What Triggers Google AI Overviews? Query Signals and Conditions Explained

A Google AI Overview is triggered when a query benefits from synthesis rather than a single retrieved answer, based on query intent, length and phrasing, and topic category. Ahrefs found 20.5% of 146 million SERPs carried an AI Overview. These trigger conditions are definable query signals, not randomness. Once triggered, Google decomposes the query and compresses several retrieved pages into one summary placed above the organic results. Those same signals can be checked for a keyword before anything is published, making trigger prediction the foundation of AI Overview Optimization.

The published figures below come from Ahrefs’ study of 146 million desktop SERPs collected in September 2025. Everything after them tests whether those figures transfer. Across a Bengali-language health corpus, a US SaaS corpus, a 20-query intent panel, 15 matched query pairs, and a three-arm local experiment run in Dhaka in August 2026, the baseline held in some places and broke in others. Where it broke, the direction and the size are reported here, along with the method that produced them.

What triggers a Google AI Overview?

A Google AI Overview triggers when a query benefits from synthesis rather than simple retrieval. Triggering is not random. Definable query signals drive the decision.

AI Overviews trigger when generative AI provides greater value than standard Search results.

Source: AI features in Search, Google Search Central

AI Overviews appear on 20.5% of all SERPs, 29,991,998 of 146,122,391 analyzed, and on about 21% of keywords. That equals roughly 1 in every 5 searches (Ahrefs, Ryan Law analysis of 146 million SERPs). Ahrefs drew these figures from 146,122,391 desktop SERPs collected in September 2025. Every first-party measurement on this page was taken in August 2026, eleven months later.

Even the baseline is contested. Ahrefs reports AI Overview presence at 20.5% across 146 million SERPs; Semrush’s tracking, as reported by Search Engine Land, puts it closer to 13% of all searches. Two large vendors measuring the same thing differ by more than half. Any single baseline is a starting point for your own measurement, not a substitute for it.

One structural reason to re-measure rather than import. Google rolled out AI Overviews on Gemini 3 globally on 27 January 2026, four months after the 146-million-SERP dataset was collected. Whether a query “benefits from synthesis” is a model judgment, so a change in model can shift the threshold without any change to the query. Every cross-niche figure quoted on this page predates that switch. The first-party measurements postdate it by seven months.

All bars are trigger rates. Composition shares are reported separately in the text. Built from published figures in Ahrefs (146,122,391 SERPs) and Whitespark (local queries); baseline 20.5%.

Five trigger signal classes shape the outcome:

• Search intent — the strongest single signal — with 99.9% of AIO keywords informational.
• Query length and phrasing: question format and word count, from 9.5% of single-word queries to 46.4% of queries with seven or more words, and 57.9% of question queries.
• Topic category: the subject area of the query, at 43.6% Science and 43.0% Health against 3.2% Shopping.
• Suppressors: query types that reduce triggering, at 7.9% for local, 6% for news, 4% for NSFW.
• Volatility: how newsy or evergreen the query is.

An AI Overview is triggered when a query is complex enough to require multi-source synthesis (combining multiple sources into a single readable summary). Most AI Overviews draw from four to eight source pages and run a few hundred words.

Google AI Mode replaces the triggered AI Overview with a back-and-forth conversational experience. The two surfaces select sources differently; this is the boundary this page stops at: see AI Overview vs. AI Mode. AI Overview Optimization treats these signals as inputs for Zero-Click Search visibility, and Generative Engine Optimization builds on the same trigger logic.

Across 142 keywords that sustho.com ranks for in Bangladesh in Bengali, pulled from DataForSEO’s SERP index with snapshots refreshed between June and August 2026, an AI Overview was present on 134 — 94.4% — compared with Ahrefs’ 20.5% all-SERP baseline. The set is a domain’s own ranked-keyword footprint rather than a curated tracking list, so it skews long-tail and informational by construction; that skew is the point. A baseline drawn across 146 million mostly English SERPs says almost nothing about what a Bengali health publisher will actually face.

