Google AI Overview: what it is and when it triggers

The AI Overview is the answer box Google displays above classic results, generated by artificial intelligence. It is now one of the first things a user sees. One question remains poorly documented: does it always trigger?

Repliq, a Swiss solution for measuring brand visibility in answer engines, put over 2,300 real questions to Google. The observed trigger rate is 73.4 %. And what matters is not the average, but what the average hides.

What an AI Overview actually is

The AI Overview is a text the engine produces at search time. It is not a page ranked first: it is a summary written for the occasion, drawing on several sources and linking to some of them.

The difference is fundamental. A classic result is a page that already exists and that Google orders. An AI Overview is an answer that did not exist before the question. A brand appears there not because it ranks well, but because the engine judged it relevant to cite while composing the answer.

Two neighbouring elements deserve distinction. The featured snippet, which Google has shown for years, extracts a passage from a single page without rewriting it. The AI Overview reformulates and combines several sources. And AI Mode is a separate conversational interface: it does not appear on the classic results page.

Since when, and why Switzerland has sixteen months of hindsight

The timeline explains much of what can be measured today.

Switzerland received AI Overviews sixteen months before France, and in all four national languages simultaneously. The French delay stemmed from negotiations between Google and press publishers over neighbouring rights, covering the terms under which content may be used in generated answers.

That gap has a practical consequence: the Swiss market holds sixteen months of observation of the phenomenon, across four languages. The figures presented below come from that market, not from extrapolations of American data.

Where it appears and how to recognise it

The AI Overview sits at the top of the results page, before the links. Three elements identify it: a summary text in paragraphs or bullets, a set of source links to the right or below, and a notice flagging its experimental nature.

Screenshot of Google AI Overview, August 2026.

The source panel deserves attention, because that is where a brand's visibility is decided. Google shows a few links, generally three to six plus a total, which do not necessarily match the top classic results below.

The same search demonstrates it. In the AI Overview source panel, Comparis appears first. In the classic results for the same question, Comparis comes third, behind Priminfo and Sanitas.

Screenshot of Google, Web tab, August 2026.

The order of cited sources is therefore not the order of ranked results. These are two separate selections, and a good organic position does not guarantee citation.

On mobile, where most searches happen, the block often fills the entire first screen. The user must scroll to reach the first classic link, which explains much of the effect on clicks.

When it actually triggers

Across these questions, the AI Overview returns an answer in 73.4 % of cases. In the remaining 26.6 %, the search produces a results page with no summary block at all.

But the average is misleading. Segmented by question intent, triggering ranges from 84.1 % to 23.4 %.

Swiss Atlas, Repliq observatory. Over 2,300 questions put to Google AI Overview in French-speaking Switzerland, April to August 2026.

On price and purchase questions, the AI Overview triggers almost four times less. It is the sharpest finding of the measurement: at the moment the user approaches a purchase, Google largely stops answering on their behalf and defers to classic results.

A caution on reading this: 23.4 % does not mean never. Out of four purchase searches, one still produces an AI Overview. Anyone testing a single price question therefore has a one-in-four chance of seeing it appear, which does not contradict the measurement. That is the difference between an observed rate on a sample and an absolute rule.

A methodological note: this segment counts 94 measurements, the smallest of the five. The gap is too wide to be accidental, but the exact value warrants a margin.

The same question in three languages, three source ecosystems

Switzerland has four national languages. We asked the same question about Swiss health insurance on google.ch in Italian, French and German, on the same day.

No shared source in first position. In Italian a comparison site leads, in French a blog for expatriates, in German a fintech company from outside the insurance sector.

Even the block's name changes: AI Overview in Italian, Aperçu IA in French, Übersicht mit KI in German.

The consequence is direct for any company operating across several language regions. Measuring visibility in one language says nothing about the others. A brand well cited in German may be absent in Italian on the same question, and nothing in a monolingual measurement would flag it.

How it works: what the corpus reveals

Beyond the trigger rate, these questions allow the AI Overview's behaviour to be described, and compared with the three other engines tracked on the same corpus.

It cites fewer sources than the others

The AI Overview is the most frugal of the four. Six sources at the median, against nine for Perplexity. The space available to a brand is therefore structurally narrower. On the other hand it almost always names its sources: only 1.2 % of its answers carry none, whereas ChatGPT answers without a source in 18.7 % of cases.

It favours the article and avoids the homepage

Comparing the type of page cited by the AI Overview with that of the three other engines on the same corpus:

This is the most directly actionable finding of the measurement. A homepage is markedly less likely to be picked up by an AI Overview than by the other engines, while an editorial article or a practical guide is markedly more likely. Optimising a homepage for the AI Overview means working on the format it retains least.

A short format and a widely dispersed selection

An AI Overview runs to 250 words at the median, roughly 2,900 characters. It is a format of synthesis, not of exposition.

