Study · AI Visibility

Which restaurants does AI recommend in Vienna?

We asked ChatGPT, Perplexity, Gemini and Google's AI Mode the same question, repeatedly, in a single day. 31 venues were named. On exactly one, all of them agreed.

🇩🇪 Diese Studie auf Deutsch lesen

More and more people no longer ask Google, they ask AI: “Where should I go to eat in Vienna?” According to a BrightLocal survey, 45% of consumers already use AI for local recommendations, a year earlier it was 6%.1 For a venue, that means: whoever gets named by AI wins the guest. Whoever is missing simply does not exist for those guests. So we tested it.

The result in one sentence

Across four AI engines, around 31 different venues were recommended, but only one, the Steirereck, was named by all of them. And in every single run at that.

The experiment

We wanted more than a gut feeling, we wanted a picture you can retrace. That is why, on 10 June 2026, from a Vienna-based browser, we asked all four major AI engines the exact same question, word for word:

The exact question to each AI
“What are the best restaurants in Vienna? Please give me specific recommendations.”
Engines
ChatGPT, Perplexity, Gemini, Google AI Mode
Runs
ChatGPT, Perplexity & Gemini 2× each, Google AI Mode 1×, seven answers in total
Date
10 June 2026
Location
Vienna, Austria (real IP, no VPN)
Counted
Every venue named by name in the answer
Payment
None, no one was paid for a mention

Why repeatedly? Because AI answers fluctuate. A single query would be a snapshot, not a statement. That very fluctuation is part of the result.

The result: agreement only at the top

The same pattern showed up across all engines: at the very top sit the same big names, and after that the lists drift far apart. Only the Steirereck made it into every one of the four engines, and into all seven answers. Amador and Silvio Nickol reached three out of four. The rest, around two thirds of all named venues, appeared in only a single engine.

Steirereck
4
Amador
3
Silvio Nickol
3
SIXTA
2
Figlmüller
2
TIAN
2
~24 more venues
1
In how many of the 4 AI engines was the venue named? (first run per engine). Steirereck the only one in all four.

And this is the list each engine gave, along with the sources it drew on:

EngineNames, among othersDraws on
ChatGPTSteirereck, Amador, Silvio Nickol, SIXTA, Figlmüller, BauernbräuGoogle reviews, Reddit, travel blogs
PerplexitySteirereck, Amador, Konstantin Filippou, Tian, Wiener WiazhausGault&Millau, austria.info, venues' own websites
GeminiSteirereck, Amador, Mraz & Sohn, TIAN, PlachuttaGoogle knowledge, editorial sources (no individual links)
Google AI ModeSteirereck, Silvio Nickol, Glasswing, Figlmüller, SIXTAGoogle business profiles (33 websites)

The same AI, two answers

It becomes even clearer when you ask the same AI twice. Between two runs, ChatGPT swapped out around half of its recommendations, with new names and a different order. Perplexity and Gemini stayed more stable, but mainly held on to their top picks. For a venue, that means: being named in one engine does not mean showing up in the others, and certainly not permanently.

Why these venues in particular?

Because AI does not invent anything, it summarizes what is documented online, frequently and credibly. The Steirereck ranks No. 33 on the “World's 50 Best Restaurants 2025”2, holds three Michelin stars and tops the Gault&Millau ranking with 19 points and 5 toques3. Across the sources it is described as Vienna's best-known fine-dining restaurant, and that is exactly what the AI builds its answer from.

And the sources differ by engine (see the table above): ChatGPT showed a Google review for each venue and additionally drew on Reddit and travel blogs; Perplexity relied at the top on Gault&Millau and austria.info; Google's AI Mode read practically straight from Google business profiles. Whoever is missing from these sources can cook however well and will still not be named by the AI.

For the AI, “best” means above all: best documented.paraphrased after Omni Search Labs, US experiment on AI restaurant search6

What larger studies show

Our Vienna test is a snapshot, but it fits larger surveys. Important: these all come from industry tools, not from peer-reviewed research, and should therefore be read as indicative.

  • Local Falcon analyzed 189,905 ChatGPT answers: 83% of restaurants never appear there, only about one in six is recommended at all. On Google, at least 86% show up somewhere.4
  • A SOCi analysis found that ChatGPT recommended only 1.2% of locations.5
  • A US experiment by Omni Search Labs asked five engines the same restaurant question; two of them shared only one common venue out of eight.6

Visibility in AI is therefore rare and very unevenly distributed.

Caution: AI makes mistakes too

You should not trust the AI blindly here. Three observations from our test:

  • Fluctuation. The same question, a different answer, as described above.
  • Confident errors. One of the AIs tested called the Amador “Vienna's only restaurant with three Michelin stars.” That is wrong: since the Guide Michelin returned to Austria in January 2025, Vienna has two three-star houses, Steirereck and Amador.7
  • Outdated data. AI models demonstrably recommend venues that closed long ago or invent awards, a documented problem.8
For a serious study that means: measure repeatedly, check sources, take nothing on trust. That is exactly how this analysis came about, every figure is cited below, and we cross-checked every restaurant award against the official guides.

What this means for Vienna's venues

For Vienna's venues, that means: an excellent restaurant can stay invisible to the AI if the digital trail is missing. Whoever appears where the AI reads becomes visible:

  • a well-maintained Google business profile with many, good reviews,
  • consistent listings on the platforms the engines draw from (Google Maps, TripAdvisor, TheFork, Foursquare),
  • presence in the sources that matter for Vienna, Falstaff, Gault&Millau, Falter's “Wien, wie es isst”,
  • a machine-readable website (opening hours, menu, address as structured data),
  • and timeliness: whoever closes, relocates or changes the menu has to update it everywhere.

Visibility in AI can be built deliberately. This approach is called Generative Engine Optimization (GEO), and that is what KIGEO specializes in.

Does AI know your venue?

We check for free whether ChatGPT, Perplexity, Gemini & Google recommend your restaurant, and show you who gets named instead.

free AI visibility check →

Methodology & sources

All engines were queried on 10 June 2026 with the exact wording quoted above from a Vienna-based browser (ChatGPT, Perplexity, Gemini 2× each, Google AI Mode 1×). Named venues were recorded by name; all restaurant awards were checked against the official guides. Sources used:

  1. BrightLocal, Local Consumer Review Survey 2026, 45% use AI for local recommendations (up from 6%).
  2. The World's 50 Best Restaurants, Steirereck, No. 33 (2025).
  3. Gault&Millau Österreich, Guide 2026, the best restaurants in Vienna (4 five-toque establishments).
  4. Local Falcon, The AI Visibility Crisis, 83% of restaurants invisible in ChatGPT (189,905 results, February 2026).
  5. SOCi, Local Visibility Index 2026, via Search Engine Land, ChatGPT recommended 1.2% of locations.
  6. Omni Search Labs (Chris Sheehy), Restaurant Search Experiment, 2026, “best = most documented”.
  7. Guide Michelin, Return to Austria (January 2025); Vienna: two three-star houses.
  8. The LO Times (Ryan Sutton), “Don't Ask ChatGPT for Restaurant Advice”, AI recommends closed venues / invents awards.

Download this study's raw data as CSV ↓

K

About KIGEO

KIGEO is an AI visibility (GEO) and SEO agency from Vienna. We make sure businesses get named in ChatGPT, Perplexity, Gemini and Google AI, and we measure it honestly. This study was produced with the same methods we use for our clients. Request a free check →