Almost every AI-search statistic in circulation describes a US search. Whether Google even shows an AI Overview, and how many sources it cites when it does, looks very different from one country to the next. We measured it across 15 markets and the six AI engines cloro monitors, over a 30-day window. AI search is far less uniform than the "48% of queries" headlines suggest, and "cited by AI" means something completely different depending on which engine you ask.
One correction since publication, in the methodology below and worth knowing before the numbers: each market here runs its own prompts, so the gap between two countries mixes Google's behaviour with the questions asked in each. A controlled panel puts most of the apparent gap down to the questions. Treat the country spreads as upper bounds. The engine comparison is unaffected.
Published July 5, 2026 · Updated July 28, 2026 · cloro monitoring corpus · 30-day window · ~61,000 sampled AI Overview results
67% – 93%
Among the countries where Google shows AI Overviews, the measured trigger rate ranges from 67% (Germany) to 93% (US, Austria). Each market runs its own prompts, so part of that 26-point spread is the questions rather than Google, and a controlled panel suggests most of it is. A US figure is still not your market's figure.
~12 vs ~1
Citation depth is an engine trait, not a constant: ChatGPT backs an answer with ~12 sources; Gemini often cites just one. A single "AI citation" is not a comparable unit across engines.
Austria vs Germany
Same language, very different rates: Austria triggers an AI Overview 93% of the time in our data, Germany 67%. Whatever separates them is not language. We cannot yet say how much is Google and how much is the different prompts each market runs.
France: 0% → 70%
When we first published, France returned zero AI Overviews, because it sat outside Google's launched markets. Google enabled them on 25 July 2026 and the rate hit 69.6% within a day on the same panel. A 0% market is a rollout boundary, not a permanent state.
"AI Overviews now appear on roughly half of Google searches" is the stat everyone repeats. It's a US number, and it hides the real story: among the markets where Google has switched AI Overviews on, how often they actually fire ranges from 93% (United States, Austria, Japan, Canada) down to 67% (Germany). The English-speaking markets cluster tightly at the top — US, Canada, Australia, and the UK all sit at 76% or above — while continental Europe is where the rate falls off.
The sharpest illustration is the German-speaking bloc. Austria (93%), Switzerland (82%), and Germany (67%) share a language but span a 26-point range. Whatever drives the difference, it isn't linguistic, so a brand that measures its AI visibility in one market and assumes the rest of the region looks similar is measuring the wrong thing. One German-language tracking setup covers one of these three markets.
Report AI visibility per country: a regional average built from these numbers puts Austria and Germany in one cell while their searchers see different pages.
Update — 28 July 2026: France switched on
Google now lists France and French in its availability documentation, and we caught the change in flight. On 25 July 2026 AI Overviews began appearing on French searches for the first time. Across the three complete days since (25–27 July), 69.6% of 730 sampled French searches returned an AI Overview, citing 8.3 sources on average.
France's row throughout this study — trigger rate, chart, and citation matrix — is that post-launch window. Every other market is the original 30-day window. We've kept the two apart rather than averaging them, because a rate measured across the switch would describe neither the old France nor the new one.
When this study first published, France was the one market in the sample that returned zero Google AI Overviews — and that wasn't an artifact of our corpus. Google did not offer the feature in France at all: its availability documentation listed the launched countries and languages, and France wasn't among them, consistent with the French and EU regulatory environment around Google's use of publisher content. The 0% was a documented rollout boundary, not a measurement gap.
Then it changed, and it changed all at once. French AI Overviews went from nothing to 67.7% on their first day and have held near 70% since, with no gradual regional ramp in between. France now sits mid-table, between the UK (76%) and Poland (68%), in the same continental-European band as the markets it was excluded from weeks earlier.
The exclusion is still the more useful lesson, because it generalizes. Throughout it the non-Google engines answered French queries normally — ChatGPT cites ~13 sources per answer, Perplexity ~10, and Copilot ~5. "AI search in France" was a live surface long before Google's arrival; it just wasn't Google's. Any team that had written France off for want of AI Overviews spent that time invisible on the engines that were answering all along.
Take the same brand cited twice on the same day. A Gemini answer in Germany names 1.0 sources, so the brand is the answer. A ChatGPT answer in Austria names 13.6, so it is one line in a dozen. Both land in a dashboard as "1 citation."