Across 449 keywords with 2,000 or more monthly searches from a Gmail mail-merge SaaS’s US ranked-keyword footprint, an AI Overview was present on 416 — 92.6% against Ahrefs’ 20.5% all-SERP baseline. That volume threshold is the filter: the domain’s unfiltered US footprint is far larger, and the 92.6% figure describes the filtered slice, not the whole footprint.

The gap tracks the set composition rather than any disagreement with the measurement. Of that set, 99.8% was non-branded, and 59.2% was question-shaped: the two traits the same dataset identifies as raising the rate. The single branded query in the set, the brand name itself, with 6,600 monthly searches, returned a knowledge panel but no AI Overview.

That is what a category average cannot carry. A rate drawn across 146 million SERPs describes the mix inside those SERPs, and any real keyword set is a different mix.

A study by Wang and Hagar at Generative AI Newsroom points to the mechanism behind that. Across 30 sites and 30,000 keywords, they found that the more sites a keyword appears across, the less likely it is to trigger an AI Overview: broad, widely covered terms trigger less; narrow, specific ones trigger more. Their site-category split runs 20.9% for informational sites, 10.5% for news, and 8.0% for popular-culture sites.

That explains the two high-rate corpora measured here rather than explaining them away. Both are long-tail footprints of specific, low-competition phrasings. Specificity predicts a high rate; the footprints deliver one. Building a footprint of that shape is a coverage problem before it is a trigger problem, and the method is on topical authority.

Intent is the dominant trigger. The next question follows from the classifier work behind which query intents trigger AI Overviews.

Which query intents trigger AI Overviews?

Informational intent triggers AI Overviews the most, with commercial, transactional and navigational intent trailing far behind. Four intent classes clear the trigger bar at sharply different rates:

• Informational intent: 99.9% of all AIO-triggering keywords fall here.
• Commercial intent: 5.5% of AIO keywords.
• Transactional intent: 1.2% of AIO keywords.
• Navigational intent: 0.1% of AIO keywords.

These values come from Ahrefs’ analysis of 146 million SERPs. Ahrefs classified the same keywords against the four intent categories Google names in its Search Quality Evaluator Guidelines. In that mapping, 99.9% of AI-Overview-triggering keywords fall into the Know category, while Website queries account for 0.9% and Visit-in-person queries for 7.1%. If you need the category definitions before applying them, you can find them at search intent.

Non-branded queries trigger AI Overviews 1.9× more often than branded ones: 24.9% against 13.1%. Ahrefs also reports that 76.3% of AIO-triggering keywords are non-branded, compared with 23.7% branded. Semrush’s tracking from October 2024 through May 2025, as reported by Search Engine Land, shows the share of keywords triggering an AI Overview rising across commercial, transactional and navigational intent.

Ahrefs publishes two intent tables, and they tell different stories. The composition table shows commercial at 5.5% of triggering keywords and transactional at 1.2%, which suggests that commercial triggers more often. The rate table gives commercial 4.3% and transactional 2.1%, which says the same thing more directly: across their 146 million SERPs, commercial keywords triggered about twice as often as transactional ones.

On the SaaS set measured here, the ordering reverses. Across 366 transactional keywords, 319 (87%) carried an AI Overview. Across 159 commercial keywords, 73 did, 46%. Transactional was triggered nearly twice as often as commercial, the opposite of the cross-niche result. The classification is DataForSEO Labs’ single-label main-intent field, read from SERP snapshots refreshed through August 2026.

Two explanations fit. The intent labels may not mean the same thing at the corpus level as they do inside one product’s keyword footprint, where “transactional” is dominated by how-to-do-a-task phrasings rather than buy-now phrasings. Or the vertical genuinely differs. Either way, the cross-niche rate is the wrong number to plan a SaaS content calendar against, and the composition figure is the wrong number twice over.

Intent decides most triggers, but length adjusts the odds. Does a longer query raise the chance of an AI Overview?

