Dental patients crossing a border are close to a best case for assistant use. The purchase is expensive, the stakes feel high, the patient usually knows nobody who has done it, and the questions are exactly the kind people are now comfortable asking a chatbot instead of a forum.
Which means the answers matter more here, sooner, than in most categories.
This is a checklist rather than an argument. Fourteen checks, ordered by how much they move the outcome. The order is the useful part — the first four outweigh the rest combined.
First: get a baseline
1. Ask the ten questions your patients actually ask. Not best dentist in [city]. The real ones: is it safe, what does a crown cost there, how do I check credentials, what happens if something goes wrong afterwards, how long do I need to stay. Ask each across ChatGPT, Claude, Gemini and Perplexity.
2. Ask each one three times, in fresh sessions. This is the step people skip, and skipping it invalidates everything after. In our Tijuana run the top recommendation changed between repeats in 159 of 160 cases. One ask tells you essentially nothing about your position.
3. Write down the citations, not just the names. Which pages did the assistant read? After thirty answers a short list of five to ten pages will repeat. That list is your actual territory, and it is almost never the list a clinic would have guessed.
4. Note which questions you appear in and which you do not. Visibility is question-shaped. Clinics are often named on safety questions and absent from which clinic should I choose — or the reverse. The gap tells you what to fix.
The sources layer
This is where the outcome is mostly decided.
5. Check your listing on every page from step 3. Not whether you are listed — whether the details are right. Old phone number, previous address, wrong specialty, hours from before the remodel. Correcting a listing on a page that already gets cited is the highest-return single action available to you.
6. Verify the professional register entries. Dental association membership, licensing body records, specialty board listings. These carry unusual weight because they are dense, consistent, and read as legitimacy. Many clinics are members and have never looked at their own public entry.
7. Find the "best clinics in [city]" articles and read them. They exist, they are frequently years old, and they get cited far past their expiry. If you are missing from one, contact the author. This works more often than clinics expect, largely because nobody asks.
8. Check the medical tourism aggregators specifically. Some are cited constantly in cross-border questions; others never appear despite being large. Determine which is which before spending anything.
9. Look at where the public conversation happens. Reddit threads, patient forums, review platforms. For cross-border dental this is disproportionately influential, because a thread where five patients describe real experiences is high-quality material for a model. Note that closed groups do not exist for this purpose — only public pages count.
The entity layer
10. Standardise your clinic name to one exact form. Including accents, abbreviations, and whether "Dental" comes before or after. If you appear as four variants, a model has four weak entities instead of one strong one.
11. Standardise the address and phone the same way. One form, everywhere, including the format of the street number and the country code.
12. Standardise how you describe the specialty. Clinic, center, practice, dental group used interchangeably across listings means competing against yourself for the word patients actually use.
The site layer
Lower leverage than the two above, and still worth doing.
13. Put the facts in text, not images. Address, phone, hours, credentials, languages spoken, insurance and payment accepted. A phone number inside a JPEG does not exist to a crawler. Add an FAQ written the way patients ask, including the uncomfortable questions about complications and follow-up.
14. Serve the language your patients ask in. If they ask in English, you need English pages at their own URLs — a real page, not a translation widget. A Spanish-only clinic site competing for American patients is absent from the conversation deciding its market.
What not to bother with
Hidden instructions for the AI on your pages. Paying anyone for "ChatGPT placement." Repeating best dentist in [city] through your copy. The first two do nothing and can hurt; the third is an SEO habit that transfers badly.
The realistic expectation
Fixing listings can show up within weeks, because assistants read those pages live. Getting into a comparison article takes as long as the writing takes. Anything sitting in training data does not get corrected on your timetable — the practical remedy is to make the live-search layer strong enough to outweigh it.
And the honest ceiling: in a market as fragmented as the Tijuana data showed — 61 clinics named across 440 answers, none reliably first — the goal is not to own the answer. Nobody owns it. The goal is to be in the set that gets named, consistently enough that a patient asking three times sees you at least once.
If you want your own numbers rather than the market's — how often you were named, on which assistants, and which sources put competitors ahead of you — that is the audit.