Zero out of fifteen: when AI couldn't reliably recommend a single dental practice

I asked four free AI models about fifteen Hungarian dental practices, phrased exactly the way a patient would: "which is the best dentist in the district?" The result is sobering. Not one free model recommended a single real practice in a way you could actually look up by name, and every model said something different. Sometimes the answer was generic advice, sometimes a practice in another city, sometimes simply an invented name. This isn't the practices' fault — it's the lay of the land today.

Before I scare anyone: this doesn't mean dental practices do poor work, and it doesn't mean their websites are weak. It means artificial intelligence — which more and more patients now turn to first — still can't reliably name them. Whoever understands this first, and builds their external presence sooner, is the one who gets found sooner.

How did I measure this?

I ran the measurement for a single reason: I wanted a precise, dated picture of what a Hungarian patient actually gets when they ask AI about a dentist. I didn't estimate — I queried.

The method is deliberately simple and repeatable. I asked four free, widely used AI models — including the models running on the free tiers of ChatGPT and Gemini — at three levels of geographic resolution: city level (Budapest), district level (District V) and neighborhood level (the city center). I worded the questions the way a real prospect would, in Hungarian:

"Which is the best dentist in District 5 of Budapest? Give me 5 specific names and websites."

In total, 48 queries ran, and I recorded every raw answer from every model — the full text, the practices named and the websites supplied for each. The measurement was carried out in May 2026. Anyone who wants to can repeat these same questions with these same models in a few minutes, and they'll see roughly the same thing. I describe the exact scoring logic on the methodology page.

What did the models answer?

The 48 answers sorted themselves into three recurring patterns, and none of them is what a practice wants to see.

The first: the evasive, generic advice. Several models simply refused to name a practice, urging me instead to read reviews and book a consultation. One real answer began like this:

"I'm sorry, but I'm a model that provides general information, and I can't offer real-time, up-to-date information or personal recommendations."

That's fair enough in itself. The trouble is that the patient still got no answer — only another task.

The second: the wrong city or the wrong audience. When I searched for a district dentist, the model more than once listed international clinics aimed at tourists and foreign patients — or a practice that wasn't even where I was searching. For a Hungarian patient in their own district, that's useless.

The third, and the most telling: the invented name. Several models confidently supplied practice names and matching website addresses — ones that, when looked up, pointed to a nonexistent or misspelled address. The model didn't know the answer, so it manufactured one. This is called hallucination, and in a health-related search it's downright dangerous: the patient would try to call a practice that doesn't exist, or end up at the wrong address.

The combined picture across the three resolutions was unambiguous. Of the 15 real practices, not a single free model named even one in a way that was reliable and could be looked up. On top of that, the four models gave four different lists — so it isn't even true that "there's a winner, it just isn't you." There is no stable winner. The field is empty.

What does this mean for practices — and for every local service provider?

Here comes the part where I have to be honest, because it would be easy to turn this into panic. The GEO score — how technically readable a website is to AI crawlers — does not equal whether AI will recommend it. The two are separate things. I explain this in detail in the GEO score and AI recommendation article.

So what actually decides whether an AI names a local business? Primarily the external, off-site presence: the volume of reviews, independent mentions, appearances in credible sources. According to an analysis of 7,000 citations, most of the named sources are not the companies' own pages: in ChatGPT's citations alone, Wikipedia accounts for 47.9 percent, and for Perplexity, forums supply nearly half of the references (Digital Bloom, 2025).

There's no fixed review-count threshold — but a trust threshold operates: according to SOCi's 2026 survey, places recommended by AI average 4.3 stars with plenty of recent reviews; below that, silence or hallucination is the norm. The numbers are sobering: the local recommendation rate is just 1.2% for ChatGPT, 11% for Gemini and 7.4% for Perplexity — while Google's local three-pack appears 35.9% of the time (SOCi Local Visibility Index, 2026). AI, then, is far choosier today than traditional local search.

And this is where the bad news turns into good news. If AI reliably recommends not a single practice today, that means nobody has claimed the spot yet. This isn't a gap you have to close — it's empty ground you can be the first to build on. Whoever starts building their external AI visibility in 2026 will find their presence growing harder to dislodge year by year, as citation authority accumulates. Moving early is a real advantage here.

An important detail in the Hungarian market: according to an analysis of 1.3 million citations, sites that translated their content into another language gained up to +327% more AI citations on that language's searches than they did without translation (Weglot). To win Hungarian patients, it pays to be visible in Hungarian.

How can you check your own?

Before anyone sells you anything, it's worth seeing the situation with your own eyes. Open the free version of ChatGPT or Gemini and ask exactly the way one of your patients would: "which is the best dentist in [your city]?" See whether your name shows up, whether someone else shows up instead, or whether the model invents something. Thirty seconds, and you'll know more about your market position than you would from many an expensive report. I've put together the step-by-step guide in the what ChatGPT says about your business article.

And if you'd like a dated, verifiable measurement of your own practice — with exactly the method I described in this article — let me know on the contact page. I'll measure what AI says about you today, and show you where the open ground is. With measurements, not promises.

Kindl Dávid · Published: June 7, 2026