What model powers Google AI Overviews?

The AI summary above Google’s search results.

As of July 2026, Google’s AI Overviews are powered by Gemini 3 — Google made it the default worldwide on 27 January 2026. Google discloses only the model family: AI Overviews run on “a custom version of Gemini,” and Google never names the exact variant or which one produced a given overview. So this is the vendor’s stated family, not a per-answer model Discoverably can observe.

Current model
Gemini 3 (a custom variant)
Evidence
Vendor-stated
As of
Jan 2026
What we capture
the AI Overview on a live google.com results page

The vendor discloses this default; the engine never labels which model produced a given answer, so we report the vendor’s word with its date, not an observation.

How AI Overviews pick their model

AI Overviews are the AI summary that can appear above Google’s normal search results. They don’t run on a model you can see or choose: Google describes them as powered by “a custom version of Gemini,” and as of 27 January 2026 that custom model is Gemini 3, made the default worldwide. Google discloses the family, never the exact variant — so the honest answer is the family, with its date.

AI Overviews are always built from retrieved web pages — there is no memory-only mode — using Google’s core Search ranking plus query fan-out. Google changes the underlying model on its own schedule; the January 2026 move to Gemini 3 shifted citation patterns noticeably. Discoverably follows the default experience rather than pinning a version.

Model history

Every family-level change to the model behind Google AI Overviews, newest first, each linked to its primary source.

Custom Gemini
LaunchVendor-stated

Google expands AI Overviews to 200+ countries and 40+ languages, powered by a custom version of Gemini.

Google ↗ (opens in a new tab)
Gemini 3 (custom variant)
Model changeVendor-stated

Gemini 3 becomes the default model for AI Overviews worldwide.

Google ↗ (opens in a new tab)
  1. Gemini 3 (custom variant)

    Gemini 3 becomes the default model for AI Overviews worldwide.

    Google ↗ (opens in a new tab)
  2. Custom Gemini

    Google expands AI Overviews to 200+ countries and 40+ languages, powered by a custom version of Gemini.

    Google ↗ (opens in a new tab)

What decides whether Google AI Overviews names your brand

Conditional, Search-grounded Gemini summaries shown only when Google judges an overview additive — so trigger incidence is itself a factor, and citation must be tracked apart from rank.

How answers form & what drives brand mentions

Conditional, Search-grounded, fan-out. VENDOR-DOCUMENTED AIOs are Gemini-generated summaries integrated with Google’s core ranking/quality systems, may use query fan-out like AI Mode, and surface supporting links chosen during generation; they appear only when Google’s systems predict added value. Eligibility/controls are identical to AI Mode (indexed + snippet-eligible; no special markup; structured data optional and must match visible text; Merchant Center/Business Profile freshness for commerce/local). Google-Extended does not control AIOs.

AI features and your website ↗ (opens in a new tab)

Citation vs ranking. INDEPENDENT STUDY (primary-verified) Ahrefs analyzed 4M AIO citation URLs across 863k SERPs (2026-03): ~37% of cited URLs ranked top-10 for the same query, ~26% ranked 11–100, ~37% outside the top 100 — methodology differs from its top-10-heavy 2025 result, so the two are not a clean time series. OUR INFERENCE: strong organic ranking remains the best single predictor of AIO citation, but it’s a probabilistic input to a volatile selection layer, and citation must be tracked as its own metric.

Ahrefs, 2026-03-02 ↗ (opens in a new tab)

Market context — the stakes. INDEPENDENT STUDY (context, not a brand-selection factor) Pew’s behavioral panel (~900 US adults, ~68k searches, Mar 2025) found users clicked a traditional result on ~8% of searches with an AI summary vs ~15% without, and clicked a source inside the summary ~1%. Counterweight: Seer Interactive measured being cited inside an AIO associated with ~35% more organic clicks (and ~91% more paid) vs not being cited on the same SERPs — correlation, not causation. On AIO-present queries, citation inside the overview is the primary visibility asset — classic rank alone no longer captures the value.

Pew, 2025-07-22 ↗ (opens in a new tab)Seer, 2025-09 ↗ (opens in a new tab)

Factor weights

Weight and confidence are our editorial estimate of each factor’s influence, and of the evidence for its direction — not a vendor’s ranking weight. No engine publishes a brand-selection formula.

