AI Answers Reached Your Export Markets on Different Dates

On 22 July 2026 Google's French blog carried a brief announcement: Aperçus IA and Mode IA, the French names for AI Overviews and AI Mode, were being rolled out in France from that day, on mobile, on desktop and in the Google app. The same post mentioned a new Search Console setting, available in France, letting site owners decide whether their pages appeared in either. As corporate prose goes, it was unremarkable. It also closed a gap that had been open, in one form or another, for the best part of two years.

Consider what that meant the following morning. A buyer in Lyon searching in French for a supplier of PTFE-lined hose assemblies might now be shown a generated paragraph above the links, with its sources attached, or might switch to a conversational mode and describe the application in full. A counterpart in Manchester had been able to see the summaries since the late summer of 2024; one in Bilbao or Bologna, since the spring of 2025. None of these buyers is a real person. They are a device for illustrating a timeline. The timeline, however, is entirely real, and for anyone selling the same product into several of those countries it matters rather more than it first appears.

Five announcements, two years apart

Taken from Google's own announcements, the sequence for the markets a European exporter is most likely to care about runs as follows.

• United Kingdom. AI Overviews began rolling out on 15 August 2024, one of six countries added that day, with the expansion phased over several weeks. AI Mode followed on 28 July 2025, as a tab on the results page and in the Google app.
• Germany, Italy and Spain. AI Overviews arrived in a European expansion announced on 25 March 2025, in German, Italian and Spanish as well as English, for signed-in users aged 18 and over. AI Mode reached all three in the expansion Google announced on 7 October 2025, which its German, Italian and Spanish blogs covered the following day.
• France. Both features on 22 July 2026, at once, with nothing in between.

Put as intervals, a British buyer had AI Overviews for roughly seven months before a Spanish one, and for very nearly two years before a French one. AI Mode opened in Britain about ten weeks ahead of Germany, Italy and Spain, and a full year ahead of France. These are start dates, not completion dates. Several of the announcements describe a rollout rather than a switch, so the day a particular person first saw the feature could be later still.

There is a further wrinkle the dates alone do not capture. Availability runs by country and by language together, and the two are not always in step. The March 2025 expansion brought AI Overviews to Switzerland in French, German and Italian, some sixteen months before the same feature reached French speakers on the other side of the border. A language being supported somewhere is not the same as its being supported in the market you actually sell into.

What the lag does, and what it does not

It would be tidy to conclude that British buyers therefore behave differently from French ones: that two years of summaries have taught them to trust generated answers, or to skip the links. Nobody outside Google holds the evidence to say that with any confidence, and this piece will not pretend otherwise. What can be said is narrower, and sturdier.

Exposure is a precondition of habit. A buyer cannot have formed a view about a feature they have never been shown. So in July 2026 the set of buyers who had ever used AI Mode to draw up a supplier shortlist was, by construction, a different set in France from the one in Britain, where the option had existed for a year. Whatever those buyers do with the feature, and it may well be very little, they have had different lengths of time in which to do it.

The second point concerns the web rather than the buyer. When a generative feature appears in a market, it is reasonable to expect the people who publish in that language, trade titles, distributors, industry associations, to notice it and adjust, and some of them to opt out. The Search Console setting that arrived in France on launch day is the obvious instrument for that. The pages available to a summary written in French in late 2026 therefore come from a market that has only just begun adjusting, while the English-language web in Britain has been at it for considerably longer. The direction of the effect is not something anyone can state in general. The difference in elapsed time is not in dispute.

The assistants are not uniform either

Google's search features make the cleanest example because the announcements carry dates. The same unevenness runs through the assistants themselves, and it runs along two axes at once: where the user is, and what the user pays.

Gemini is the best documented case. Google's Gemini Apps help page on limits and upgrades sets out usage limits and context windows that rise with each paid tier, lists features reserved for subscribers, shows that the paid plans are not sold in every country, and sets a higher minimum age for them in the European Economic Area, Switzerland and the United Kingdom. Two buyers putting the same question to Gemini may therefore be using materially different products, depending on plan and country, before either has typed a word.