On the source side, the 10,759 citations recorded spread across 2,002 distinct domains. The top ten gather only 14.1 %, the top hundred 42.7 %. Reddit and YouTube lead the ranking, but at just 2.6 % and 2.0 % of citations: the figure illustrates dispersion rather than preference. For comparison, ChatGPT relies on Reddit more heavily (3.1 %) than the AI Overview does.

In other words, no small group of sites controls access. Space is won across a very large number of domains, not by targeting a handful of platforms.

Brand or category: a 13-point gap

A second useful segmentation, between searches naming a brand and searches describing a need.

When a user types a company name, Google produces a summary in more than eight cases out of ten. That summary describes the brand in words the brand did not choose, from third-party sources.

It is a point marketing departments often overlook: the first description of the company a prospect reads no longer comes from the company website, but from an answer the engine assembles out of what others wrote.

The thirteen-point gap has a plausible explanation. A brand search has a stable, verifiable answer the engine can summarise at low risk. A category search of the «which solution should I choose» type calls for a comparative judgement it takes on less willingly.

What changes in practice

Three consequences emerge from the data.

Visibility is no longer measured in positions alone. A company can rank well among links and be absent from the generated answer, or the reverse. These are two distinct measurements, and tracking only one gives a partial picture.

Third-party sources weigh more than the brand's own site. The AI Overview cites what the engine judges reliable, and rarely the brand site alone. Swiss Atlas measurements show comparison sites are the second most cited source category, at 16.1 % of citations.

Purchase intent remains classic territory. Since the AI Overview triggers in only 23.4 % of these searches, traditional search optimisation retains its full value there. A rarely stated piece of good news: the part of the journey closest to conversion is also the least affected.

How to appear in an AI Overview

There is neither a button nor a purchasable slot. Sources are chosen by the engine as it composes the answer. What is observable on regularly cited pages:

1. Short sections. Cited pages are structured in blocks of roughly 220 to 250 words, each answering one precise point. An engine extracts a passage, not a whole article.
2. Verifiable data. Figures, dates, amounts. A page without data is not cited on a question that calls for some.
3. One question per page. Pages covering a topic in general are cited less than those answering a precise question.
4. Third-party sources. Being picked up by a comparison site or the specialist press weighs more than a well-written internal page.

One clarification against a common assumption: there is no source oligopoly. In Swiss Atlas measurements, the hundred most cited domains gather only 34.3 % of citations. The remaining two thirds spread across thousands of sites.

This work of optimising for generated answers has a name, GEO (Generative Engine Optimization).

How to turn the AI Overview on or off

A frequent question with a simple answer: the AI Overview is not a feature users switch on. Google shows it when it judges it relevant, which explains the 26.6 % of searches without a block.

There is no official option to disable it permanently. You can, however, get classic links only via the «Web» tab in the filter bar beneath the search box. That tab shows results with no generated block.

For anyone measuring visibility, this tab is also a control instrument: it allows comparing what the generated answer says with what the classic results say on the same question.

Method and limits of interpretation(4)

1. French-speaking Swiss perimeter. These questions come from the Swiss Atlas, Repliq's longitudinal observatory, across ten sectors of the French-speaking Swiss market. Questions are asked in French. These results do not hold for other markets or for Switzerland as a whole.
2. Definition of triggering. A measurement counts as triggered when the search returns an AI Overview with non-empty text. Empty responses count as not triggered.
3. Intent segmentation. Deterministic rule applied to the question text, identical to the one used in production. The purchase/price segment counts 94 measurements, the smallest.
4. Window. From 20 April to 6 August 2026, excluding the technical ramp-up week.

In summary

The AI Overview triggers on roughly three searches out of four, but on only one in four when the user approaches a purchase. It describes brands in more than eight brand searches out of ten, in words the company did not choose. And the sources it draws on lie largely outside the brand's own site.

Measuring presence in these answers has become an exercise distinct from position tracking. That is the subject of AI visibility measurement.

These figures are not a snapshot. The Swiss Atlas is updated monthly, and the results by sector and by source are public: observatory of French-speaking Swiss sources.

Measure my AI visibility→

FAQ

What is Google's AI Overview? The artificial-intelligence answer Google displays above classic links, produced at search time from several sources.

When does it trigger? In 73.4 % of cases across over 2,300 Swiss questions. From 84.1 % on practical questions to 23.4 % on purchase intent.

How do you turn it off? There is no permanent option. The «Web» tab in Google's filter bar shows results without a generated block.

Does it cost traffic? It can reduce clicks, but shifts visibility into the answer itself. The two are measured separately.

How do you appear in one? Short sections, verifiable data, one question per page, and pickup by third-party sources.

---

Read next:

• What is GEO?
• Share of Model: definition, calculation and reliability
• The 3 essential GEO KPIs