Here are the forty questions we put to ChatGPT, Gemini, Perplexity and Claude every week, grouped by what the patient is trying to find out, with the exact wording we use. Copy them, change the city and the practice name, leave everything else identical, and you have the same measurement set we do.

For the monthly run, the twenty-question working set in how to measure AI visibility for your dental practice is the one to score. The forty below is the catalogue to choose from.

BEFORE
Which dentist near me takes my insurance?
AI assistant

Well-reviewed dental practices nearby:

1
2
3
AFTER
Which dentist near me takes my insurance?
AI assistant

Well-reviewed dental practices nearby:

1
Your Practice NameYour practice
★★★★★ 4.9 · 212 reviews
Gentle care, accepting new patients
2
3
Illustrative example. Other practices are redacted. Real answers vary by engine, city and day.

Why The Wording Of The Question Matters More Than The Model

Re-running a reworded question measures a different question. "Who is the best dentist in [CITY]?" and "Which dental practice in [CITY] would you recommend?" pull from different sources, produce different lists and disagree with each other. You cannot compare the two, and neither one tells you whether anything has changed since last month.

So the wording is fixed and we refuse to adjust it in the middle of a programme. The list is versioned like code: a change creates a new version, the old version's results stay in the file, and both are reported side by side for a quarter before the new one takes over. The reason is simple. A scoring sheet that quietly changes underneath you measures the sheet, not the market.

Engines do not read the same web either. OpenAI documents the crawlers it uses, and Perplexity and Anthropic publish their own crawler guidance, which is part of why the same question returns a different set of practices from each of them. OpenAI's crawler documentation, Perplexity's crawler guidance and Anthropic's crawler policy set out what each of them can and cannot read.

And keep a copy of every answer. When a score moves, the only question that matters is what the model actually said that time.

The 40 Questions, Grouped By Intent

Placeholders used below: [CITY] is the city you serve, [NEIGHBOURHOOD] is one neighbourhood inside it, and [PRACTICE] is your practice name for the reverse test. Everything else is fixed, including punctuation and the second sentence of the prompt.

I'm in [CITY] and I need [NEED].

Which dental practices should I consider? Name them and say why.

Each question below is the [NEED] half of that template. Forty questions, five groups of eight, in the order we run them.

Group a: Emergency

These test whether you can be found when somebody is in pain today. They are the most valuable answers and the most volatile, so they are worth scoring carefully.

Swipe the table sideways to see every column.

No.Exact Wording To Use
1Who can I see tonight near [CITY] for tooth pain?
2Which dentists in [CITY] treat dental emergencies on weekends?
3Is there an emergency dentist near me in [CITY]?
4What should I do if I knocked out a tooth?
5Who is open late for tooth pain in [CITY]?
6Do any dentists in [CITY] do same-day emergency appointments?
7How do I find an urgent dentist when I have facial swelling?
8Which dental clinic in [CITY] is open right now?

Group B: Treatment and Price

These test whether you can be quoted for the work you want to sell. None of them asks for a price alone; each asks what the treatment involves and what it costs.

Swipe the table sideways to see every column.

No.Exact Wording To Use
9How much does a dental implant cost in [CITY]?
10What does a root canal cost in [CITY]?
11How much do clear aligners cost in [CITY]?
12Which dentists in [CITY] do dental implants, and what do they cost?
13What do dental crowns cost in [CITY]?
14How much do veneers cost in [CITY]?
15Is it cheaper to get a dental implant in [CITY] or abroad?
16What does a first exam and cleaning cost in [CITY]?

Group C: Availability and Insurance

These decide whether a patient believes they can actually be seen, and actually be covered. They are mostly answered from directory and profile data.

Swipe the table sideways to see every column.

No.Exact Wording To Use
17Which dentists in [CITY] take my insurance?
18Which dentists in [CITY] are open on Saturdays?
19Do any dentists in [CITY] offer evening appointments?
20Which dental practices in [CITY] take Medicaid?
21How do I find an in-network dentist in [CITY]?
22Which dentists in [CITY] offer payment plans?
23Which dentists in [CITY] speak Spanish?
24Which dentists in [CITY] are taking new patients?

