No one has published an AI-visibility dataset for German dining before. This is the first one. Over the 26 days from 1 to 26 July 2026, we tracked 22 Berlin restaurants and wine bars across 35 real dining questions, run daily on three AI engines, ChatGPT, Gemini and Google AI Overview, and scored 823 answers for who got named and which sources built the answer. The headline finding surprised us: Berlin's most influential AI dining source is not a food magazine. It is visitberlin.de, the city's own tourism board.
TL;DR
- This is the first published AI-visibility dataset for German dining. Across 823 scored answers, visitberlin.de, Berlin's official tourism board, was retrieved in 24.9% of all answers and 44.9% of ChatGPT answers, ahead of every magazine, listings site or booking platform we tracked.
- Restaurant Tim Raue leads the leaderboard at 18.8% visibility, the highest score in Berlin, and it still misses more than four in five relevant answers.
- Three of the 22 tracked restaurants, Wild Things, Kin Dee and Estelle Dining, were never named once across all 823 answers.
- Which AI engine a diner uses changes the Berlin answer more than which restaurant is "best": Facil appears in 26.8% of ChatGPT answers but only 3.4% of Gemini answers.
- berlinfoodstories.com, a one-person independent blog, is cited harder per retrieval (2.36) than any national title in the dataset, including Falstaff (0.62).
What did we measure, and how?
We tracked 22 Berlin restaurants and wine bars (a 23rd tracked entity was our own control brand and is excluded from every restaurant figure here) across 35 dining questions of the kind a real diner types: best restaurant for a special occasion, where to eat near a district, a wine bar for a date. Those questions ran daily across ChatGPT, Gemini and Google AI Overview for the 26 days from 1 to 26 July 2026, producing 823 scored answers: 385 on ChatGPT, 385 on Gemini and 53 on Google AI Overview. For each answer we recorded whether a restaurant was named and which source domains the engine pulled in. This is our own daily multi-engine tracking, not a survey and not a self-report.
Three caveats before the numbers. The Google AI Overview sample is small, 53 answers in total, so any AI Overview figure below is directional, not a settled read. This is a young dataset, a first look, and we will refresh it quarterly rather than treat it as a fixed ranking. And visibility here means the share of tracked answers a restaurant was named in; not every one of the 35 questions suits every restaurant, so no venue should be expected anywhere near 100%.
Which Berlin restaurant is most visible in AI search?
Read the top of the table twice. Even the leader clears less than one in five answers.
| Restaurant | Visibility | Mentions | Avg position |
|---|---|---|---|
| Restaurant Tim Raue | 18.8% | 237 | 2.7 |
| Katz Orange | 17.1% | 164 | 3.1 |
| Rutz | 16.2% | 208 | 2.7 |
| Facil | 14.3% | 152 | 3.1 |
| Cookies Cream | 10.6% | 115 | 6.4 |
| 893 Ryotei | 7.8% | 71 | 2.4 |
| Nobelhart & Schmutzig | 7.2% | 73 | 4.7 |
| Grill Royal | 6.3% | 74 | 2.0 |
| Jaja | 4.9% | 50 | 4.3 |
| Freundschaft | 4.5% | 46 | 5.0 |
| Horvath | 4.1% | 39 | 3.5 |
| Otto | 3.7% | 34 | 6.0 |
| Borchardt | 2.9% | 33 | 3.0 |
| Lovis | 2.4% | 21 | n/a |
| Barra | 2.2% | 20 | 5.0 |
| Lorenz Adlon Esszimmer | 1.7% | 16 | n/a |
| Ottorink | 0.9% | 7 | 4.0 |
| Sphere by Tim Raue | 0.7% | 6 | n/a |
| Bocca di Bacco | 0.4% | 3 | n/a |
| Wild Things | 0% | 0 | n/a |
| Kin Dee | 0% | 0 | n/a |
| Estelle Dining | 0% | 0 | n/a |
Restaurant Tim Raue tops the Berlin index at 18.8%, and that ceiling is instructive on its own: the single most AI-visible restaurant in the German capital still goes unnamed in more than four out of five answers to questions it should be a strong candidate for. Katz Orange and Rutz follow closely, both above 16%, and Facil rounds out a clear top four.
Grill Royal is the outlier worth flagging. Modest visibility at 6.3%, but the best average position in the whole dataset at 2.0, meaning when an engine does name it, it names it first or second, not buried at the bottom of a list of six.
One more read worth a beat: two of the 22 tracked venues carry the Tim Raue name. Restaurant Tim Raue leads the entire index at 18.8%; Sphere by Tim Raue, the second venue under that name, sits at just 0.7%. Same chef, same brand, radically different AI visibility, which is close to the venue-cannibalisation pattern we cover in restaurant groups and ChatGPT.
At the other end, three of the 22 tracked restaurants, Wild Things, Kin Dee and Estelle Dining, were never named once across all 823 answers. Zero mentions, zero visibility, across every question and every engine. That is the sharper way to read a visibility index: it is not only who leads, it is who is entirely absent from the conversation an AI engine is having about where to eat in Berlin.
Why does a tourism board beat the magazines in Berlin's AI answers?
