Which dentist does AI recommend in Vienna?
We asked ChatGPT, Perplexity, Gemini and Google's AI Mode twice: once generally for a good dentist, once specifically for a dental implant. Of 17 recommendations from the general question, only 2 survived the switch to the buying question.
🇩🇪 Diese Studie auf Deutsch lesenMore and more people ask an AI first when they need a dentist. It only gets interesting, though, when you ask twice: once generally, and once the way someone asks who is about to spend money. That is exactly what we did, on the same day, in the same four AIs.
On the general question the four AIs named 11 practices between them, not one of which appeared in all four. And when we switched to the implant question, only 2 of 17 recommendations were left.
The experiment
On 18 August 2026 we asked all four major AI engines two identical, word-for-word questions from a Vienna browser, each in a fresh, empty chat:
(“I am looking for a good dentist in Vienna. Can you recommend a few specific ones, each with address, phone number and website?”)
(“I need a dental implant in Vienna. Which dental practice can you recommend?”)
Question 1 is deliberately built exactly like the one in our vet study from June, so the two sectors can be compared.
Finding 1: No practice makes all four
Across the four engines the general question produced 11 different dental practices. The best result was Dr. Mathias Lidauer with 3 out of 4 mentions, missed by Gemini of all engines. Named by all four: nobody.
Answer length varies a lot as well: Perplexity named 6 practices, ChatGPT 5, Google AI Mode and Gemini 3 each.
Finding 2: Google's two AIs disagree again
Just like with the vets in June: Gemini and Google's AI Mode, both from the same company, did not name a single practice in common on the general question. They overlapped for the first time only on the implant question, with exactly one name.
| Engine | Names on question 1 | Relies on |
|---|---|---|
| Perplexity | 6 practices: Hallmann, Azzawi, Stanzl, Stadtpark, Lindenberger, Lidauer | 10 sources, all 10 of them the practices' own websites |
| Google AI Mode | 3 practices: Azzawi, Lidauer, Zgkouri | Google Business Profiles, the Vienna dental chamber register, one YouTube video |
| ChatGPT | 5 practices: Lidauer, Lindenberger, Cerny, Stadtpark, University Dental Clinic | practice websites plus the local business listing |
| Gemini | 3 practices: Hallmann, Glaser, Fuchs | no visible sources, map data only |
One detail stands out: Gemini recommended two practices rated 4.0 and 4.2 stars, while every other engine stayed between 4.4 and 5.0. Gemini clearly does not follow the star ranking.
Finding 3: The buying question throws the list away
This is the genuinely new part. Same engines, same day, only a different question. How many recommendations from the general question show up again for “I need an implant”?
| Engine | Recommendations on question 1 | of those, back on question 2 |
|---|---|---|
| Perplexity | 6 | 1 |
| Google AI Mode | 3 | 1 |
| ChatGPT | 5 | 0 |
| Gemini | 3 | 0 |
| Total | 17 | 2 |
Being recommended for “a good dentist in Vienna” says almost nothing about being recommended for “I need an implant”.
The reverse holds too: of the 13 practices named on the implant question, 11 did not appear on the general question at all. And on question 2 as well, not a single practice made it into all four engines.
Why this happens: the sources shift too
The reason sits in the source lists. On the general question, Google's AI Mode drew mainly on Google Business Profiles, plus the official register of the Vienna dental chamber. On the implant question, the same service switched to explainer content, that is, pages describing the treatment and above all what it costs.
That is the practical lever, and it is uncomfortably clear:
- The general question is won by the profile. A well-kept, current Google Business Profile with plenty of reviews gets you named for “a good dentist in Vienna”.
- The buying question is won by the content. Whoever explains what an implant costs, how long it takes and what insurance covers gets named for “I need an implant”. A good profile alone will not do it.
Perplexity was striking too: all 10 sources it cited were the practices' own websites. No directory, no review portal, no press article. If you have no usable website, you do not exist for Perplexity on this question.
The fact-check: no dead links this time
We opened all 11 websites the engines explicitly named on question 1. All 11 work. Two pages blocked our fetch script through bot protection but loaded normally in the browser. Unlike the vets in June, where two sites were broken, there was not a single dead link here.
One side observation that still explains a lot: Google's AI Mode recommended one practice explicitly for emergencies. On the day of the test, that practice's website carried a holiday notice, the surgery was closed for two weeks. That is not the practice's fault. It just shows that an AI recommendation says nothing about whether anyone is there today.
What this means for Vienna businesses
Dentist, tax adviser or trade business, the mechanics are the same everywhere:
- One check is not enough. Testing one engine with one general question measures next to nothing. Four engines and at least one real buying question are the minimum.
- Profile and content are two separate jobs. The Google profile wins the general questions; explainer content about the specific service wins the questions where someone intends to buy.
- Your own website counts more than many assume. In this sector, practice websites were the only source Perplexity used.
This approach is called Generative Engine Optimization (GEO), and it is what KIGEO does.
Does AI name your business, including on the question that brings money?
We check for free whether ChatGPT, Perplexity, Gemini and Google name you, on the general question and on the concrete buying question, and who gets recommended instead.
get a free AI visibility check →Methodology & sources
All four engines were queried on 18 August 2026 from a Vienna browser (logged in, one fresh chat per question and engine, a single run) using the two identical questions quoted above. We counted every practice recommended by name in the answer text; practices that appeared only in a map widget or in the source list were noted separately and not counted as recommendations. We then fetched all 11 websites named by the engines and opened the two flagged ones in the browser. A sample, not a full survey.
What we did not check: in this edition we did not verify the addresses and phone numbers individually against the practice websites, unlike in the vet study. We therefore deliberately make no claims here about whether addresses and numbers are correct.
Noted honestly: the first ChatGPT run on question 1 broke off; the answer announced a search and then named no practice at all. We counted the second run, in a new, empty chat, with the same question.
- Data collection and raw data: KIGEO, own measurement of 18 August 2026, in full in the CSV below.
- Comparison data from the same series: KIGEO vet study Vienna (13 June 2026), same method, same question structure.
- Practice websites opened for the check: zahnarzt-hallmann.at, zahnmedizin-hernals.at, zahnarzt-stadtpark.at, zahnarzt-lindenberger.at, zahnarztwien.at, dr-cerny.at, zahnarzt23.at, zahnarztfuchs.at, jacobs-implantate.at, azzawizahnaesthetik.at, dr-zgkouri.at.
Download the raw data of this study as CSV ↓
Note: the CSV is in German.