Google has not published how AI Overviews pick the sites they link to. What it has published is an eligibility rule: a page has to be indexed and eligible to be shown in Google Search with a snippet, and that is the whole technical requirement. How Google chooses among the pages that meet it is not documented anywhere I could find.
So we looked at what Google actually linked. We asked Google 50 B2B software buying questions on 1 and 2 October 2026, kept 100 AI Overviews and their 683 source links, and lined every link up against the ordinary results on the same page. Ranking is part of the picture. It is a smaller part than most advice assumes.
- 100
- AI Overviews
- 683
- source links
- 50
- questions, each asked on 2 days
- 3
- overviews with no links
Captured from Google results pages on 1 and 2 October 2026, US English, desktop. The questions came from 5 B2B software categories with no brand names in them.
What Google says, and what it leaves out
Google's page on AI features and your website (last updated 10 December 2025) is short on specifics. To appear as a supporting link, it says, "a page must be indexed and eligible to be shown in Google Search with a snippet." Then: "There are no additional technical requirements." And a warning: "Indexing and serving isn't guaranteed."
It is just as direct about extra work. You "don't need to create new machine readable files, AI text files, or markup to appear in these features," and "There's also no special schema.org structured data that you need to add."
It also describes how an answer gets built. Both AI Overviews and AI Mode "may use a 'query fan-out' technique," and "while responses are being generated, our advanced models identify more supporting web pages." The page adds that AI features "display a wider and more diverse set of helpful links" than classic search.
That is where the documentation stops. The AI Overviews help page tells readers to click the links to supporting information to check what an overview says. I could not find a Google page that says how it picks among eligible pages.
From here I keep two kinds of statement apart. "Google says" is a quote from its documentation. "We found" comes from our captures and describes one sample.
What we captured
Each of the 50 questions came from one of five B2B software categories (sales tools, CRM, workflow automation, ERP and marketing automation), written the way a buyer types them, with no brand names. Each was asked once on 1 October and once on 2 October, from a US location in US English on desktop. For every capture we kept the overview text, its source links in the order they came back, and the first ten ordinary results beside it. The same run, across five engines, is described in which sources ChatGPT, Gemini and Perplexity cite and on the research page.
Ninety-seven of the 100 overviews had source links. The other three, all from 2 October, came back with two to three thousand characters of text and no links: the best AI phone agent for answering business calls, the best ERP for ecommerce, and the best financial management software for mid-market companies. We cannot tell from the capture whether Google showed them that way or the links failed to load, so they are left out of the link counts below. The 97 averaged 7 links each, from 2 to 19.
Ranking is part of it, and position 1 is not special
Of the 683 links, 241 (35%) were a page that also appeared among the ordinary top 10 results for the same question on that capture. Eighty-six of the 97 overviews linked at least one such page. So the two lists overlap, often. They are not the same list.
Share of cited links, %. 97 AI Overviews with source links, from 50 B2B buying questions asked on 1 and 2 October 2026, US English. "In the top 10" means the same URL was among the ordinary results on that capture. Video or forum is a YouTube, Reddit, Quora or LinkedIn page that was not.
Another 31 links (4.5%) were a different page on a site that had a top-10 result. The remaining 60% came from sites with nothing in that question's top 10, or were YouTube, Reddit, Quora or LinkedIn pages that were not in it.
If Google simply copied its own ranking, position 1 would be cited most and the rate would fall away down the page. That is not what the captures show.
Share of overviews that cited the page at that position, %. Same 97 overviews. Positions 8 and 9 have fewer rows because some results pages showed fewer than nine ordinary results next to the overview.
The result at position 1 was linked in 25 of 97 overviews. The result at position 4 was linked in 37. Every position from 1 to 9 landed between 21% and 38%, and with 97 overviews a few points either way is noise. The claim I would defend is narrow: there is no steep slope. Being first did not make the overview more likely to link you.
Other studies reach a similar place by a different route. Ahrefs reported 76% of cited pages in the top 10 in July 2025, then published an update in March 2026 that put it at 37.9% across 4 million AI Overview URLs. By Ahrefs' own account, AI Overviews had moved to Gemini 3 in January 2026 and it had changed how it parsed them, so the two figures are not a clean before and after. A second test in the update, counting blue links only, gave 37.1%. Ours was 35.3%, from far fewer overviews and one kind of question, so I would call it close and nothing more precise.
The first links rank. The later ones are videos and forums
An overview's source list is ordered, so we looked at how the top-10 share changed down the list.
Share of links at that place that were also an ordinary top-10 result, %. Same 683 links, in the order Google returned them. The lists ran from 2 to 19 links, with 7 as the median.
Of the first links, 55% were top-10 pages. By the seventh link or later it was 15%. One caution: we know the order the capture returned the links in, and we have not checked that this matches the order a person sees on screen in every case, so read this as a pattern in the data's order.
The tail is where video and forums sit. There were 88 YouTube links in the 100 overviews, and only 5 of them were a video that also appeared in the ordinary results. YouTube gets into the overview through a different door from the one the blue links use. Ahrefs found the same shape at a larger scale: 18.2% of citations from outside the top 100 were YouTube URLs.
Which videos and threads show up depends on the question
Across all 100 overviews YouTube was linked in 51 and Reddit in 48, which is what put them at the top of the earlier post. Split by category, the picture is less uniform.
CRM
AI agents and workflow automation
Marketing automation
ERP, accounting and finance
Sales tools
YouTube went from 3 of 20 overviews on sales tools to 15 of 20 on CRM.
Share of 20 overviews per category. 10 questions per category, each asked on 1 and 2 October 2026. Counts are overviews that linked at least one YouTube or Reddit page.