The other assistants publish their own availability and plan pages, and those change often enough that any summary written here would be stale before it was read. The principle survives without the detail. A buyer's experience of an assistant depends on account tier and market as well as on the underlying model, and a measurement taken through an API sees none of that variation. That is a limit of every method of this kind, ours included, and it is better stated than discovered.

A result in one market is not a forecast for another

The tempting inference for an exporter runs like this. We are named reliably in English; the British market has had these tools longest; Britain therefore shows us where France, Germany and Spain are heading, and we can plan accordingly. It is an attractive idea, because it turns one market into a crystal ball. It does not survive much inspection.

What an assistant says about a supplier in a given language depends heavily on what has been written about that supplier in that language: documentation, distributor pages, trade press, forum threads, comparisons written by people with no particular reason to be kind. Those bodies of text are built separately, market by market. A strong English-language presence does not seep into the German web with the passage of time. Somebody has to write the German material, and in most exporting companies that somebody reports to a different manager, often in a different building, occasionally in a different company altogether.

Timing then adds a second layer on top. Even if two markets were identical in what had been written, they would not be at the same point in their exposure to generated answers, nor in the adjustments their publishers had made. The lead market is not a preview. It is a separate market that happens to be further along one particular calendar, and different on most of the other dimensions that matter.

Cross-market comparisons need a date on them

This is the practical consequence, and the easiest to act on. A sentence such as we are named more often in Spanish than in French is incomplete without two further facts: when the measurement was taken, and which surfaces were measured. A comparison made in June 2026 set a French market without Google's generative features against a Spanish market that had had them for well over a year. The same comparison made in October 2026 is comparing something else, even if the spreadsheet looks identical.

That argues for a few habits which cost very little.

• Date every figure that crosses a market boundary, on the slide as well as in the workbook. A number without a date will be quoted a year later as though it were current.
• Keep a short log of launches by market and language for the surfaces your buyers use: the day a feature arrived, taken from the vendor's own announcement rather than from recollection. The five lines above are a reasonable start for Google.
• Compare a market with itself before comparing it with its neighbours. Movement within France across a launch date can be interpreted. A gap between France and Spain on the same day blends content, language and calendar into a single figure.
• Name the surface. A result from a chat assistant and a result from a search summary are different measurements, and a cross-market comparison that mixes the two is mixing things twice over.

Why the unit of measurement is language and market

All of which is why PSentry reports by language and market rather than as one global figure. Each language and market gets its own AI Visibility Score, its own trend over time and its own list of the competitors named in your place, so that a strong result in one never quietly averages away a weak result in another. Segmenting by language and market is not the same thing as geolocating anybody: the tool does not simulate a user standing in a particular town, and it has nothing to do with local search.

The boundaries deserve equal plainness. PSentry measures four assistants: ChatGPT, Claude, Gemini and Perplexity. It does not monitor AI Overviews or AI Mode, so the French launch described at the top of this piece is not something it would have seen at all. Scans run twice a month, which is ample for following how a market moves across a quarter and no use whatever for catching a launch on the day it happens; for that, the vendor's own announcement remains the better record. And it reports what the assistants say about you. It does not alter it.

What exporters tend to ask about uneven rollouts

If AI Mode is now available in all five markets, does the history still matter?

For any comparison that looks backwards, yes. A trend line spanning July 2026 in France, or October 2025 in Germany, Italy and Spain, crosses a change in what buyers there could be shown. For measurements taken from now on, the search features are at least nominally level across those markets, though differences of plan and feature inside the assistants remain.

Should we prioritise the market that had generative search first?

Not on that ground alone. The arrival date is one fact about a market among many, and seldom the one that decides where the revenue is. Prioritise on the usual commercial grounds, then use the dates to read the measurements correctly.

Can our English results serve as a benchmark for the other languages?

As an illustration of what good looks like, possibly. As a prediction of where the other languages will end up, no. The material an assistant draws on is built separately in each language, and a lead in one does not transfer to the next.

Where is the reliable record of what launched where?

The vendors' own announcements and help pages, including the local-language editions, which sometimes name countries and conditions that the global post leaves out. Trade coverage is useful for finding them, and rather less useful as the record itself.