Google AI Mode Sources: How Query Fan-Out and Link Selection Work

Google's documentation for site owners on AI features covers AI Mode and AI Overviews together. As of October 2026, it makes four statements that matter for sources:

• Fan-out. AI Mode may use a "query fan-out" technique, issuing multiple related searches across subtopics to build one answer.
• Eligibility. To be shown as a supporting link in AI Overviews or AI Mode, a page "must be indexed and eligible to be shown in Google Search with a snippet." Google adds: "There are no additional technical requirements."
• No guarantee. The same page says "Indexing and serving isn't guaranteed."
• Different surfaces. AI Overviews and AI Mode "may use different models and techniques," so they can return different links for the same question.

That is the documented core. Google also says there are no special optimizations necessary to appear in either surface. Everything else you read about how AI Mode cites is inference. For the full digest of the guidance, see the comparison of AI Overviews and AI Mode.

Query fan-out means AI Mode does not run one search for one question. It may split the question into subtopics and search each one. Your page can enter the answer through a subtopic query you never targeted.

Say you run a hypothetical bike shop in Denver and someone asks AI Mode which commuter bike to buy for snowy winters. The engine might search separately for tire options, frame materials and local service. A page that answers one of those subtopics well could be retrieved even if it never mentions the full question.

Two practical consequences follow:

• You can't see the fan-out queries. Google does not publish them for AI Mode. Any tool that tests prompts is testing a proxy for an input nobody outside Google observes. This is the same problem covered in whether AI search runs the exact question a user types.
• Topical fit beats keyword lists. Google's optimization guide says you don't have to worry about missing "long-tail" keywords or every variation of a query. Research on already-retrieved content points to topical relevance as one of the most reproducible factors, though that work was measured in lab setups, not in AI Mode.

The next move: make sure each page clearly covers its subject, and stop trying to map every phrasing.

AI Mode and AI Overviews draw on the same index and share one gate. If a page fails indexing or snippet eligibility, it is out of both. If it passes, it is eligible for both, but eligible is not the same as selected.

Google's statement that the two surfaces "may use different models and techniques" explains why links differ. As of Google's May 2026 announcement, AI Mode defaults to a Gemini 3.5 Flash model. Google publishes only "Gemini 3" for AI Overviews, with no finer version. Different models, different retrieval steps, different answers.

So don't assume that a citation in one surface predicts a citation in the other. Ahrefs found that 37.9% of URLs cited in AI Overviews also appeared in the first 10 organic results (March 2026). That study covered AI Overviews only. The evidence base has no equivalent figure for AI Mode, so treat any AI Mode overlap number you see with caution. For the AI Overviews side, see how Google picks AI Overview supporting links.

Be plain about the gaps. As of October 2026, Google does not publish:

• How AI Mode ranks or chooses among eligible pages.
• The fan-out queries it ran for any answer.
• How many sources it considers before citing a few.
• Any position or rank metric for AI Mode. Search Console's generative AI report shows impressions only, with no clicks, no CTR and no position.
• A list of signals that raise a page's chance of selection.

This means any "AI Mode rank" sold to you has no ground truth behind it. Google's own guidance also rules out several common tactics. It says you don't need AI text files such as llms.txt, special markup or a specific writing style for its generative AI features. Structured data is not required either. Ahrefs' matched test found schema produced no major uplift in citations on any platform it tracked. Existing markup does no harm, but it is not a lever for AI Mode.

Your levers are access controls. They work by removing blocks, not by adding signals.

Two details catch teams out. First, Google-Extended governs Gemini Apps training and grounding. Google says it does not affect inclusion in Google Search, so blocking it does not remove you from AI Mode. Second, inclusion is the default. You don't opt in, as explained in whether you need to opt in to AI Overviews. The Search Console control is still rolling out to a subset of owners, and child properties inherit from the parent by default. A property can be excluded without anyone on the team choosing it.

One thing you don't need to worry about here: Googlebot renders JavaScript, unlike most AI crawlers. A snippet directive left in a template is a far more common silent block, for example:

<meta name="robots" content="max-snippet:0">

Run these checks in order. Each one tells you whether a block exists, not whether you will be cited.

1. Confirm robots.txt allows Googlebot on the pages that matter.
2. Confirm the pages are indexed, using Search Console URL Inspection.
3. Search the HTML and headers for noindex, nosnippet, max-snippet and data-nosnippet, and check each one was intentional.
4. Check the Search Console generative AI setting on every property, including inherited settings on child properties.
5. Read impressions in the Search Console generative AI report. Avoid trends from mid-August 2026, when a Google-side bug depressed impressions.
6. Look at the live answers. The free Google AI Mode Checker shows which links AI Mode returns for a query, which you can compare with AI Overviews.

For what to do once the access checks pass, see the AI Mode SEO guide.