YouTube appeared in 15 of 20 CRM overviews and in 3 of 20 on sales tools. Reddit moved less, from 6 to 11 of 20 in every category, and it was the more common of the two on ERP and on sales tools. There are only 10 questions per category, each asked twice, so treat this as direction. It does say that "what does the overview link for my category" is a question with a different answer in each one, and it is cheap to look before you decide what to write.
What the cited top-10 pages looked like
Among the ordinary top-10 results on the same captures, we compared the titles of pages the overview linked with those it did not. Our questions all start with "what is the best", so almost every results page is full of roundups, and that limits what the comparison can show.
"Top 10" or "N best" in the title
"best" in the title
2025 or 2026 in the title
"how to", "what is" or "guide" in the title
241 top-10 pages the overview cited, 578 it did not.
Share of pages with that wording in the title. Ordinary top-10 results on the same 97 captures, split by whether that overview linked them. A description of this sample, not a test of cause.
Titles with "Top 10" or "N best" in them were less common among the cited pages (24%) than among the ones skipped (47%). I would not read that as "roundups lose". More than half the cited top-10 pages still had "best" in the title. A better reading is that, among very similar roundups, a title built around a list number did not help a page get picked. We did not test why, and 241 against 578 pages is a modest sample.
Day to day, the list moves
The same questions, asked a day apart, gave overviews that linked different pages.
Mean overlap of pages (shared divided by all on either day), %. 47 questions whose overview had source links on both 1 and 2 October 2026. The ordinary top 10 is the same 47 questions on the same two captures.
For the same question, the pages linked on 1 October and on 2 October overlapped by 38% (pages in both, divided by all pages linked on either day). The ordinary top 10 for those 47 questions overlapped by 54%. By site rather than by page, the overview overlap was 42% in the earlier run. One reading on one day is a weak basis for deciding that a page is in or out.
How to rank in AI Overviews: what this data supports
Nothing here proves what causes a citation. These are the steps I would take, in order, and the reason for each.
- Check that the page can be shown with a snippet. Google's eligibility rule is indexed plus snippet-eligible, and its own list of controls that limit what Search shows includes nosnippet, data-nosnippet, max-snippet and noindex. The URL Inspection tool in Search Console will tell you whether a page is indexed.
- Read the overview for your actual question before writing anything. Note whether the links are vendors, roundups, videos or threads. The free AI Overview checker lists the sources for up to three keywords and marks yours.
- If the overview mostly links videos and threads, a blog post is competing for the smaller share of its links. In our CRM questions, YouTube appeared in 15 of 20 overviews.
- Answer the question the way the neighbouring searches would. Google says fan-out finds "more supporting web pages", which may be how pages outside the top 10 for the typed question get in: they rank for a nearby search. We did not test that. The query fan-out tool lists the related searches for a question so you can see what the page would need to cover.
- Check again on other days. With the cited pages overlapping only 38% from one day to the next, a single look tells you little in either direction.
If your category is B2B software, the older advice to write another "Top 10" post is not supported by this sample. The GEO guide covers what else tends to matter, and how AI engines choose brands to recommend covers the part about being named, which is a separate question from being linked.
Tracking it
CueScout's AI Overview tracker checks Google AI Overviews every day for the buying questions you choose. For each check it stores whether Google showed an overview, every link in it in order, and where each linked page ranked in the ordinary results on the same page, which are the three fields this post was built from. The free AI Overview checker does one live reading of up to three keywords without a signup.
Limits of this sample
- 50 questions, every one a "what is the best" question about B2B software, asked on two days in US English on desktop. How-to, local, health and other query types will behave differently.
- One capture per question per day. Google can show a different overview to the same person a minute later.
- "Position" is the order of the ordinary results in the capture, ten deep. Videos, forum blocks and other features beside them are not counted in it.
- A link counts as "in the top 10" only when the URL matches. YouTube links are matched on the video ID. A cited page that ranks for a different query does not count.
- Titles were compared by wording, and the comparison describes this sample, not a cause.
- Three overviews had no links and are excluded from the link counts. Our earlier post gave 6.5 pages per overview for AI Overviews because it folded all YouTube watch links into one URL. Counting each link separately, the figure over all 100 is 6.8.
Treat the percentages as a snapshot from October 2026, not a constant.
Frequently asked questions
How does Google AI Overviews choose which sites to cite?
Google does not publish the selection rules. Its documentation says a page must be indexed and eligible to be shown in Google Search with a snippet, and that there are no additional technical requirements. In our October 2026 run of 100 overviews, 35% of cited links were pages from the ordinary top 10 for the same question, so ranking is related to the choice but does not decide it.
Do you have to rank in the top 10 to be cited in an AI Overview?
No. In our sample 65% of cited links were not among that question's top 10 results. Ahrefs, which looked at 4 million AI Overview URLs in March 2026, found 37.9% in the top 10. Its July 2025 study had reported 76%; Ahrefs says AI Overviews moved to Gemini 3 and its own parsing changed in between.
Does ranking number one guarantee an AI Overview citation?
No. In our sample the result at position 1 was linked by the overview in 25 of 97 cases (26%). The result at position 4 was linked in 37 of 97 (38%).
Do structured data or an llms.txt file help you get cited in AI Overviews?
Google says there is no special schema.org structured data you need to add, and that you do not need to create new machine readable files, AI text files or markup to appear in AI Overviews or AI Mode. We did not test either in this run.
How can I see whether my site is cited in a Google AI Overview?
Search the question in Google and open the overview's source links, or run it through the free AI Overview checker, which lists the sources and marks yours. Tracking the same questions every day is what an AI Overview tracker is for, because the cited pages change quickly.