Conflict-of-interest disclosure
This study was conducted by NEXTIWEB, a Montreal web agency that sells the very services its findings make necessary. We state this upfront because it is the first question we should be asked. The next section explains how the clinics were drawn, before a single figure.
Why a web agency ran this study
No one had measured it. There is a great deal of talk about dental clinics' visibility — and a great deal sold to them — yet no public record existed to rely on. I wanted to know what was actually true, on clinics drawn at random rather than hand-picked.
I am publishing what I found, including what does not flatter me: a criterion I had to invalidate along the way, a measurement that separates poorly near its threshold, and the fact that on some sites the responsibility does not belong to the clinic.
This document asks for nothing. There is no form, no offer, and no clinic is named.
How the clinics were drawn
The clinics were not chosen. They were calculated. The starting point is the official directory of the Ordre des dentistes du Québec, list by list: six municipalities, then the administrative regions. Each list is sorted by last name.
In a list of N members, N is divided by the number of clinics wanted. The result is a step. The ranks drawn are that step, then its double, its triple, and so on. These are position numbers. They are calculated before any name has been seen. We then read who occupies that rank and keep their main practice clinic.
Example: the Longueuil list
79 members listed, 5 clinics to draw. 79 divided by 5 gives a step of 15. The ranks are therefore 15, 30, 45, 60 and 75, fixed before opening the list.
Two ranks did not qualify. Rank 45 was a specialist, so we take 46. Ranks 75 and 76 were a public institution and then another specialist, so we take 77. Each substitution is recorded with its reason.
None of these five clinics was chosen by anyone. They occupied a rank.
The grounds for exclusion are set in advance: specialist, public institution, duplicate, or a member whose main place of practice lies outside the target territory. A discarded rank is replaced by the next one, and the substitution is logged.
No clinic came from a Google search. That is deliberate. Going through Google would have selected the clinics that already invest in their visibility, and the study would have measured its own bias instead of the market.
115 ranks were drawn in total. 100 clinics had a working website and make up the audited sample. 14 had no website at all. One had its server go down during the study and was handled separately.
A reader can redo the draw: same directory, same list, same step, same rank, same clinic.
A rank only means something with its list, its size and its date, because the directory is a living document. The Montreal list had 1,144 members on August 25, 2026 and 1,145 on the 28th. All three are recorded for every clinic drawn.
The finding
None of the 100 audited clinics reaches 8 out of 10. The best tops out at 7. The average stands at 3.8 out of 10.
This single number covers three very different situations. Confusing them would be unfair to the clinics, and inaccurate.
What amounts to neglect
Four criteria concern Law 25: publishing a privacy policy, naming a person responsible for the protection of personal information, offering a genuine cookie-refusal button, and informing the patient at the moment they enter their data. None requires technical skill or a budget. These are obligations in force since 2023.
92% of clinics do not tell the patient what they do with their information at the moment it is entered. The form asks for a name, a phone number, sometimes a date of birth or a reason for the visit. Nothing explains where it goes. This is the most serious finding of the study: it touches a legal obligation and health data, and it has no technical excuse.
What amounts to a choice
One clinic in two has not answered a single Google review in six months. Six in ten show opening hours on Google that do not match those on their own website. In one case, Google announces the clinic closed four days out of five when it is open.
Answering a review is free. Correcting your hours takes two minutes. No one is stopping it. The panel shows a median of 70 reviews and an average rating of 4.6 out of 5: these clinics are visible and well-liked, and half of them do not speak to the people who recommend them.
What comes down to circumstance
96% of clinics take more than three seconds to display their main content on a phone, and 86% have no valid structured data — the markup that lets a machine understand it is looking at a dental clinic, where it is located and when it is open.
Here the reproach carries less weight. A site's speed depends on a platform chosen years ago, often by a supplier. Google's criteria have changed since. And the conversational assistants that now read websites in place of patients did not exist when these sites were built.
These clinics are not behind out of neglect. They have been overtaken by a change no one announced to them.
What they have already done
94 clinics out of 100 offer online appointment booking reachable in three clicks. Fewer than one in ten offers a genuine calendar of availabilities. In almost every other case, the path ends with a form the clinic must process by hand: the patient proposes their availability, the clinic calls back.
The clinics made the gesture, not the system.
A number that appears in none of the four blocks
Of the 115 clinics drawn from the directory, 14 have no working website. They are not counted in this study's percentages, which cover only the 100 audited clinics. So one clinic in eight drawn at random has nothing to audit.
