How to rank in AI Overviews: how Google picks sources and a 13-point checklist.

1. How Google picks AI Overview sources.

An AI Overview is the generated answer that sits above the normal results on some Google searches, with a small set of linked sources beside it. It is a feature of Search, built on the same index as the blue links. Gemini writes it: Google said on January 27, 2026 that it was making Gemini 3 the default model for AI Overviews globally. Google AI Overviews SEO starts with knowing the four steps behind that box.

1. Google decides whether to show one. Its guidance says AI Overviews appear only when its systems determine the overview is additive to classic Search, and that they often do not trigger. Long, question-shaped searches trigger them most.
2. It fans the query out. Google describes query fan-out as issuing multiple related searches across subtopics and data sources. One typed query becomes several background searches. Google does not publish them.
3. It retrieves pages from the index. Each background search pulls candidate pages from Google's normal index. Google adds that while a response is being generated, its models identify more supporting pages. There is no separate index for AI answers.
4. It writes the answer and attaches links. The model composes the answer from passages on those pages, and a few of the pages are shown as sources.

How often step one says yes depends on who is counting and what is being searched. Pew Research Center tracked the browsing of 900 US adults in March 2025 and found that 18% of their Google searches produced an AI summary. Length and phrasing mattered most: 8% of one or two word searches got a summary, against 53% of searches of ten words or more, and 60% of searches that began with a question word. Semrush tracked more than 10 million keywords through 2025 and saw the share with an AI Overview rise from 6.49% in January to about 25% in July, then fall back to 15.69% in November. The practical reading: informational, question-shaped queries are where AI Overviews live, and the rate moves when Google retunes it.

Two things follow from the mechanism. There is no list to climb, so to rank in AI Overviews means to be one of the linked sources. And the match is made on the background searches, which are narrower than what the user typed. One point of vocabulary before going further. AI Overviews are the block on the normal results page. AI Mode is the separate conversational tab, and our Google AI Mode guide covers it. Google says the two may use different models and techniques, so the links they show will vary. The same work, across every assistant and engine, is what we call AI search optimization.

2. The one rule Google publishes: indexed and snippet-eligible.

Google's Search Central guidance on AI features is short and blunt. To be eligible as a supporting link in AI Overviews, a page “must be indexed and eligible to be shown in Google Search with a snippet.” It adds: “There are no additional technical requirements.” The same page rules out the shortcuts people sell. You do not need new machine readable files, AI text files or markup, and there is no special schema.org structured data to add. Anyone selling SEO for AI Overviews as a file you upload is selling something Google says it does not read.

That makes the crawler question simple, and it is where most confusion sits. Five user agents get mixed up. Only one of them decides whether you can appear in an AI Overview.

Google-Extended is the one people get wrong. It is a control token, and it governs whether your content is used to train Gemini models and to ground answers in Google's other AI products. Google's crawler documentation says: “Google-Extended does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search.” Blocking it will not take you out of AI Overviews, and allowing it will not put you in. The other three bots belong to other products. They matter for the answers those products give, which our guide to ranking in ChatGPT, the ClaudeBot guide and the Perplexity SEO guide cover one by one.

# robots.txt
# Google Search: required for AI Overviews
User-agent: Googlebot
Allow: /

# Gemini training and grounding outside Search:
# your call, no effect on AI Overviews
User-agent: Google-Extended
Disallow: /

<!-- In the page head -->
<!-- Not eligible: a page with no snippet cannot be a source -->
<meta name="robots" content="nosnippet">

<!-- Eligible: indexable, no limit on snippet length -->
<meta name="robots" content="index, follow, max-snippet:-1">

The second half of the rule is the snippet. Google's guidance says that to limit the information shown from your pages in Search you use the nosnippet, data-nosnippet, max-snippet or noindex controls, and those controls apply here. So three checks decide eligibility.

• Indexed. Run the page through URL Inspection in Search Console. Google's own best-practice list starts with making sure crawling is allowed in robots.txt and by any CDN or hosting infrastructure, because a firewall rule can block Googlebot without a line of robots.txt changing.
• Snippet allowed. Search the page source and response headers for nosnippet and max-snippet. A nosnippet rule left on a template makes every page that uses it ineligible.
• Text in the HTML. Google lists making sure important content is available in textual form. An answer inside an image, a PDF or a tab that loads on a click is harder to retrieve and quote.