Does query length affect whether an AI Overview triggers?

Yes — query length affects whether a Google AI Overview triggers, and the effect is directional. Query length increases the likelihood of an AI Overview trigger. AI Overviews appear on 9.5% of single-word queries and on 46.4% of queries with seven or more words, per Ahrefs’ analysis of 146 million SERPs.

Question-shaped keywords make up 33.5% of AI-Overview-triggering keywords, not the majority. What matters more is the rate: 57.9% of question queries trigger an AI Overview, compared with 15.5% of non-question queries, a 3.7× gap. Question form is a strong predictor, even though questions make up a minority of triggering keywords. The complement is explicit: 42.1% of question queries fail to trigger.

The distinction is worth flagging because the published write-up of this dataset states the composition figure the other way round, giving 66.5% as the question share. The underlying table gives 33.5% for questions and 66.5% for non-questions. Composition and rate answer different questions, and here the two got crossed in the reporting.

The reason, definition, and bool classifiers come from Google query-classification labels surfaced by Mark Williams-Cook through a bug-bounty disclosure. Reason queries trigger at 59.8%, the highest rate in the set; bool follows at 57.4% and definition at 47.3%.

Testing 15 matched query pairs on 27 August 2026 — same topic, same informational intent, verified against DataForSEO’s classification — varying only length and question form, through DataForSEO’s live SERP endpoint at US desktop: fourteen of the fifteen pairs triggered an AI Overview on both the short and the long form. The exception ran in the direction Ahrefs’ gradient predicts: “eeat signals”, at two words, returned no AI Overview, while “what are eeat signals and how does Google measure them”, at ten, did.

Inside a topic already near saturation, length moved the outcome once in fifteen attempts. That is not zero, and it is not the 9.5%-to-46.4% spread the cross-niche data shows either. The length effect persists at the edges of a saturated vertical and disappears in its middle. Where the base rate is already high, the open problem is citation, not activation.

Longer, more specific queries raise the stakes for AI Overview Optimization because they include phrasing that turns a search into Zero-Click Search visibility and shapes Generative Engine Optimization results.

Google’s query intent classification ties phrasing to trigger odds. Length and question form still leave a gap, and topic category fills it.

Which topic categories cross the AI Overview trigger threshold?

Topic categories that cross the AI Overview trigger threshold are led by Science and Medical/Health queries, while Shopping, Real Estate and News queries stay well below the 20.5% baseline. Google does not summarize every subject equally: the trigger rate varies by topic category (Ahrefs, 146 million SERP analysis, 20.5% baseline). Ahrefs measured AI Overview share across named topic categories, benchmarked against the 20.5% all-SERP baseline.

YMYL (Your Money or Your Life) status raises the odds. AI Overview triggers on 34.3% of YMYL queries versus 17.2% of non-YMYL queries, and 32.3% of all AIO-triggering keywords are YMYL.

Medical YMYL crosses the highest at 44.1%. Safety YMYL follows at 31%, Legal YMYL at 23.6%, and Financial YMYL at 22.9%. The spread matters: a health page sits in the category Google summarizes most readily, a finance page in one it summarizes half as often.

Measured against a Bengali-language health set on 27 August 2026, 142 keywords that sustho.com ranks for in Bangladesh had an AI Overview on 134 (94.4%), compared with Ahrefs’ 43.0% for the Health category. No English-language dataset reports a Bengali figure, and the direction is worth naming: the concern with a non-English market is usually that AI Overviews have not rolled out. Here, the rate runs more than double the English-language category figure.

Two things sit inside that number and cannot be separated. Health queries are YMYL, and the same dataset puts them at 34.3% vs. 17.2% for non-YMYL, so category and YMYL effects are tangled. And the set is a domain’s own ranked-keyword footprint, which, by construction, skews toward long-tail and informational.

News at 15.1%, Real Estate at 5.8%, and Shopping at 3.2% sit below baseline, evidence that some query attributes actively suppress the AI Overview.

What query attributes suppress an AI Overview?

