Do AI assistants recommend clinics? AI visibility pilot in Spain (October 2026)
Hardly ever. In our pilot of October 1, 2026, 46 of the 78 dental implant and aesthetic medicine clinics we audited in Madrid, Valencia and Málaga (59%) did not appear in a single answer from ChatGPT, Gemini, Perplexity or Claude, and their average AI visibility score was 5.1 out of 100. It is a pilot by Elevatec Techlearn, an artificial intelligence agency and fractional CTO, with 35 questions and a single run: a first data point, not a definitive study.
- Answer date
- October 1, 2026, a single run
- Questions
- 35 distinct, 140 valid answers (35 per engine)
- Clinics audited
- 78: 38 dental implant, 39 aesthetic medicine and 1 general dental
- Cities
- Madrid (27 clinics), Valencia (31) and Málaga (20)
- Engines
- ChatGPT, Gemini, Perplexity and Claude, with web search
- Type
- Pilot. A larger study is under way
What this pilot is and what it is not
We asked four AI assistants the questions a patient would ask when looking for a clinic, for example which is the best implant clinic in their city or where they can get a treatment at good value. Then we checked whether each audited clinic appeared in the answers and in which position. There are 35 questions across 7 niche and city combinations, answered once on October 1, 2026.
The number of clinics is enough for a first picture. The number of questions is not, and with a single run we cannot measure how much answers change from one day to the next. That is why we publish it as a pilot. We are already running a larger study, with 6 cities and 3 waves of questions spread over time, and will publish it here when it is finished.
1. Six in ten clinics never appear
Of the 78 clinics audited, 46 (59%) do not appear in any of the 20 answers for their niche and city (5 questions to 4 engines). Only 32 appear at least once in any AI.
By engine, 19 clinics appear in ChatGPT, 20 in Gemini, 16 in Perplexity and 8 in Claude, always out of 78.
2. The average visibility score is 5.1 out of 100
The score gives 100 points for coming first, 80 for second, 65 for third, 45 for lower down or only as a cited source and 0 for not appearing, averaged over each clinic's 20 answers. The mean is 5.1 and the median is 0. The highest score was 61.5.
| Score (0 to 100) | Clinics | Share of total |
|---|---|---|
| 0 | 46 | 59% |
| 0.1 to 9.9 | 18 | 23% |
| 10 to 24.9 | 9 | 12% |
| 25 to 49.9 | 4 | 5% |
| 50 or more | 1 | 1% |
3. When they appear, it is rarely on the podium
Crossing each clinic with every answer for its city and niche gives 1,560 combinations. The clinic appears in 137 (8.8%) and in the top three only in 48 (3.1%): 15 times first, 18 second and 15 third. The other 89 appearances are lower down or only as a cited source.
4. Each AI recommends different clinics
In the 28 questions where ChatGPT, Gemini and Perplexity all gave a list, the first recommendation did not match in any of them. Only in 8 of the 35 questions did all three name at least one clinic in common, and the similarity between two engines' lists for the same question is very low (an average Jaccard index of 0.06 to 0.10).
Across all questions, those three engines named 307 distinct businesses. 232 (76%) were named by just one engine and only 26 (8%) appeared in all three. Doing well in ChatGPT says nothing about how you do in Gemini or Perplexity.
| Engine | Audited clinics that appear (of 78) | Answers naming no clinic (of 35) | Distinct businesses named |
|---|---|---|---|
| ChatGPT | 19 | 7 | 140 |
| Gemini | 20 | 1 | 122 |
| Perplexity | 16 | 0 | 146 |
| Claude | 8 | 0 | 77 |
ChatGPT is the one that most often answers without naming anyone: in most of those answers it asks for more details before recommending. Claude names fewer distinct businesses because it repeats the same ones more.
5. Clinics' own websites are the most cited source
We count each cited domain once per answer. That gives a total of 1,004 citations across the 140 answers.
| Source type | Citations | Share of total |
|---|---|---|
| Website of a clinic or healthcare group | 468 | 47% |
| Other directories, rankings and booking platforms | 244 | 24% |
| Doctoralia and Top Doctors | 146 | 15% |
| Press and magazines | 81 | 8% |
| Institutions, professional bodies and scientific societies | 37 | 4% |
| Other (unclassified, social media, Google Maps) | 28 | 3% |
Top Doctors appears as a source in 75 of the 140 answers and Doctoralia in 71. Perplexity cited both in all 35 of its answers; ChatGPT cited Top Doctors in 4 and Doctoralia in 13. Google Maps was cited only once, but read that with care: we queried via API, and consumer apps do tend to use maps.
