Explainer · 4 min read
How AI Search Tools Choose Which Sources to Cite
No AI company publishes its full source-selection method. Here is what is documented, what studies suggest, and where the answer is still unknown.
Short answer
No AI search company publishes its full method for choosing sources. What is documented: Google's AI features may run several related searches and need pages that are indexed and snippet-eligible, and ChatGPT search ranks results on multiple factors with placement not guaranteed. Studies of which pages get cited change quickly, so treat them as snapshots.
Key takeaways
- Google says its AI features may issue several related searches (query fan-out), so a page can be cited for a sub-question.
- The documented technical bar for Google is an indexed page eligible to show a snippet, with no extra requirements.
- OpenAI, Perplexity and Anthropic each run a search crawler that site owners can allow or block.
- Ahrefs found the share of AI Overview citations that also rank in the top 10 fell from about 76% to about 38% between two studies.
- Repeated runs of the same question often produce different sources, so single results mislead.
What the companies document
Each company tells site owners a little about how its product finds pages. None describes how it ranks one source above another. This is the full public picture as of the update date on this page.
| Product | What is documented |
|---|---|
| Google AI Overviews and AI Mode | A page must be indexed and eligible to appear in Search with a snippet. There are no additional technical requirements. The features may issue several related searches across subtopics and data sources, which Google calls query fan-out, so they can show a wider set of supporting links. |
| ChatGPT search | OAI-SearchBot surfaces sites in ChatGPT's search features. OpenAI says ChatGPT ranks results using multiple factors intended to help users find relevant, reliable information, and that placement is not guaranteed. Responses that use search may include citations and a Sources list. |
| Perplexity | PerplexityBot is designed to surface and link websites in Perplexity search results. Perplexity-User fetches pages when a person asks a question, and generally ignores robots.txt because the person requested the fetch. |
| Claude | Claude-SearchBot analyzes content to improve the relevance and accuracy of search results. Claude-User retrieves pages in response to a person's request. |
| Microsoft Copilot and Bing | Bing Webmaster Tools reports citations in Copilot, Bing AI summaries and some partner experiences. It does not report placement within an answer or any ranking. |
The sources for this table are listed at the bottom of the page. For the crawler details, see AI crawlers and robots.txt.
Query fan-out and what it means for your pages
Query fan-out is the most useful idea in the documentation. When a person asks a broad question, the system may break it into several narrower searches. A person asking about a car accident might trigger separate searches about deadlines, insurance, medical bills and what to do at the scene.
Google does not expose the fan-out queries it generates, so you cannot see them. You can still design for them. Our advice is to give each page one clear job and answer it directly, use headings that read like the sub-questions a client would ask, and link related pages together. The content playbook shows how to build that set of pages.
What the studies show, and why they disagree
Vendors have measured which pages AI answers cite. We cite only the studies whose method we could check, and the one below shows how quickly such numbers can move.
- Ahrefs, March 2026. Across 863,000 keyword searches and about 4 million cited AI Overview URLs, about 38% of cited URLs also ranked in the top 10 for the same search. In its July 2025 analysis the figure was about 76%. The Ahrefs author said that fan-out searches are a likely explanation, since the extra searches pull in pages that do not rank for the original query.
Our reading: ranking in the top 10 is neither required nor sufficient for a citation. A page that answers a narrow question well can be cited without ranking first for the broad term. That is a reason to publish specific answers, not a reason to stop doing ordinary SEO.
Why results change between runs
In a January 2026 study, SparkToro's Rand Fishkin and Patrick O'Donnell of Gumshoe.ai had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times. The chance that two responses contained the same list of brands was under 1 in 100, and the chance of the same list in the same order was under 1 in 1,000. Ahrefs makes a similar point about AI Overviews: they are probabilistic, and their citations and fan-out searches change with every query.
For a law firm, the lesson is practical. A screenshot showing your firm named once, or missing once, tells you almost nothing. Measure appearance across many runs, as described in the AI answer audit.
What you can do with this
- Make sure the basics are in place: pages indexed and eligible for snippets, and the search bots of each major product allowed.
- Write specific pages that answer one question each, and connect them with internal links.
- Keep your identity consistent so a system that finds your page can tell who stands behind it. See attorney bios and firm identity.
- Measure appearance over many runs, and watch Bing's AI Performance report for citations.
What nobody can tell you yet
No company publishes the weights it gives to page quality, links, freshness or third-party mentions in AI answers. Any vendor that claims to know the exact formula is guessing or testing on a sample. Ask what the sample was, how many runs it covered and whether the method is public.
Common questions
Does ranking on page one still matter for AI citations?
It still helps to be indexed and well regarded by the underlying search systems, but it is not the whole picture. Ahrefs reported in March 2026 that about 38% of URLs cited in AI Overviews also ranked in the top 10, down from about 76% in its earlier study. The rest came from lower positions or from pages that did not rank for that exact search.
Can I see which of my pages an AI tool cited?
Partly. Microsoft's Bing Webmaster Tools has an AI Performance report, in public preview, that shows citations of your pages in Copilot and Bing AI summaries. Google counts AI Overviews and AI Mode traffic inside the Web search type in Search Console, without a separate line for it. See metrics that matter.
Why does my firm appear in one answer and not the next?
Answers vary run to run. In SparkToro's January 2026 study the chance of getting the same list of brands twice was under 1 in 100. That is why we recommend measuring how often you appear across many runs. See the AI answer audit.
Sources
- Google Search Central: AI features and your website
- OpenAI Help Center: Searching the web with ChatGPT
- OpenAI: Overview of OpenAI crawlers
- Perplexity: Perplexity crawlers
- Anthropic Help Center: Does Anthropic crawl data from the web, and how can site owners block the crawler?
- Microsoft Bing: Introducing AI Performance in Bing Webmaster Tools
- Ahrefs: Update, 38% of AI Overview citations pull from the top 10
- SparkToro: AIs are highly inconsistent when recommending brands or products
Informational only. This article is general education about search and AI visibility. It is not legal advice, it does not create an attorney-client relationship, and we do not guarantee that any statement in it is accurate, complete or current. No one can guarantee that an AI tool will name your firm. Lawyer advertising rules differ by state, so check your state bar's rules and talk to a qualified attorney before you publish claims about your firm. The update date above shows when we last checked.