Five query attributes suppress an AI Overview: restricted and harmful topics, breaking-news freshness, local-pack-dominant intent, pure navigation, and tool intent. A near-zero trigger rate on these categories is an active suppression layer, not the absence of a trigger. Each attribute maps to Google’s query intent classification, where a query’s intent decides whether synthesis is allowed at all.

• Restricted and harmful topics. AI Overview is suppressed on NSFW, drug, and scam-related queries. In the Ahrefs 146-million-SERP analysis, 4% of NSFW queries trigger an AI Overview, and 99.1% of AIO-triggering queries are classified as not NSFW; within that category, drug queries trigger at 12.6% and scam queries at 11.3%.
• Breaking-news freshness. AI Overview is largely suppressed on very newsy, fast-moving queries. Only 6% of news queries trigger an AI Overview. Very newsy queries trigger at 6.3% and make up 0.07% of triggering keywords, against a News category share of 15.1%, so suppression concentrates in the very-newsy subset.
• Local-pack-dominant intent. AI Overview rarely triggers for pure local intent (near-me intent, where the user wants a nearby business), but often for advisory local queries. Local search queries trigger at 7.9%, compared with 22.8% for non-local queries. Whitespark’s local-search research shows pure “service + location” queries at 15%, business-type queries at 68%, and educationally framed local queries at 92%.
• Pure navigation. AI Overview is suppressed when a query targets a single destination. Navigational queries make up 0.1% of AIO keywords, Website-query rate is 0.9%, and Visit-in-person queries account for 7.1%. AI Overview is often skipped when a Know Simple query has a single, unambiguous factual answer, since Google determines that the query does not benefit from synthesis.
• Tool intent. Tool-intent queries return the tool, not a synthesis. This is the one suppressor that showed up in both corpora measured here.

In the Bengali health set, both non-triggering keywords were BMI calculator queries. In the US SaaS set, the misses were an AI tone generator, a free email writer, and a business-account setup query that returned a featured snippet in place of an AI Overview.

Two languages, two markets, one shape: where the searcher wants an instrument rather than an explanation, Google supplies the instrument. Ahrefs’ cross-niche data points in the same way, finding citations that are absent from simple calculations.

CPC does not predict AI Overview triggering. Ahrefs found no consistent gradient across cost bands: keywords above $20 CPC triggered at 15.8% and keywords between $0.20 and $0.50 at 12.4%, both below the 20.5% baseline but not ordered by price. Commercial and Shopping keywords trigger less often than the baseline, but that is an intent and category effect, not a bid-value effect.

The local suppression set was 30 matched service topics tested in the Dhaka market on 27 August 2026, run in three arms. Service-plus-location queries triggered an AI Overview in 2 of 30 cases (6.7%), compared with 7.9% in English-language data. The advisory-framed version of the same services triggered on 25 of 28, or 89.3%. Two of the thirty topics returned a server error on the advisory arm and are excluded from that denominator rather than counted as failures.

A third arm separates the cause. Run without any location at all, the same services triggered on 4 of 30, or 13.3%. Adding the location roughly halved that. Adding advisory phrasing, while keeping the location, multiplied it by more than twelvefold. The suppression attaches to the phrasing, not the locality.

A local pack appeared on 27 of the 30 location queries, so this is not a case of Google holding thin local data for the market. The pack is there, and where it is, the AI Overview mostly is not.

Do ads and rival SERP features affect triggering?

Ads and rival SERP features do not change the likelihood of a trigger; they change display placement. A crowded results page does not crowd out an AI Overview from being generated, because the trigger decision happens before layout. SERP features act as a display constraint, not a trigger rule.

AI Overview shares the search results page with four other SERP features:

• Local pack (map-based local business results) occupies vertical space above AI Overview.
• People Also Ask sits below or beside AI Overview.
• Organic listings run under AI Overview.
• Ads take the top slot ahead of AI Overview.

Identical queries still show AI Overviews inconsistently even when SERP features and ads match, which layout competition alone cannot explain.

