the top 10 results for the query, down from roughly 76 percent in its earlier analysis. Ahrefs attributes the drop to fan-out: AI Overviews increasingly cite pages that rank for the related searches rather than the original one. YouTube was a notable source among the cited pages that did not rank in the top 100 at all.

The second number is why citation matters. Ahrefs' February 2026 update, based on 300,000 keywords and December 2025 Search Console data, found that the presence of an AI Overview correlated with a 58 percent lower clickthrough rate for the page in position one, up from 34.5 percent in its April 2025 study. Pew Research Center's analysis of 900 US adults' real browsing in March 2025 found people clicked a traditional result on 8 percent of visits when an AI summary appeared, against 15 percent when it did not, and clicked a link inside the summary on just 1 percent of visits. For informational queries, the overview is now the page one result, and being named inside it is the new top position.

The third is about which brands get named. In Ahrefs' study of 75,000 brands, branded web mentions correlated with appearing in AI Overviews at 0.664, while backlinks correlated at 0.218. Brands the web talks about are the ones the overview mentions, which is a reason to treat earned mentions as part of AI Overview strategy rather than a separate PR exercise.

A practical sequence for earning citations

1. Confirm eligibility. The page must be indexed, must not carry nosnippet or a tiny max-snippet value, and its key content must be in the HTML Google renders. Check indexing in Search Console's URL Inspection tool.
2. Map the fan-out. For each target query, write down the sub-questions a thorough answer would cover: definitions, costs, comparisons, steps, risks, and "best for" questions. The People Also Ask box and the follow-up suggestions in AI Mode are a free view of them.
3. Answer each sub-question in its own passage, under a heading that states it, with the answer in the first sentence. Our guide to writing citable passages shows the pattern.
4. Add specifics with sources: figures, dates, named standards, and quotes from people with real expertise. These are the elements engines can attribute with confidence.
5. Cover the topic, not just the page. A cluster of pages that each answer one sub-question well gives fan-out several ways to find you.
6. Earn mentions on the sites that are already cited for your topics, such as industry publications, comparison pages, and community threads.
7. Keep facts current. Prices, statistics, and product details go stale fast, and an outdated figure is easy for an engine to pass over in favor of a fresher source.

A worked fan-out example

Take the query "is a heat pump worth it in Maryland." A complete answer touches at least five sub-questions, and each one is a separate chance to be cited: what a heat pump costs to install in the state, how much it saves compared with a gas furnace, which rebates and tax credits apply, how well it performs in a Mid-Atlantic winter, and how to choose an installer. A contractor whose site has one generic "Heat pumps" page competes for none of them cleanly. A contractor with a pricing page that states typical installed costs, a rebate page that names the programs and their dates, and a cold-weather page with real performance figures has three pages that each answer one sub-question better than anyone else nearby.

Measuring AI Overview traffic in Search Console

Google reports AI feature traffic inside the normal Web search type, so there is no AI Overview filter to isolate it. The practical proxy is to watch, for informational landing pages, impressions rising while clickthrough rate falls at a stable position. That pattern usually means an AI Overview is answering above you. Pair it with direct observation: search the query yourself on a few days and record whether you are cited.

What does not seem to matter much

• Page length on its own. A long page with no clean answers gives fan-out nothing to cite; a short page with one excellent answer can be cited for that sub-question.
• Exact-match keyword repetition. The Princeton-led GEO study found keyword stuffing lowered visibility in generated answers.
• Special files and tags. Google's documentation rules out llms.txt and special markup as requirements.

How SEO for AI Agents measures this

Because AI Overviews are part of Google Search, the technical floor is the same one our audit checks for every page: indexable, server-rendered content, measured by the AI crawler readability check, and clean structured data, validated by the JSON-LD validity check. On top of that, the passage extractability check shows which paragraphs on a page are shaped to be quoted, with the word count and heading behind each one.

For the outcome, we ask AI Overviews and the other engines the questions your buyers ask and record, over repeated runs, whether you were cited, with the verbatim answer and timestamp as the receipt. The engine divergence view then shows where AI Overviews cites you and another engine does not, or the reverse, which is often the quickest way to spot a fan-out sub-question you have not answered. The glossary entry on AI Overviews covers the feature's history and rollout.