You can do this yourself in a few minutes per engine: open a fresh ChatGPT session (logged out or an unpersonalized temporary chat), turn on web search, and ask like a patient — "best cosmetic dentist in [your city]." Note whether it names you, a competitor, or no one, then run the same question through Perplexity, Gemini and Google AI Overviews, and again over a few days, because AI answers shift. This page is the step-by-step guide; if you would rather have the scan run for you instead, use the AI visibility checker.
How do I check what ChatGPT says about my practice?
Open a clean ChatGPT session with web search turned on, then ask a patient-style question such as "who is the best cosmetic dentist in [your city]?" Read which practices it names, save the answer with the date, and run it a few more times. The goal is to see your practice the way a prospective patient sees it, not the way you already know it.
Follow these steps in order. Each one removes a way the test could quietly mislead you.
Current controls: OpenAI Temporary Chat guidance and ChatGPT Search guidance. OpenAI describes search partners rather than one fixed Bing-only pipeline.
- Start a clean session. Log out, or select Unpersonalized before starting a temporary chat. A personalized temporary chat can still use memories and custom instructions. This reduces account personalization; it does not reproduce every patient's location or context.
- Turn on web search. ChatGPT can search automatically. For this test, choose Search from the tools menu before sending the query.
- Ask like a patient, not an owner. Use the phrases patients actually type, such as "best cosmetic dentist in [city]." Do not search your own practice name yet; that only tells you it exists, not whether you get recommended.
- Read who gets named. Note whether the answer lists your practice, competitors, directories like Yelp or Healthgrades, or says it can't recommend a specific dentist.
- Save the test conditions. Keep the exact question, date, engine or model shown, city, search setting and account personalization setting with your screenshot. Open cited pages to verify the practice, address and services. Record a practice-name mention separately from a citation to your own website.
- Repeat. As a starting check, run each query three to five times across different days. This is a practical suggestion, not a validated minimum for a reliable trend.
If you'd rather not do this by hand, our free AI visibility checker runs these queries across ChatGPT, Perplexity, Gemini and Google AI Overviews for you and returns a single read in about two minutes.
Which queries should you test?
Test the phrases patients actually use, not your practice name. Patients describe a need ("Invisalign dentist near me"), a location, and sometimes a constraint like insurance or Saturday hours. Run one query per intent, swap in your city and neighborhood, and check whether a real result would send that patient to you.
These seven examples cover different patient needs; they are not a measured ranking of what brings in most patients. Replace the bracket with your city, then run the neighborhood version too.
| Patient intent | Example query to test | A strong result looks like |
|---|---|---|
| Cosmetic / veneers | "best cosmetic dentist in [city]" | Your practice named in the shortlist |
| Invisalign / aligners | "Invisalign dentist near me in [city]" | You listed among the Invisalign providers |
| Implants | "dental implant specialist in [city]" | You named, with your specialty noted |
| Availability | "[city] dentist open Saturday" | You appear when hours match the request |
| New patient | "good family dentist in [city] accepting new patients" | You listed as open to new patients |
| Insurance | "dentist in [city] that takes [insurance]" | You named for the plans you accept |
| Brand check | "is [your practice] any good?" | An accurate summary of your reviews and services |
The brand check is the one query where using your name is the point. Use it to check what an AI tells someone who has already heard of your practice, separately from an unbranded recommendation test.
Why isn't one test enough?
One run is a snapshot, not the weather. AI answers are rebuilt each time and pull from a shifting set of sources, so the same question can name you today and skip you next week. According to a 2025 Profound analysis of about 80,000 prompts per engine, roughly 40–60% of the domains in July responses had not appeared in June responses to the same prompt set. This measures newly appearing domains across the tested responses, not the chance that a particular practice disappears. Judge your visibility by how often you appear, not by a single result.
Here is the share of July cited domains absent from June, by engine, in that study. The takeaway is not the exact figure but the size of the swing: checking once and stopping is like reading one day's weather and calling it the climate.
| AI engine | July cited domains absent from June |
|---|---|
| Google AI Overviews | ≈59% |
| ChatGPT | ≈54% |
| Perplexity | ≈41% |
Source: Profound domain-drift analysis, 2025 (≈80,000 prompts per engine, June–July). tryprofound.com
Start with three to five checks over a couple of weeks, keeping the query and test conditions fixed. Treat that as an initial record, not a statistical confidence threshold. Do not infer repeat consistency from a single city-scan snapshot. You can see the method behind that in how we measure AI visibility.
The reference material behind Citevio's checks is released for independent review. Citevio's dental market research is published openly under CC-BY-4.0 with a DOI, so the numbers can be checked and reused; client data is never published.
How do you check Perplexity and Gemini too?