Why does the same query trigger an AI Overview inconsistently?

The same query triggers an AI Overview inconsistently because per-session volatility factors change the result between sessions and over time. Two people searching for the same words see different SERPs, so Google adjusts AI Overview generation based on per-session signals.

• Personalization signals (account-level history feeding the result).
• Geography and location (the market where the query is issued).
• Content freshness (source pages Google re-selects over time).
• Staged rollout tests (Google enabling AI Overview generation for a fraction of users at a time).

Of these four, the measurement below covers freshness and rollout; personalization and geography would need a different design.

Pulling every SERP snapshot DataForSEO holds between February and August 2026 for fifteen US SEO keywords, twelve had more than one snapshot, with a mean of 3.1 each. Two returned no history at all, and one returned a single snapshot.

The AI Overview state changed on 2 of those 12. On “keyword cannibalization,” an AI Overview was present on 17 March, absent on 21 May, and present again on 3 July. On “content decay,” it was absent in February and present at every snapshot after: a switch that stayed switched, not a flicker. Over months, one query in twelve genuinely oscillated.

A parallel run against fifteen Dhaka-market queries failed on a malformed location parameter and returned no data. Whether third-party SERP history exists for that market remains an open question, not a finding.

The volatility set was measured through a depersonalized API, so these are volatility and index freshness, not account personalization. At a mean of 3.1 snapshots across seven months, any change that happened and reverted between two crawls is invisible here. Whether the same query flickers within a single day is a different measurement, and this is not it.

Two user-side variables are the most testable: login status and geography.

Does login status change AI Overview triggering?

Login status changes AI Overview eligibility, but account state is an experiment gate rather than a ranking factor. Three personalization variables influence eligibility: login state, search history, and device (mobile or desktop). Account state and device context can shift whether a Google AI Overview renders, without changing how any single source page ranks. Personalization and login state were not varied in these tests. A signed-in user on mobile may see a Google AI Overview that a signed-out user on desktop does not, so the surfaced overview is an account-conditioned view of the results page. AI Overview triggering also depends on the market and language where the query is issued.

How does Google generate an AI Overview once a query triggers one?

Once a query is triggered, generation is a separate process with its own signals, covered in full on query fan-out. In outline: Google fans the query into sub-queries, retrieves passages against each, and synthesizes. This page stops at the trigger decision.

The generative synthesis draws from four to eight source pages and produces one overview a few hundred words long. Passage ranking (matching individual passages to the query) selects the specific text spans that feed this compression. For the retrieval mechanics behind that step, see RAG. Once created, the AI Overview shifts position by query type: it appears above organic results and above the local pack for local queries.

Because the summary answers the query on the page, the Google AI Overview drives Zero-Click Search behavior, where searchers read the overview and never click through. AI Overview Optimization and Generative Engine Optimization both target this generative synthesis, so a page becomes one of the four to eight sources. If you need the click-loss numbers rather than the mechanism, what this does to click behavior is measured on the AI Overviews and Zero-Click Search pages.

A keyword can be identified as trigger-likely before any query runs, raising the question of how to interpret that signal in advance.

How do you tell if a keyword will trigger an AI Overview?

Four checks, in order, each with a threshold you can apply before writing.

1. Classify the intent. Informational triggers at 21.4%, commercial at 4.3%, transactional at 2.1%, navigational at 0.9% in cross-niche data. Any SERP tool that exposes an intent label will do this; DataForSEO returns it as main_intent on ranked-keyword calls. Anything outside information starts below baseline, and you should verify against your own corpus rather than trusting the label, because on the SaaS set measured here, the transactional and commercial ordering is reversed.

2. Read the phrasing. Question form is the single cheapest signal: 57.9% against 15.5%. Reason phrasings (“why”, “how come”) reach 59.8%. Word count moves with it, 9.5% at one word to 46.4% at seven or more. Check: Does the keyword ask something, and does it take more than six words to ask it?