6. Nobody owns the recommendations
In each niche and city combination, the 3 most named clinics take between 17.5% and 29.4% of mentions, and the top 10 between 40.8% and 58.7%. Of the 344 businesses named, 191 (56%) came up only once.
Dental implants in Madrid is the most concentrated combination: the top 3 take 29.4% of 109 mentions. Aesthetic medicine in Valencia is the most open: 17.5% of 103 mentions, spread across 57 distinct businesses.
7. Technical signals alone do not explain who appears
- 71 of the 78 clinics already have structured data on their website. 41% of those with it appear, and 43% of those without (71 and 7 clinics).
- Clinics with an llms.txt appear somewhat more: 14 of 28 (50%) versus 18 of 50 (36%). With this sample the difference is not significant (p = 0.24).
- More Google reviews help a little: 31% appear in the third with fewer than 155 reviews versus 46% in the other two (26 clinics per third). The correlation with the score is weak (0.17) and the average Google rating has no effect (−0.09).
With 78 clinics none of these differences is statistically significant. We report them as hints, not causes.
What it means for a clinic
First, that not appearing is the norm today, so there is a lot of ground to gain. Second, that clinics' own websites are the source assistants use most, ahead of directories: what you publish about your treatments, your team and your prices counts. Third, that Doctoralia and Top Doctors weigh heavily in Perplexity and little in ChatGPT, and that each engine recommends different clinics. Looking at only one gives an incomplete picture.
If you want to know where your clinic stands, the free AI visibility audit applies this same method to your city and treatment. And if your concern is serving the patients who already message you, see the AI receptionist for clinics.
How it was done
The questions were written the way a patient would write them, with no brands, varying the intent: the best, trustworthy, good reviews, a specific treatment, value for money and near me. ChatGPT, Gemini and Perplexity were queried via API with web search turned on. Claude's answers were written by an agent with web search, not the claude.ai app, so they are best read separately.
A language model extracted from each answer the ordered list of recommended businesses, leaving out directories and platforms, and we grouped variants of the same name. The scoring is our GEO methodology. The clinic audits were run between October 1 and 7 and all use the answers from October 1.
Limits
- Small sample and a single run: 35 questions answered once. These engines' answers change between runs and over time, and that variation is not measured here.
- API versus apps: what a user sees in each assistant's app can differ, with another model, more searches, their history or their exact location.
- Claude was not queried in its app, but through an agent with web search.
- ChatGPT and Perplexity received an approximate location (the city); Gemini did not. Near me questions have no exact point.
- Name extraction and grouping are automatic and may miss or merge a name.
- The clinics are not a random sample: there is a bias towards those with a website, a public email and plenty of reviews.
- Only three cities. It does not support claims about all of Spain without caveats.
Frequently asked questions
Very little, according to our pilot of October 1, 2026. Of 78 dental implant and aesthetic medicine clinics in Madrid, Valencia and Málaga, 46 (59%) did not appear in any answer from the four assistants and the average visibility score was 5.1 out of 100.
35 distinct questions, 140 valid answers (35 per engine) and 78 clinics: 38 dental implant, 39 aesthetic medicine and 1 general dental, spread across Madrid (27), Valencia (31) and Málaga (20).
No. In the 28 questions where ChatGPT, Gemini and Perplexity all gave a list, the first recommendation did not match in any. Of 307 businesses named by those three engines, 232 (76%) were named by just one.
Mostly clinics' and healthcare groups' own websites (47% of 1,004 citations), followed by other directories and booking platforms (24%), Doctoralia and Top Doctors (15%) and the press (8%).
No. 71 of the 78 clinics already had structured data and appeared just as rarely as the rest. Clinics with an llms.txt appear somewhat more (50% versus 36%), but with this sample the difference is not significant.
Because it is 35 questions answered once, in three cities. Measuring how much answers change requires repeating them. We are running a larger study, with 6 cities and 3 waves, which we will publish on this same page.
Ready to work together?
Schedule a meeting or tell us about your project. We respond quickly, no strings attached.