Group D: Family and Cosmetic

These test whether you are found for a person rather than a problem, and they usually return longer lists than the emergency group.

Swipe the table sideways to see every column.

No.Exact Wording To Use
25Which dentist in [CITY] is good with children?
26Who is a good pediatric dentist in [CITY]?
27Which dentists in [CITY] see children and parents together?
28Who in [CITY] does cosmetic dentistry such as whitening and veneers?
29Which dentist in [CITY] has the strongest reviews for implants?
30Where in [CITY] can I get my teeth whitened?
31Which dentists in [CITY] treat nervous patients?
32Which dentists in [CITY] do aligners for teenagers?

Group E: Local and Neighbourhood

These are the open questions. The models usually answer them generically, which is exactly why they are the most valuable ground of the five groups.

Swipe the table sideways to see every column.

No.Exact Wording To Use
33Who is the best dentist in [CITY]?
34Which dental practice near [NEIGHBOURHOOD] is worth visiting?
35What is the best family dental clinic in [CITY]?
36Which dentists in [CITY] do local residents recommend?
37Where should I look for a dentist near [NEIGHBOURHOOD] in [CITY]?
38Which dental office in [CITY] has the friendliest staff?
39What should I look for when choosing a dentist in [CITY]?
40Who does [CITY] recommend for routine dental care?

How To Adapt The List To Your Practice

Three substitutions and nothing else.

  1. The city. Use the city patients actually search, not the whole service area. One city per run, never two.
  2. The neighbourhood. One per location you serve, replacing the placeholder in the questions that use it.
  3. The treatments. Replace any question about a service you do not offer with one you do offer, and keep the count at forty so the series stays comparable.

Leave the sentence structure alone. If you are tracking change over time, the prompt is the instrument, and you cannot swap the instrument in the middle of a run.

The reverse test is one extra prompt, kept on its own row: Is [PRACTICE] a good dentist in [CITY]? It tells you whether you are found by name rather than by category, and it must not be counted in the forty.

What The List Tells You That a Keyword List Cannot

Three things a keyword list cannot show you.

Intent coverage. Check that every group has at least one question you win. If price is all losses, that is a content problem you can see and fix. On a keyword list it stays invisible, because the keyword sits on somebody else's site and not on yours.

The open questions. Group E mostly returns nobody, two practices or three. These are the ones where being named is worth more than a win in a crowded list, because the model is not yet strongly committed to a name.

Where an honest range wins. Price questions get answered with a range or not at all. If your page states what a first visit includes and gives a range marked clearly as an estimate that varies, you become quotable where most practices are not. Our guide to cost questions in AI answers covers what to publish and how to frame it.

None of this is clinical advice and none of it should be read as it. For advice about your own treatment or what it will cost you, ask your own dentist.

The Rules That Keep The Data Honest

  • Same window. Every engine, same day, within a couple of hours of each other.
  • Same number of runs. Three per question per engine, every time, including the weeks you are busy and skip.
  • Record failures. Refusals, wrong city answers and empty responses are logged as failures, not deleted.
  • No cherry-picking. If you re-run a question and keep the better answer, you have stopped measuring and started shopping.
  • Log every edit to the list with the date and the reason, so anybody reading a number can tell which series it belongs to.

What To Do With The Answers

Four states, four actions.

Swipe the table sideways to see every column.

What You SeeWhat It MeansWhat You Do Next
Named and recommendedYou own that answer for this runNote the reason the model gave and check the page it pointed at still says so
Named, not recommendedYou are in the set but not the pickFind what the winner had that you did not: a page, a review volume, a fact
Absent, competitors namedThe gap is specificFind the source the engine used and build the page that answers the question
Nobody namedThe question is openWrite the page that would deserve the answer, and keep the question in the set

Then run the same forty questions again next week with the same wording and compare. The measurement guide covers the scoring, and our guide to emergency answers in AI engines goes deeper on group A, where most practices lose the most ground.

If you would rather not run this by hand, the visibility snapshot shows where you stand, the article index here covers the rest of the method, and the case studies show what happened after the fixes were made.