Two surveys, one picture
If you want to show up in AI answers, the first question is mundane and rarely asked: which AI are we talking about? For Germany there are solid figures, and they are not ours.
Bitkom Research, commissioned by the Bitkom association, surveyed 1,003 people aged 16 and over by telephone, representative, in calendar weeks 8 to 11 of 2026. Those who said they use AI were then asked about individual services.
| Service | Share of AI users |
|---|---|
| ChatGPT | 71 % |
| Gemini | 50 % |
| Microsoft Copilot | 43 % |
| Meta AI | 35 % |
| DeepSeek | 8 % |
| Perplexity | 7 % |
| Grok | 6 % |
| Claude | 5 % |
| LeChat | 4 % |
As a cross-check: Splendid Research surveyed 1,000 people aged 18 to 69 through an online panel, quota-representative, from 18 to 26 June 2026. There ChatGPT reaches 73.7 percent and Gemini 49.3 percent of AI users. Two different methods, four months apart, and practically the same result for the two largest services. That is why these figures are quoted here at all.
What the figures do not say
They are not market shares. People were asked which services they use, and multiple answers were possible. That is why the values add up to more than 200 percent. Turning the 71 percent into a 71 percent market share says something the survey does not support.
And neither survey asks about local search. Someone using ChatGPT is not necessarily using it to find a roofer in Lüneburg. They may be writing emails with it. For the question of whether your trade in your town gets found through an AI, only a real query helps, and our article on checking it yourself shows how to run one.
What this means for your business
ChatGPT and Gemini first. Together they cover most AI use, and they are two different doors: one sits in the browser and the app, the other in the Android phone and next to Google search.
Perplexity is smaller than its reputation. The industry mentions it a lot, we do too, and that does not match 7 percent. It remains worth measuring because it discloses its sources and therefore shows what an answer rests on. As reach among your customers it is small.
Google's AI Overview is something else again than Gemini. The overview is a feature inside search, Gemini is the assistant. Both come from Google, both behave differently, and we keep them apart in our measurements for that reason. More in our article on AI Overviews for local businesses.
Where we measure ourselves, and where we do not
For our clients we query three services every month: ChatGPT, Gemini and Perplexity. Per the table above that is 71, 50 and 7 percent. Copilot at 43 percent and Meta AI at 35 percent can only be queried by hand and signed in. We have collected nothing there so far, and if we did, those figures would stay separate from the anonymous measurement. This is stated here because a measurement is only worth what you know about its limits.
For Copilot the reason can be named. Our tooling can reach the service, but since 8 September 2026 Copilot shows its own sign-in page to signed-out visitors instead of an answer. A sign-in page is not a measurement, and we do not count it as one: it would otherwise pass as "not mentioned" and turn a failed measurement into a decline. Querying automatically while signed in is something Microsoft forbids in its terms of use. And even if it were allowed, the result would hang on an account and its history rather than on the market.
Meta AI is nearly the same case, although we described it differently here for a long time. There is a web interface, at meta.ai, in German since 2025. You still cannot ask it anything without signing in: next to the input field sits a prompt to sign in, and the field stays locked. Meta forbids automated queries as well, in section 3.2 of its terms of service.
The honest sentence is this: three services we measure automatically and signed out, the other two not at all. For those we have a way to record an answer by hand and signed in, and we have not used it yet. Should such a figure arrive, it would stay separate from the anonymous measurement and never be added to it, because a signed-in query can also reflect the past of an account while a signed-out one reflects the market. Anyone receiving a figure from us therefore knows what it refers to. How reliable a single such measurement is at all we investigated separately, which is why we run three passes per service rather than one.
How current these figures are
Both surveys are from February to June 2026. AI use shifts quickly, which is why the date sits in the text rather than in a footnote. We check these figures once a year and update the article rather than leaving it to age. If you are reading this much later, check the linked sources for a newer survey.
Common questions
Are these market shares?
No. They are usage shares among people who use AI at all, and multiple answers were possible. That is why the values add up to more than 200 percent. Passing them on as market share says something the survey does not support.
Does Gemini at 50 percent mean I matter there as much as on ChatGPT?
The number does not say that. It says how many people use the service at all, not what for. Neither survey asks about searching for a business nearby. For your trade and your town, only a real query helps.
Why do you measure Copilot differently from the rest?
Because Copilot no longer answers anonymously. Since 8 September 2026 the service shows its own sign-in page to signed-out visitors, and a sign-in page is not a measurement. Querying automatically while signed in is expressly forbidden by Microsoft's terms of use. That leaves recording an answer by hand and signed in. We have that route and have not used it so far, and such figures would stay separate from the anonymous measurement.
How long do these numbers hold?
The surveys are from February to June 2026. AI usage shifts quickly, which is why the date is stated here and why we check the figures once a year. If you are reading this much later, look at the sources for a newer survey.