If you want out, the same controls work in reverse, page by page. For a whole site, Search Engine Journal reported on August 31, 2026 that a Search generative AI control in Search Console settings had reached all sites. It excludes a site's links and content from AI Overviews, AI Mode and generative AI features in Discover, and per Google it is not used as a ranking signal elsewhere in Search. Check that nobody on your team has switched it to exclude. To audit the crawl side properly, work through the crawl checks in our technical SEO audit checklist.

3. Do you need to rank in the top 10?

Less than you used to. In March 2026 Ahrefs analyzed 863,000 keyword results pages and about 4 million URLs cited in AI Overviews. It asked one question: where does each cited page rank for the query that produced the overview?

• 37.9% of cited URLs were in the top 10.
• 31.2% ranked in positions 11 to 100.
• 31.0% did not rank in the top 100 at all.

Eight months earlier, in July 2025, an Ahrefs study had put the top 10 share at about 76%. Ahrefs names three reasons the number fell: query fan-out pulling from the results of related searches, the January 2026 switch to Gemini 3, and its own improved detection of citations. The first one is the mechanism from section 1 showing up in the data. A page that ranks 40th for the query you track can be the best result for one of the background searches, and you never see that search.

Read the numbers both ways. More than a third of citations still go to top 10 pages, which is far more than ten results out of hundreds would get by chance, so classic ranking work is not wasted. But nearly two thirds go elsewhere, so a ranking report for one keyword tells you little about whether you are cited. The same study found that YouTube made up 5.6% of all cited URLs and 18.2% of the cited URLs that did not rank in the top 100, and that YouTube is the most cited domain in AI Overviews. Text is not the only format in the candidate set.

The working rule: stop asking whether you rank for the head term, and start asking which sub-questions sit behind it and whether you have the best short answer to each. That is the shift our piece on how GEO differs from SEO describes, and it is why search everywhere optimization treats YouTube and forums as part of the same job.

4. AI overview SEO: what makes a page citable.

Google does not publish how it weighs one retrieved page against another, and nobody outside the company has the formula. What exists is Google's own list of what still matters, plus studies of the pages and brands that get picked. Together they point at six properties.

• The answer comes first, in a passage that stands alone. The model lifts passages, not whole pages. A heading phrased like the question, followed by two or three sentences that make sense with nothing around them, is the easiest thing to lift. That is the core of answer engine optimization.
• It covers the sub-questions. Fan-out means the definition, the how to, the cost, the comparison and the benchmark are each a separate search. Give each one its own section, or its own page when it deserves one, and link them together.
• The facts are specific and sourced. The GEO benchmark presented at the KDD conference in 2024 tested content edits against generative engines and found that adding statistics, quotations and cited sources raised visibility in generated answers by up to 40%. It did not test Google's system, so treat it as direction. The generative engine optimization guide walks through that study.
• The page is current. Show an updated date and keep the figures behind it true. An overview that quotes a stale price or an old limit is a bad answer, and the system has other pages to choose from.
• The markup tells the truth. Google's list includes making sure structured data matches the visible text on the page. No schema type unlocks AI Overviews. Markup that disagrees with the page is still a fault worth fixing.
• The brand is known. Ahrefs studied 75,000 brands in May 2025 and found that branded web mentions had the strongest correlation with how often a brand appeared in AI Overviews (0.664). Backlinks came in at 0.218. Brands in the top quarter for web mentions averaged 169 AI Overview mentions. The next quarter averaged 14. That is correlation, and big brands score high on everything, but the direction is clear.

None of this is a trick. It describes a good reference page, and Google's guidance says as much: existing SEO fundamentals continue to be worthwhile. The broader discipline of writing for machine readers is what we mean by LLM SEO. The hard part is volume. One topic fans out into five or six questions, each one needs a direct answer with a source behind it, and the set has to stay current.

5. What an AI Overview citation is worth.

Be honest with yourself about this before you spend a quarter on it. A citation is visibility first and traffic second.

• Fewer clicks on the results below. In Pew's March 2025 panel, people clicked a traditional result on 8% of visits to a results page with an AI summary, against 15% on pages without one.
• Very few clicks on the sources. In the same data, people clicked a link inside the summary on 1% of visits.
• The top result loses the most. Ahrefs compared 300,000 keywords across December 2023 and December 2025 and found that an AI Overview correlates with a 58% lower click-through rate for the top-ranking page. Its April 2025 version of the same study put the figure at 34.5%.

Google's position is that the clicks which survive are better ones. Its guidance says clicks from results pages with AI Overviews “are higher quality (meaning, users are more likely to spend more time on the site).” It has also been adding ways out of the box. On May 6, 2026 Google announced more links placed directly within AI responses next to the text they support, suggestions for where to go next at the end of many responses, and a preview of the site when you hover over an inline link on desktop.

