Which AI models power Google AI Mode and AI Overviews?

Google has published a default model for one of its AI search surfaces and only a family name for the other.

• AI Mode: Gemini 3.5 Flash has been the default since May 19, 2026, per Google's Search I/O announcement. That was still the stated default in our 21 September 2026 snapshot of what the consumer apps run.
• AI Overviews: Google has described the model as "Gemini 3", with that family as the default since January 27, 2026. This comes from a 9to5Google report quoting Google, not a Google documentation page. Google publishes nothing finer than the family name.

So nobody outside Google can tell you which exact model wrote a given AI Overview. Treat any article that names a specific AI Overviews version as guessing.

Google's own AI features documentation adds the point that matters most for site owners: AI Overviews and AI Mode "may use different models and techniques." They are two products. Expect them to behave like two products. Our side-by-side comparison of AI Overviews and AI Mode covers the rest of the differences.

A different model is only one reason the link lists diverge. Three other mechanisms sit on top of it.

Different techniques

Google says AI Mode may use a "query fan-out" technique, issuing multiple related searches across subtopics before it writes an answer. That means AI Mode retrieves from a wider and different set of results than the single query you typed. The AI Mode query fan-out explainer walks through how that changes source selection.

Different triggering

AI Overviews often do not appear at all. AI Mode always answers, because the user chose it. A query can have an AI Mode answer with five sources and no AI Overview, so "cited in Google AI" can be true on one surface and meaningless on the other.

The queries are not the ones you test

Both surfaces generate their own retrieval queries. You can test the prompt a buyer types, but you can't see the searches Google actually ran on their behalf. That gap applies to every prompt-based tracker, including ours.

The practical result: the same page can be a supporting link in AI Mode and absent from the AI Overview for the same query, with nothing wrong on your site.

"Google AI" usually blurs three separate things. Keep them apart.

The Gemini API's Grounding with Google Search is a fourth path. It is what developers call, and it automatically generates its own search queries. It is not AI Overviews or AI Mode, so results from the API are a proxy at best for what Search shows.

This matters for controls. Google-Extended governs training and grounding for Gemini Apps, and Google says it does not affect a site's inclusion in Google Search. If you are wondering whether blocking it hides you from Search features, read whether blocking Google-Extended removes you from AI Overviews.

Different models do not mean different entry rules. Google documents one gate for both: to be eligible 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 that there are no additional technical requirements and no special optimizations necessary.

The controls follow from that:

• robots.txt: there is no token that blocks AI Overviews or AI Mode specifically. Googlebot rules are the control.
• Snippet controls: noindex, nosnippet, max-snippet and data-nosnippet keep content out of AI answers. Accidental template use is a real failure mode.
• Search Console control: inclusion in Google's generative surfaces is on by default for all properties. Child properties can inherit an opt-out from a parent. See whether you need to opt in to AI Overviews.

Eligibility is not inclusion. Google also states that indexing and serving aren't guaranteed. What you control is whether you are blocked.

Knowing the model name gives you nothing to optimize. Google's guidance says you don't need to write in a specific way for generative AI search, and no published study shows a page-level technique that reliably moves AI citations on these surfaces. Spend the effort where the evidence is strong.

1. Confirm the gate. Check that key pages are indexed, carry no noindex or snippet restrictions, and that Googlebot can fetch them.
2. Check the Search Console control, including any parent property a child inherits from.
3. Measure each surface separately. An answer from AI Mode tells you nothing about the AI Overview for the same query. A free way to compare AI Mode and AI Overview answers shows the gap for your own queries.
4. Read your own impressions. The Search Console generative AI report covers both surfaces, as impressions only, with no clicks or position.
5. Recheck after model changes. Google swaps defaults without notice to site owners. Date every observation you record.

Does knowing the model help me get cited in AI Overviews? No. Google says there are no special optimizations needed to appear in AI Overviews or AI Mode beyond being indexed and snippet-eligible. The exact AI Overviews version isn't published, and defaults change. Your controllable levers are access ones: indexing, Googlebot permission, snippet controls and the Search Console inclusion setting. None of them depend on which Gemini model is running.

Is AI Mode the same as the Gemini app? No. As of our September 2026 snapshot, AI Mode defaulted to Gemini 3.5 Flash, while the Gemini app listed Gemini 3.6 Flash as selectable and stated no free-tier default. They also retrieve differently: AI Mode draws on the Google Search index, while the Gemini app is governed separately, including by Google-Extended. Testing one tells you little about the other.

Can Search Console show which surface showed my page? Partly. Search Console's generative AI report covers AI Overviews and AI Mode, but as of October 2026 it reports impressions only, with dimensions for pages, countries, dates and devices. It shows no clicks, no click-through rate and no position. It also does not tell you which model produced the answer.

Why does the same query give different sources a week later? Several things move at once. Google can change the default model, AI Mode fans out into multiple searches that can vary, and AI Overviews may not trigger every time. The underlying search index also changes. A single observation is a snapshot, so record the date and surface with every result and compare like with like.