What the study measured
Ten binary criteria, of equal weight, defined and frozen before collection began. Each can be checked in a few minutes by anyone, on the clinic's public website or its Google listing.
| Criterion | Compliant | NON-comp. | 95% range |
|---|---|---|---|
| 1. Privacy policy published | 74% | 26% | 17 to 38% |
| 2. Privacy officer (RPRP) contact shown | 63% | 37% | 25 to 50% |
| 3. Cookie banner with a genuine refusal | 24% | 76% | 65 to 84% |
| 4. Collection notice at the point of entry | 8% | 92% | 84 to 96% |
| 5. Mobile load under 3 seconds | 4% | 96% | 90 to 99% |
| 6. Online booking in 3 clicks or fewer | 94% | 6% | 2 to 13% |
| 7. Real-time appointment availability | 6% | 94% | 87 to 98% |
| 8. Machine-readable structured data | 14% | 86% | 69 to 95% |
| 9. Answered a Google review within 6 months | 52% | 48% | 36 to 60% |
| 10. Google hours identical to the website | 39% | 61% | 50 to 71% |
Criteria 5 and 7 cover 99 and 98 clinics. One site could not be measured, and two booking paths require an identity before showing any availability.
Criterion 5 separates poorly near the threshold. A second reading of the four compliant clinics, nine days later, places them at 2.1 · 2.8 · 3.0 and 3.1 seconds. The first measurement was kept.
How to read the range
The percentages come from 100 clinics, not from every clinic in Quebec. The range says how much the result could move if we started over with 100 other clinics drawn the same way: the true rate would fall within this range about 95 times out of 100.
Example: 26% of clinics have no privacy policy, and the range runs from 17 to 38%. Redoing the study, we would almost always get a figure between those two bounds.
Why 95 and not 100. Total certainty would require announcing a range from 0 to 100%, which would say nothing. The more certainty you want, the wider the range grows. 95% is the usual convention, the one where the range stays narrow enough to be useful.
These ranges account for the sample size and for the fact that some clinics share the same site template.
Two findings behind the numbers
When a site speaks to machines, it is rarely by choice
A search engine, or an assistant like ChatGPT, does not read a site the way a patient does. It looks for markers placed in the page's code: the type of establishment, the address, the hours. Some sites add a file that summarizes the clinic for these machines.
Seventeen clinics in the panel have this summary file. Only one wrote it itself, by hand — the equivalent of about fifteen pages. The other sixteen received it without asking: their platform or an extension produced it on its own.
The markers placed in the code follow the same logic. Fourteen clinics have valid ones. Ten owe them to a group's or an agency's template.
When a clinic is machine-readable, it is almost always its supplier that made it readable, not the clinic.
The work is 80% done, and it doesn't count
An incomplete marker is no better than a missing one. Nine sites correctly declare their name, address and phone number, then forget the hours — the one thing a patient actually looks for. Another wrote the word "dentist" in lowercase where the capital is required: the marker is there, and the machine misses it.
Appointment booking follows the same path. The button exists almost everywhere, the calendar almost nowhere.
What the study does not say
It says nothing about Quebec. It covers 100 audited clinics, not the province. The sample is not built to be representative and no figure should be read as such.
It does not judge the quality of care. A slow website has never treated anyone badly. The study measures what a patient encounters before walking in, and what a machine understands of the site.
It does not always distinguish the clinic from its supplier. Twenty-two clinics in the panel do not control their own site: it belongs to a group or a platform, and several of them share the same model. When one of these sites fails a criterion, the responsibility lies elsewhere. The patient, though, encounters the same site.
This sharing has a second consequence. Sites built on the same model do not count as separate observations: it is a single finding repeated. The table's ranges are widened to account for this.
We also checked two other things that could have skewed the results.
First, the draw itself. Since it starts from a list of dentists, a clinic where several practise was more likely to be drawn than a solo practice. We counted how many dentists practise in each of the 100: size has no link to the score obtained.
Second, our way of counting. The ten criteria count equally: a privacy policy is worth one point, site speed is worth one point. That is a choice, and another choice was possible. So the table gives each criterion separately: the average out of 10 is only a summary.
The detail of these checks is available on request.
What comes next
The study is designed to be redone. The same 100 clinics will be re-measured in six months, on the same criteria. The state of each site at the time of the first measurement is archived with an independent third party. Whatever has changed will therefore be demonstrable.
The full report, in a printable version.