3. Place the topic. Science 43.6%, Health 43.0%, Pets 36.8% sit well above baseline; News 15.1%, Real Estate 5.8%, Shopping 3.2% sit below. Medical YMYL runs highest at 44.1%. Check: Is your category above or below 20.5%?

4. Test for suppression, then re-test for volatility. Run the query through the five suppressors: restricted topic, breaking-news freshness, local-pack intent, pure navigation, tool intent. Then pull the SERP history. In the fifteen keywords tracked here across seven months, twelve had more than one snapshot, and the AI Overview state changed on two: one oscillated, one switched and stayed. A single check on a single day is not a reading. For the tools that record this per keyword at scale, see AI visibility tracking.

The rate at which these predictions hold depends on your corpus, not on the published averages. Both corpora measured here ran near 90% against a 20.5% baseline, so the checks above set a floor, not an expectation.

Where the numbers on this page come from

Two sources, kept separate throughout.

The cross-niche figures come from Ahrefs’ analysis of 146,122,391 desktop SERPs collected in September 2025, published in November 2025. They classified 86 factors across intent, phrasing, topic, and volatility, comparing each category’s rate against a 20.5% dataset baseline. Where this page quotes a rate, it is their rate; where it quotes a share, it says so.

The first-party figures were measured in August 2026 from desktop Chrome, logged out. Five tests:

What limits these numbers

The corpora are footprints, not samples. A domain’s ranked keywords are selected by that domain’s content, and they cluster: in the Bengali set, six near-duplicate phrasings of one condition share a single SERP and a single ranking URL. The effective number of independent observations is well below the keyword count.

The AI Overview flag has two provenances. The footprint figures are read from DataForSEO’s SERP index, where each keyword carries its own crawl date. The panel, pair, and local tests read it from live SERPs on a single day. Both returned rates are in the same range, which corroborates the methods rather than representing a single measurement repeated.

Detection is not perfect. A SERP tool’s reading of whether an AI Overview rendered can differ from what a browser shows, and AI Overviews are not stable within a query: the volatility test found the state changing on 2 of 12 keywords across six months. The cross-niche rates have their own limitations: they are category averages against a single 20.5% baseline, and each snapshot is point-in-time.

Two runs failed. Two of the thirty local topics returned a server error on the advisory arm and are excluded from that denominator. A parallel volatility run against Dhaka-market queries failed on a malformed location parameter and produced no data.

The samples are small, and the geography is narrow. Bangladesh and the United States, desktop only, one language pair. Nothing here estimates a global rate.

Everything above is about whether a query produces an AI Overview at all. The questions below sit just outside that: they come up whenever triggering is discussed, but each belongs to a different part of AI Overview Optimization and links to the page that covers it.

Related questions about AI Overview triggering

Can I stop an AI Overview from using my page?

Yes, with the same controls that govern snippets. nosnippet, max-snippet and data-nosnippet remove or limit the text Google can lift, and they apply to AI Overviews as they do to featured snippets. There is no AI-Overview-only opt-out that keeps you in classic results. For the crawl and index conditions that sit behind those controls, the full mechanics are on AI Overview eligibility.

Does an AI Overview replace the featured snippet?

Not always, and the two can appear on the same SERP. In the trigger test here, “set up business Gmail account” returned a featured snippet and no AI Overview, so the surfaces are selected independently rather than one displacing the other.

Does triggering differ on mobile?

The published data does not say. Ahrefs’ 146-million-SERP dataset is desktop-only, and every measurement on this page was taken on desktop Chrome. Mobile trigger rates are unmeasured ground.

What to do with this

Trigger probability is a property of the query, and it is checkable before you write a word. Classify the intent, read the phrasing, place the topic, then re-test after a few weeks because the state moves. The cross-niche baselines are a starting point and nothing more: in the two corpora measured here, the rate ran four times the published average, and the exceptions clustered on tool-intent queries rather than scattering at random. Measure your own set. The checklist for what to do once a query triggers is on how to optimize for AI Overviews.