Citation depth is a property of the engine, and for some engines it swings hard by market. Plotting three engines across all 15 markets (sorted by AI Overview trigger rate) shows the shape at a glance: ChatGPT holds a steady line, Perplexity runs almost flat, and Gemini collapses outside the English-speaking markets.
| Country | ChatGPT | Perplexity | AI Overview | AI Mode | Copilot | Gemini |
|---|---|---|---|---|---|---|
| Austria | 13.6 | 9.6 | 13.6 | 11.3 | 6.5 | 5.1 |
| United States | 12.5 | 9.8 | 10.2 | 12.2 | 5.0 | 6.6 |
| Japan | 12.3 | 9.6 | 8.0 | 7.9 | 4.2 | 1.0 |
| Canada | 12.0 | 9.5 | 12.9 | 17.8 | 4.4 | 4.8 |
| UAE | 12.6 | 9.4 | 7.0 | 5.0 | 4.2 | 1.9 |
| Brazil | 11.6 | 9.6 | 9.5 | 9.2 | — | 0.6 |
| Australia | 13.0 | 9.6 | 9.8 | 14.0 | 4.9 | 6.8 |
| Israel | 10.9 | 9.5 | 7.4 | 5.4 | 4.6 | 0.8 |
| Mexico | 12.3 | 9.6 | 10.7 | 10.1 | 3.3 | 4.7 |
| Switzerland | 12.4 | 9.3 | 12.7 | 10.0 | 6.7 | 2.3 |
| Denmark | 11.6 | 9.2 | 8.3 | 5.7 | 4.5 | 1.1 |
| United Kingdom | 11.4 | 9.3 | 10.3 | 13.3 | 5.0 | 6.7 |
| France | 12.7 | 10.2 | 8.3 | 14.4 | 5.1 | 0.0 |
| Poland | 11.7 | 9.3 | 9.6 | 6.1 | 4.5 | 1.0 |
| Germany | 11.6 | 9.5 | 11.2 | 8.8 | 4.9 | 1.0 |
Three patterns fall out of it:
The practical consequence: being "cited by AI" is not one achievement. Earning a citation where an answer draws on a single source (Gemini in Germany) is worth far more per slot than being one of a dozen (ChatGPT in Austria). Comparing raw citation counts across engines, without normalizing for how many sources each one pulls, produces a misleading scoreboard.
One market's AI visibility tells you almost nothing about the next, not even within a shared language. Measure AI Overview presence and citations per target country, and don't extrapolate a US baseline onto Europe.
Normalize by engine. A citation is 1-of-~1 in Gemini-Germany and 1-of-~12 in ChatGPT-Austria; those are not the same win. Weight citation share by each engine's sourcing depth in the market you care about, or your dashboard will over-reward the engines that simply cite more.
Don't hardcode "AI Overview always present" or "never present." It fires only two-thirds of the time in Germany, and France went from a hard 0% to ~70% on a single day, so any rule keyed to the country code was wrong within 24 hours. Detect the block per response. Google is not all of AI search: a France pipeline watching only Google reported an empty market that ChatGPT and Perplexity were answering in all along.
Figures come from cloro's own monitoring corpus, queried over a 30-day window (~61,000 sampled AI Overview results across the 15 markets). Trigger rates are measured on Google's AI Overview surface; only countries with more than 200 sampled results are reported. Citation counts are the mean number of cited sources per answer, per engine, per market. A "—" marks a surface that returned no result in that market during the window.
The corpus is customer AI-visibility monitoring: conversational, branded and multilingual prompts. An earlier version of this page told you to read the absolute rates as directional and the cross-market comparisons as the finding, on the grounds that relative differences are robust to prompt mix. That was an assumption and we have since checked it. It does not hold here.
Almost every prompt in this corpus runs in exactly one country. In the window behind this page, 2,096 of 2,100 distinct prompts were measured in a single market. So a market's rate reflects both what Google does there and what the customers tracking that market happen to ask, and the difference between two markets cannot be assigned to Google alone.
We can size the problem rather than just flag it. Four prompts do run identically across 25 markets, a small controlled panel. On those, AI Overview presence spans 91.2% to 98.4%, a seven-point spread, against the thirty-five points seen when each market runs its own questions. Holding the questions still closes most of the gap. Read every market-versus-market figure on this page as an upper bound on the real difference, and see the SERP feature census for the same correction applied to Google's other page features.
A dedicated study has since put a number on it. SERP Features by Country runs one fixed 150-query basket across six markets with both country and interface language pinned. Roughly half the apparent difference between markets survives control and half does not, the market ordering changes, and query archetype moves AI Overview presence across an 88-point range against 18.7 points for the market. That is the correction to apply to the country figures on this page.
The engine comparison is not affected. ChatGPT against Gemini on citation depth holds the prompts constant by construction, because every engine answers the same prompt set. Only the country rows carry the confound.
France is measured on a later window than the other 14 markets, and every France figure on this page reflects it. At publication France returned a flat 0% — Google did not offer AI Overviews there. Google enabled them on 25 July 2026, so France's row was re-measured over 25–27 July 2026 (730 sampled AI Overview results, 269 prompts, same panel and same surface as before) and updated here. The two windows are reported side by side rather than blended: a rate averaged across the switch would describe neither period. Read France's row as current and the other 14 as the original window until the next full-window refresh brings them back onto one basis.
cloro. (July 5, 2026). AI Overviews Around the World: Trigger Rates. cloro Research. https://cloro.dev/research/ai-overviews-around-the-world/
More studies from cloro's monitoring corpus are in the research index.
cloro monitors AI Overviews, ChatGPT, Perplexity, Gemini, and Copilot by country and query, live. Start with 500 free credits a month.