This is the headline, and it is the opposite of what we found in the other two cities we track. In Berlin, the single most retrieved source across all three engines is visitberlin.de, the city's official tourism board, pulled into 24.9% of all 823 answers, ahead of every booking site, listings platform and food magazine we tracked. Add berlin.de, the city's own portal, retrieved in 8.6% of answers, and the institutional layer is the backbone of the Berlin dining answer.
| Domain | Type | Retrieved in | Citation rate |
|---|---|---|---|
| visitberlin.de | Institutional (city tourism board) | 24.9% | 1.84 |
| opentable.de | Booking | 17.4% | 0.62 |
| falstaff.com | Editorial | 15.2% | 0.62 |
| top10berlin.de | Editorial | 14.0% | 1.16 |
| tripadvisor.de | Listings | 13.4% | 0.76 |
| mitvergnuegen.com | Editorial | 13.0% | 0.73 |
| tagesspiegel.de | Editorial (news) | 10.6% | 0.59 |
| reddit.com | Community | 10.1% | 1.99 |
| berlin.de | Institutional (city portal) | 8.6% | 1.96 |
| michelin.com | Guide | 6.6% | 1.20 |
On ChatGPT specifically the pattern is even sharper: visitberlin.de was retrieved in 44.9% of ChatGPT's Berlin dining answers, 411 retrievals and 336 citations, a citation rate of 1.94. Nearly one in two ChatGPT answers about where to eat in Berlin pulls from the tourism board's own pages.
We track two other cities the same way, and neither looks like this. In London, Time Out, an editorial title, was retrieved in about 64% of ChatGPT dining answers; we published the full source study here. In New York, three editorial titles led instead, Eater, The Infatuation and Time Out, retrieved in 40%, 36% and 37% of answers respectively. Berlin inverts that pattern entirely: a government tourism body outranks every magazine, and Reddit is markedly weaker here too, retrieved in 10.1% of Berlin answers against 28.4% in New York. If you are used to reading US or UK GEO advice that says earn the food magazines first, that advice does not transfer cleanly to Berlin. Our UK Restaurant AI Visibility Index is a useful side-by-side; the source hierarchy each city produces is genuinely different.
Does the AI engine you use change the Berlin answer?
Yes, more than which restaurant is objectively best does. Look at the same six restaurants across three engines.
| Restaurant | ChatGPT | Gemini | AI Overview (small sample) |
|---|---|---|---|
| Katz Orange | 12.5% | 18.7% | 39.6% |
| Facil | 26.8% | 3.4% | 3.8% |
| Restaurant Tim Raue | 21.0% | 15.3% | 28.3% |
| Nobelhart & Schmutzig | 4.2% | 8.1% | 22.6% |
| Rutz | 16.1% | 15.3% | 22.6% |
| Cookies Cream | 9.4% | 12.7% | 3.8% |
Facil is the sharpest case in the dataset: strong on ChatGPT at 26.8%, almost absent on Gemini at 3.4% and on AI Overview at 3.8%. Katz Orange runs close to the opposite shape, growing from ChatGPT to Gemini and then further again on AI Overview, though that AI Overview column comes from only 53 answers in total and should be read as directional, not settled. A Berlin restaurant that only checks how it looks on one engine, most often ChatGPT, because it is the one people open first, can be sitting on a real blind spot on the other two.
Why does a one-person blog out-cite the national food magazines?
The other finding worth real space is about intensity, not just reach. Citation rate measures how hard an engine leans on a source once it has been retrieved, and it can run above 1.0 because an engine can cite the same domain more than once inside a single answer. berlinfoodstories.com, a single independent Berlin food blog, was only retrieved in 5.7% of answers, a modest reach next to visitberlin.de. But its citation rate is 2.36, higher than Falstaff (0.62), higher than Tagesspiegel (0.59), higher than every national masthead in the dataset. Reddit shows a similar shape at 1.99, and berlin.de at 1.96.
The read here is not that the big magazines don't matter; visitberlin.de and Falstaff both carry real reach. It is that when an engine finds small, specific, well-structured local writing, it leans on it harder per retrieval than it leans on a big masthead. For an independent Berlin restaurant without a national press relationship, one well-written, well-structured local blog post can carry more weight per mention than most operators assume.
What should a Berlin restaurant do with this?
Two practical reads. First, if you are not visible, the fix is not another pass on your own website copy, it is presence in the specific sources this data names: your city's own tourism board content, the Berlin food editorial and blog layer (Falstaff, top10berlin.de, mitvergnuegen.com, berlinfoodstories.com), and an honest Reddit and community footprint. Second, check your visibility per engine and not only on ChatGPT, because the swing between engines in this dataset is larger than the swing between most restaurants. For the fuller German-market picture, and what changes as Google AI Overview access widens, see our Germany GEO guide for restaurants, and for the Kiez-by-Kiez mechanics of how a Berlin answer gets built, see where AI sends diners in Berlin. If your venue is drinks-led, our wine bar GEO guide covers the category-labelling problem specifically.
Disclosure: Schmitdy is our own AI search service at AI Heroes, and this index comes from the same daily multi-engine tracking we run for the restaurants we work with. We are publishing it because a German dining benchmark did not exist anywhere, and we will refresh it quarterly rather than let it go stale. If you want your own restaurant's number, and which of these sources currently name you, the free AI search audit shows it.
