Six pillars, ten markets, 1,200 answers. We expected the big clinics to own AI visibility. Instead we found nobody does — and that the three main AI tools barely agree on anything.
Last month we scanned thirty GTA wellness websites and found that not one met Google’s speed threshold. This is the second edition of that research, and it asks a different question — the one clinic owners have actually started asking us.
When a patient opens ChatGPT and types “who are the best dentists in Mississauga,” does your name come up?
We ran that question, and its med spa equivalent, 1,200 times across five GTA cities and two verticals. Then we took the 121 clinics that Google Maps ranks in its top twelve for those exact searches, and checked how often AI named each one.
The short answer: almost never, and almost never the same way twice.
Nobody is winning
Here is every clinic in the study, sorted by how often AI search named it in its own market across 120 answers.
| How often AI named them | Clinics |
|---|---|
| Nearly every time (90%+) | 0 |
| Usually (60–89%) | 3 |
| Sometimes (20–59%) | 18 |
| Rarely (1–19%) | 63 |
| Never — not once in 120 answers | 37 |
Not a single clinic out of 121 appears in even nine of ten answers about its own city.
Thirty-one percent were never named once. Not in a hundred and twenty tries. Every one of them ranks in Google’s own top twelve for the exact question being asked.
Sixty-three clinics sit in the “rarely” band — they exist in AI’s world, technically, in the way a lottery ticket technically wins.
The category is wide open, and no one in the GTA has taken it.
The clinics with a thousand reviews that AI has never heard of
We expected review count to explain everything. It is the number every clinic owner watches, and the number Google Business Profile makes most visible.
It matters — but nowhere near as much as you would think. Clinics that AI named at least once have a median of 398 Google reviews. Clinics never named have a median of 185. So reviews do separate the groups.
Then look at who was never named at all:
| Clinic | Google reviews | Maps rank |
|---|---|---|
| Parkview Family Dental Clinic | 1,300 | 8 |
| Smiles Dentistry | 1,100 | 5 |
| Zenbar Healing Studio | 896 | 5 |
| Dentistry For You – Woodbridge | 892 | 3 |
| Atlas Dental | 868 | 7 |
| City Dental on Bay | 653 | 12 |
A practice with 1,300 reviews, ranking eighth in Google Maps, appeared in zero of 120 AI answers about its own market. Another sits at Maps position three with 892 reviews and is equally invisible.
If reviews were the answer, these clinics would be everywhere. They are nowhere. Something else is running this.
Google’s own top twelve barely matters
This is the finding that surprised us most, and the one worth sitting with.
We counted every business named across all 1,200 answers — 12,351 individual recommendations, after stripping out treatment names, ratings and other noise, and merging the many spellings of each business into single entities.
Of those recommendations, 13% went to the 121 clinics Google Maps ranks in its top twelve. 83% went to businesses outside that list entirely. The remaining 4% went to large multi-location chains.
Five times out of six, AI search names a business that Google’s own local results do not rank in the top twelve for that exact query.
Ranking in Google Maps and being recommended by AI are close to unrelated problems. And in the case of AI Mode, that is Google’s own AI, drawing on Google’s own business data.
The practical consequence is uncomfortable. Every hour you have spent climbing the local pack bought you something real — but it did not buy you AI visibility.
The three AIs are describing three different cities
We ran the same questions through ChatGPT and Perplexity, 60 times each per market, then compared who they named. They agree on 7% of it.
| Market | Both named | Total named |
|---|---|---|
| Toronto med spa | 14 | 465 |
| Mississauga med spa | 12 | 304 |
| Markham dental | 25 | 236 |
| Oakville dental | 24 | 214 |
Seventy percent of the clinics AI named at all — 59 of 84 — appear on only one of the two engines. Body and Soul Day Spa was named in 60 of 60 Perplexity answers and 0 of 60 ChatGPT answers.
The reason shows up in where they look. ChatGPT leans on directories and professional bodies — Opencare, RateMDs, the Royal College of Dental Surgeons of Ontario, ThreeBestRated, Birdeye. Perplexity leans on niche aggregators and booking platforms — MedSpaFind, Fresha, TrustAnalytica, WhatClinic. Google AI Mode leans on Google Business Profile: star ratings, review counts and opening hours rendered straight into the answer.
Reddit is the only major source all of them share.
AI visibility is not one job. It is three jobs — three different source ecosystems, and work that lands you in one may do nothing for the others.
The version number changes the answer too
We ran one market twice on ChatGPT — sixty answers on one model version, sixty on the next. The two versions agree on 13% of the businesses they name. Between them they named 300 distinct clinics for the same city, and shared 40.
The older version searched the web roughly once per question. The newer one searched nearly five times. It looked harder, went deeper, and came back with a substantially different list.
Neither is wrong. They are different tools giving different answers to an identical question — and your patient has no idea which one they are holding.
Ask the same question twice, get two different answers
We also ran Google’s AI Mode on one market — Mississauga dental — ten times in a row, same question, same day. Ten runs produced ten different lists. Not one pair matched.
Nine distinct clinics appeared across those ten answers. Two showed up every single time. The rest rotated:
| Clinic | Named in |
|---|---|
| City Centre Dentistry | 10 of 10 |
| Southcommon Dental | 10 of 10 |
| Dr Arun Narang & Associates | 6 of 10 |
| Mississauga Smiles Dentistry | 5 of 10 |
| Dr Sahar Dentistry | 5 of 10 |
| Winston Churchill Dental | 5 of 10 |
| Mississauga Dental Arts | 3 of 10 |
| Urban Dental Centre | 1 of 10 |
| Churchill Smiles Dentistry | 1 of 10 |
Four of those nine are not in Google Maps’ top twelve for the query at all.