FactorWeightConfidenceBasis
Whether Google triggers an AIO for the query (brands can’t control the trigger)High (opportunity)HighVENDOR-DOCUMENTED
Google index + snippet eligibility (preview controls)High (binary gate)HighVENDOR-DOCUMENTED
Organic ranking across head query + fan-out sub-queriesHighHighVENDOR-DOCUMENTED integration; INDEPENDENT STUDY shows declining-but-real top-10 coupling
Clear, self-contained, evidence-bearing passagesHighMedOUR INFERENCE from summary mechanics; aligns with vendor guidance
Branded web mentions across the open webMed-HighMedINDEPENDENT STUDY (Ahrefs: ~0.66, the top web correlate; ~a quarter of studied brands had zero AIO mentions); correlational
YouTube mentionsMed-HighMedINDEPENDENT STUDY (Ahrefs ~0.737); correlational
Query intent class (informational triggers AIO far more than commercial)High (exposure gate)HighINDEPENDENT STUDY (multiple trackers; direction robust, figures perishable)
Merchant Center + Business Profile (commerce/local AIOs)Med-High (vertical)HighVENDOR-DOCUMENTED
Generic structured data as a direct boostLow-MedHighVENDOR-DOCUMENTED (not required; retains rich-result value)
Domain-level backlink authorityLow-MedMedINDEPENDENT STUDY (Ahrefs: backlinks ~0.22, ~a third of the mention correlation); correlational

What you can do (ranked)

  1. Defend and extend organic rankings across the question clustercross-engine within GoogleVENDOR-DOCUMENTED + INDEPENDENT STUDY

    AIO citation still couples to ranking more than any other single input — classic SEO against the full sub-question set.

  2. Earn branded mentions in independent coveragecross-engine · strongest measured coupling on Google surfacesINDEPENDENT STUDY + VENDOR-DOCUMENTED boundary

    The mention correlation (~0.66) roughly triples the backlink correlation in the same dataset; digital PR, cited data studies, podcast/YouTube appearances, category-list inclusion — inside Google’s authenticity boundary.

  3. Write the summarizable blockcross-engineOUR INFERENCE

    For each target question, one high-on-page section answering it completely in 2–4 sentences with a supporting fact.

  4. Track citation separately from rank; segment AIO-present queriesengine-specific measurementINDEPENDENT STUDY + VENDOR-DOCUMENTED tooling

    Citation churns without rank change, so rank tracking alone misreads AIO performance.

  5. Commerce/local feed + profile freshnessGoogle-specificVENDOR-DOCUMENTED

    Current Merchant Center feeds and Business Profile for commerce/local overviews.

  6. Set click expectations by intentstrategyINDEPENDENT STUDY

    Informational AIO queries → plan for citation-driven exposure with fewer absolute clicks; commercial queries → AIO presence lower, cited brands captured more clicks (Seer).

Grounding, citation & locality

VENDOR-DOCUMENTED Always grounded; links appear inside/alongside the overview, formats varying by query/device; overviews trigger selectively and their rate/length/format have changed repeatedly since launch — date-stamp any screenshot asset. Locality: runs inside Search’s market/language systems; local AIOs draw on Business Profile/local data.

INDEPENDENT ~11% of captured AIO responses lacked a visible citation in one Sept-2025 US sample (vs ~3% for AI Mode); trigger frequency swung widely across 2025 (Semrush 10M+ keyword tracking) — measure date-stamped, don’t claim a fixed coverage rate.

Ahrefs, 2025-12-15 ↗ (opens in a new tab)

Sources

Three steps to get started.

A couple of hours a month is enough to make real progress.

Enter your website.

We look up your business and the questions your customers actually ask AI.

Review what we found.

Confirm your business, your questions and your competitors. Then we ask the live apps your customers actually use: ChatGPT, Gemini, Google AI Overviews, AI Mode, Copilot and Perplexity. Not the developer API, and not the model's training data. Both give a different answer from the one your customer sees.

Get your free report.

Where you appear, who wins instead, and how AI describes you, in plain English with clear next steps. A one-time snapshot to start from.

See what AI says about you.

Get a free report: no sign-up, no credit card, results in minutes.  ·  See a sample report ›