So measure the right thing. For a query where the overview already answers the question, the choice is between being named in the answer and being absent from the screen. Count impressions and brand searches for those. Spend the most effort on queries where the reader still needs to click after the summary: a comparison, a price, a tool, a decision. Our AI SEO statistics page tracks how these click numbers are moving.

6. The how to rank in AI Overviews checklist.

Thirteen checks, in the order to do them. Eligibility comes first because nothing else counts until Google can index the page and show a snippet from it. This is Google AI Overviews SEO reduced to what you can verify. For the classic search basics underneath it, use our SEO checklist.

7. How to measure it in Search Console.

Until 2026 this was guesswork. It is now partly reported by Google itself. Three signals cover it.

• The generative AI performance report. Search Console has a report that shows impressions from AI Overviews and AI Mode, meaning how many times links to your site were shown in those features. You can group it by page, country, date and device. Google's help page says: “As of August 31, 2026, we've rolled out these insights to all websites worldwide.” Two limits: it reports no clicks, and it can be empty when a site has not had enough impressions in AI features.
• The normal Performance report. Google's guidance says traffic from AI features is included in the Performance report within the Web search type. You cannot split it out, but you can watch clicks and impressions for the pages you rewrote and the question-shaped queries they answer.
• A fixed query list. Write down 20 queries where an AI Overview shows and you want to be a source. Check them every month in a signed-out browser and record whether your page is linked. Overviews vary from run to run, so follow the trend across months. The AI Overview trackers we reviewed automate this for Google, and the AI visibility tools roundup covers the ones that watch ChatGPT and Perplexity as well.

Expect the ground to move. The share of citations going to top 10 pages halved between two Ahrefs studies eight months apart, and the model behind the overview changed in between. A page that is cited today holds that place only while it stays the best short answer to one of the background searches. Plan for a steady supply of current pages, not a one-off rewrite.

A worked example: 30 days from ignored to cited.

Take a fictional company, Kestrel Desk, that sells help desk software to small online stores. The company, its site (on the reserved .example domain) and every number below are invented to show the method. The mechanics match the documentation and studies above.

Kestrel Desk has a blog post targeting the query “what is first response time”. Google shows an AI Overview for it. The post ranks 14th and is never one of the linked sources.

The team could not see Google's background searches, so they wrote down the ones a reasonable system would run: what first response time is, how to calculate it, what a good one is for email support, how it differs from resolution time, and how to reduce it. Then the fixes ran in the same order as the checklist.

1. Day 1, eligibility. Removed the nosnippet tag from the blog template and confirmed the change with URL Inspection.
2. Days 2 to 5, the main page. Rewrote it so each of the five questions is a heading with a direct answer under it, and added a benchmark from Kestrel Desk's own ticket data with the date range stated.
3. Days 6 to 24, the cluster. Published four deeper pages, one per sub-question, each linked from the main page and back.
4. Day 25, video. Recorded a four minute walkthrough of the calculation, put it on YouTube and embedded it on the main page.

Here is the top of the main page before and after.

What worked is not clever. Removing one tag made the pages eligible. The rewrite gave Google a passage it could lift. The four deeper pages matched the background searches the single page had missed. Note the shape of the result as well: AI impressions grew far faster than clicks, which is the pattern from section 5. The tag took a day to fix. The pages took the rest of the month, and Kestrel Desk has thirty more support metrics its buyers search for.

Common mistakes when trying to rank in AI Overviews.

1. Blocking Google-Extended and expecting to leave AI Overviews. The name suggests it controls Google's AI. It controls Gemini training and grounding outside Search. To leave AI Overviews, use the snippet controls or the Search Console setting. To stay in, leave Googlebot alone.
2. Buying a file or a schema fix. It is quick to ship and easy to invoice, which is why it gets sold. Google says no AI text file, markup or special schema is needed, and our llms.txt guide shows what that file does and does not do. Spend the time on the pages.
3. Treating a top 3 ranking as a citation. It feels safe because the page already wins. But about 62% of cited pages are outside the top 10 for the query. Check the overview itself before assuming you are in it.
4. Burying the answer. Long introductions exist because writers were told word count ranks. A model looking for a passage to lift will take the page that answers in the first two sentences.
5. Writing one giant page for the whole topic. It looks thorough. But each background search is narrow, and a page that brushes ten questions is rarely the best answer to any one of them. Split the sub-questions out and link them.
6. Reporting AI impressions as traffic. The new Search Console report counts links shown, not clicks. Put it next to the normal Performance report and to signups before you call it a win.