Check ChatGPT, Perplexity, Gemini and Google AI Overviews separately, using the same patient-style question. Record who gets named and which sources are shown; a Google map-pack result is a separate comparison, not an AI Overview.
The table pairs practical checks with SOCi's reported results from a 2026 study of nearly 350,000 locations belonging to multi-location brands. These are not dental-practice probabilities, and the local 3-pack figure measures a different search surface. Do not use them to predict your own result.
| Engine | What to check in your test | Reported SOCi study measure |
|---|---|---|
| ChatGPT | Select Search, then inspect the cited pages and named practices | 1.2% |
| Perplexity | Inspect the answer and open the cited pages | 7.4% |
| Gemini | Record the Gemini answer separately from Google AI Overviews | 11% |
| Google local 3-pack (for comparison) | Record the local pack separately from the AI answer | 35.9% |
Sources: recommendation rates from SOCi's 2026 Local Visibility Index, ~350,000 locations across 2,751 brands (SOCi primary source; Search Engine Land); Damian Rollison's 2025 account of ChatGPT local search, based partly on asking ChatGPT to explain itself, not an official specification (searchengineland.com).
Citevio's own archived market evidence is identified by DOI 10.5281/zenodo.23143214.
Keep each engine's result separate; do not use a weak result in one as the result for all of them. That is the whole reason to check all four rather than assume ChatGPT speaks for every AI a patient might use.
When does ChatGPT look at the web, and which crawler has to be let in?
The same question can be answered two different ways: from a live search run at the moment you ask, or from what the model already carries. For you those are not the same event. A web search does not necessarily visit your own website for every answer. OpenAI recommends allowing OAI-SearchBot and its published IP addresses for search eligibility; access does not guarantee a mention.
OpenAI publishes which crawler does what, and the distinction is sharper than most site owners realise. These are separate identities with separate purposes:
| Crawler | What OpenAI says it is for | Refused at the server, 22 July 2026 |
|---|---|---|
| OAI-SearchBot | "used to surface websites in search results in ChatGPT's search features" | 17 of 499 (3.4%) |
| GPTBot | "used to make our generative AI foundation models more useful and safe" | 40 of 499 (8.0%) |
| ChatGPT-User | "for certain user actions in ChatGPT and Custom GPTs" | 7 of 499 (1.4%) |
OpenAI is explicit that these settings do not travel together: "Each setting is independent of the others – for example, a webmaster can allow OAI-SearchBot in order to appear in search results while disallowing GPTBot to indicate that crawled content should not be used for training OpenAI's generative AI foundation models." For a practice, that cuts both ways. A decision to keep your content out of model training does not have to cost you search visibility — but a blanket block applied without reading the difference does exactly that, and the site owner rarely knows it happened. OpenAI's documentation now also lists a fourth identity, OAI-AdsBot, used to check pages submitted as ads; it is not a search crawler and is listed here only so the name is not mistaken for one.
Limits on the refusal figures above: one day's measurement, 22 July 2026, and hosting or CDN rules change without notice, so a repeat on another date will not reproduce them. They measure access, not citation — removing a refusal was never measured to produce a mention. Which layer produced the refusal (host, security product, CDN, plugin or a human decision) was not recorded. They come from a different instrument than the robots.txt file study of 6,497 domains, which reads what a file declares rather than what a server does; the two are not interchangeable.
Can you tell from the app which path an answer took? Often not with certainty. If the reply shows sources, the web was consulted for it. If it shows none, the web may not have been consulted — which is different from proof that it was not. That is one more reason a single answer is a weak piece of evidence, and why the checks on this page ask you to run the same question more than once.
Checking your own side of it takes minutes rather than a project. The step-by-step version is on how to test if AI can read your dental website; the pattern where a file says yes and a server says no is written up on robots.txt open but AI bots blocked; and how common blocking is across dental sites is on which dental websites block AI crawlers.
For the same refusal pattern explained on video, with the three-identity check demonstrated live, see is my dental website blocking ChatGPT?
What if your practice doesn't show up?
If ChatGPT names competitors or no one, first check the test conditions, crawler access and the accuracy of the cited business details. A missing mention does not identify its cause. SOCi reported an average of around 4.3★ for locations ChatGPT recommended; that average is not an eligibility cutoff or evidence that a lower rating caused your absence.
Check access and business-information consistency, then inspect the reviews actually cited. For the full sequence, read how to get your dental practice recommended by ChatGPT. And if being invisible feels like your problem alone, it isn't: across nearly 350,000 locations, ChatGPT recommended only 1.2% of them (SOCi, 2026), that is a multi-location-brand finding, not a measured share of dental practices or a promise of an easy improvement. For a local, dentist-specific breakdown, see our AI visibility report for Charlotte dentists.
The site-side comparison comes from a separate Citevio study of 527 US dental practice websites across seven metros plus nine nearby towns, read between 6 and 26 June 2026.