A clinic sitting at five-out-of-ten is not “moderately visible.” It is a coin flip, run fresh every time a patient asks.
Who is actually writing these answers
Here is the part that changes what you should do on Monday.
Every single Perplexity answer we collected — all 600 — cited at least one clinic’s own website. On average, thirteen of them per answer. Four ChatGPT answers in five did the same.
In one Oakville search, the top-cited source was a clinic’s own blog post titled “best dentists in Oakville.” That clinic then appeared first in the recommendation.
It wrote the list. The AI read the list. The AI repeated the list, with the author at the top..
We also watched the reverse. In one Google AI Mode answer, a clinic’s website was cited as a source for claims about the Mississauga market — and the recommendation went to four other clinics. It fed the answer and got nothing.
The businesses shaping these answers are not always the ones benefiting from them.
What we got wrong
Before collecting anything, we wrote down what we expected. Three of those predictions were wrong, and the errors are more useful than the hits.
We thought review count would dominate. It matters, but a clinic with 1,300 reviews can be completely invisible. Reviews are necessary and nowhere near sufficient.
We thought the chains would be eating this. They are not, as a category — around 4% of recommendations. But the two most-named businesses in the entire study are chains: Skin Vitality (fourteen Ontario clinics) and Canada MedLaser (fifteen GTA locations). Two giants at the summit, a very long tail underneath, and almost nothing in between.
We thought model choice was a technical detail. We nearly skipped testing it to save money. It turned out to change the answer more than almost any other variable we measured.
What to do this quarter
Work out which AI your patients actually use, and start there. Given a 7% overlap between engines, spreading effort evenly across all of them is close to spreading it nowhere.
Get into the sources your engine reads. For ChatGPT that means Opencare, RateMDs, ThreeBestRated and your RCDSO listing being complete and current. For Perplexity it means MedSpaFind, Fresha, WhatClinic and booking platforms. These are unglamorous directory listings, and they are doing more work than your homepage.
Keep the reviews coming, but stop treating them as the whole strategy. They move you from invisible to occasionally visible. They will not make you reliable.
Write the comparison piece for your own market. The clinics being cited most are the ones that published a genuinely useful “best in the city” resource. Write a bad one and it does nothing; write the best one in your city and you are in the answer.
Stop reading a single AI result as a verdict. If a competitor shows up when you search once, that is one draw from a lottery. Ask ten times before you conclude anything about your position — or ours.
Method and limitations
We defined ten markets: dental and medical aesthetics, across Toronto, Mississauga, Vaughan, Markham and Oakville. For each we took the top twelve businesses Google returns for the market query — 121 clinics after excluding chains, businesses with no website, and listings in the wrong city.
We then asked each engine one question per market, in the form “Who are the best dentists in Mississauga, Ontario?” — substituting the vertical and city for each of the ten markets. Sixty times per market on ChatGPT, sixty on Perplexity, for 1,200 answers total. Every raw response was saved verbatim. Runs were randomized across markets so a mid-collection model update could not land entirely on one city. All collection happened inside one week in September 2026.
Business names were extracted as written, then merged into single entities across spelling variants, with the borderline matches adjudicated by hand rather than by similarity score.
Limitations, plainly. Sixty runs per engine per market is the floor of what published research suggests is needed for a stable visibility measurement, not a comfortable margin. Our Google AI Mode figures cover one market with ten runs — they are directional and we have labelled them as such rather than presenting them as rates. API queries carry no location signal, so each named its city explicitly; we confirmed this returns correctly-located businesses but it is not identical to a patient searching from their phone. Everything here is association, not cause — we can say what named clinics have in common, not what would happen if you changed it. And these engines update constantly. This is a photograph, which is precisely why we are running it again next quarter.
Download the full dataset (CSV, 121 clinics)
Frequently asked questions
Does my Google Maps ranking help me in AI search?
Barely. Only 13% of the recommendations we recorded went to businesses in Google’s top twelve for that same query. The two are close to unrelated, even within Google’s own products.
How many reviews do I need to be recommended by AI?
There is no threshold. Clinics that got named have a median of 398 reviews versus 185 for those never named, so it matters — but a clinic with 1,300 reviews ranking eighth in Maps appeared in none of our 120 answers about its market. Reviews are a floor, not a strategy.
Why does AI give me a different answer every time I ask?
Because it does. In ten consecutive identical searches, we got ten different lists with no two matching. This is normal behaviour for these tools, not a fault, and it is why a single search tells you almost nothing.
Should I focus on ChatGPT, Perplexity, or Google?
Whichever your patients use, because they draw on almost entirely separate sources and agree on only 7% of who they name. ChatGPT reads directories and professional registries; Perplexity reads booking platforms and niche aggregators; Google AI Mode reads Google Business Profile.
Is it worth writing a “best clinics in my city” article?
Our data says yes, with a caveat. Clinic-owned websites were cited in every Perplexity answer we collected. But being cited as a source and being recommended are different outcomes — we watched one clinic’s site inform an answer that recommended five other practices. Write the genuinely best resource in your market, not a thin one with your name at the top.
Next quarter we run this again — same ten markets, same questions, more runs, and a repeat pass so we can say whether any of this actually moved.
If you would rather know what your own clinic’s visibility looks like than the sector’s, that is what our Digital Growth Assessment measures.
