How dental and physio clinics use AI, and the privacy catch

Three to four hours a week. That is how much paperwork time Ontario family doctors and nurse practitioners got back when they let an AI listen to their appointments and draft the notes, according to OntarioMD's three-month evaluation of more than 150 clinicians. They reported spending 70% to 90% less time on documentation.

For a dentist or a physiotherapist who finishes every day with a pile of unsigned charts, that number is the whole pitch. It is also where the questions start, because the thing doing the listening is recording a patient's health information.

This post looks at what small dental and physio clinics are actually using AI for, what the numbers say, and what Ontario's privacy rules expect before you switch anything on.

How many clinics use it at all

Fewer than the vendor emails suggest, and mostly for one job.

The American Dental Association's Health Policy Institute surveyed US dentists and released the results in July. As Dentaltown reported, 43.3% use AI for at least one task, and another 26.4% plan to.

A smaller survey that includes Canada lands lower. Dental Reviewed asked 300 licensed dentists in Canada, the US and the UK between February and April 2026, and 32% said they use AI tools, with 38% considering it.

The limits matter here. The ADA survey covers American dentists only. The Dental Reviewed sample is 300 people across three countries, and the report does not split out Canada. Neither covers physiotherapists, and we could not find a published survey of AI use in Canadian physio clinics. Treat these as a rough picture of dentistry, not a census.

What they use it for: mostly not the front desk

The ADA's breakdown by task is more useful than the headline figure, because it shows where AI has actually landed in a practice.

Imaging leads by a distance. In the Dental Reviewed survey, 82% of dentists who use AI use it to help read radiographs. That is clinical software, usually bought through an imaging or practice-management vendor, not something a clinic sets up itself.

The administrative uses are smaller but more interesting for a small clinic. Insurance verification, the front desk and social media are all jobs that eat staff time without needing a clinician. Only 10.1% of dentists use AI at reception today.

The planned uses point the same way. Among the ADA's respondents, 34.8% plan to use AI for charting and note-taking and 32.6% for insurance verification. Very few want it making clinical calls: 82.6% do not plan to use AI for treatment recommendations, and fewer than 5% use it that way now.

So the realistic picture for a two- or three-chair dental office, or a physio clinic with a handful of practitioners, is narrow. AI is being tried on paperwork, the phone and insurance admin. Clinical judgement stays with the clinician.

The notes: where the time actually comes back

The AI scribe is the use with the best evidence behind it, and the one most relevant to physiotherapy, where every visit ends in a treatment note.

An AI scribe is software that listens to the appointment, turns the conversation into text and drafts a clinical note from it. The clinician reads, corrects and signs. In OntarioMD's evaluation, the participants were family doctors and nurse practitioners, not dentists or physios, but the task is the same.

Those are good numbers, with two caveats. The study is from 2024 and is a self-reported satisfaction survey, not an independent audit of note quality. And it covers primary care, where visits are conversations. A physio visit includes long stretches of hands-on treatment and exercise coaching, and a dental appointment is mostly a patient who cannot talk. How much a scribe catches in those rooms is something to test in your own practice.

The College of Physiotherapists of Ontario has already written for this case. Its guide to AI scribes tells physiotherapists to obtain informed consent from patients and to "review all AI-generated notes to ensure they are accurate and complete." It is blunt about who carries the risk: "The ultimate responsibility for patient care... remains with you, the physiotherapist."

What goes wrong

The 40% abandonment figure from the Dental Reviewed survey is the most useful number in this post. Four in ten dentists in that survey had dropped an AI product within three months of starting it, and the report names cost and poor integration with existing software as the leading barriers. A tool that does not write into your practice-management system creates copy-and-paste work instead of removing it.

Four in ten dentists had dropped an AI product within three months.

The scribes have their own failure modes. In Policy Options, York University researcher Helen Beny lists the main ones:

• Omissions. Scribes "often prioritize conversation flow over clinical nuance," so a detail mentioned once, in passing, can drop out of the note.
• Mishearing. Noise, accents and several people talking can produce transcription errors, including in a prescription.
• Hallucination. That is the term for software producing something that "appears to be correct or plausible but is false or misleading."
• Uneven accuracy. She cites research finding higher error rates for African American patients than for white patients.

None of these is a reason to avoid scribes. All of them are a reason to read every note before you sign it, which is exactly what the colleges already require.

The privacy catch

This is where a clinic differs from a restaurant or a trades business. In Ontario, patient information is governed by PHIPA, the provincial health privacy law, and both the physio college and Ontario's privacy commissioner hold clinics to it when they use AI.

In January 2026 the Information and Privacy Commissioner of Ontario published guidance on AI scribes for the health sector. A Baker McKenzie summary picks out what it expects:

• A privacy impact assessment, meaning a written review of what data the tool handles and what could go wrong, completed before you start and updated as the tool changes.
• Vendor contracts that limit what the vendor can do with your data, including a bar on training its AI models on patients' health information.
• A retention schedule for the audio recordings.
• Trained people reviewing the AI's notes before they enter the patient record.

For a sense of what retention looks like in practice, Policy Options reports that Sunnybrook keeps recorded clinician-patient conversations for 30 days before destroying them, and that explicit patient consent is required before use. A small clinic should know its own number, and the vendor's, before the first recording is made.

The same questions apply to an AI receptionist. A phone agent that books appointments hears names, dates of birth and the reason for the visit. That is health information too, and it goes to the vendor.

Where it is not worth the trouble: a solo practitioner who already finishes notes during the visit, or a clinic whose practice-management software has no supported integration. The Dental Reviewed abandonment figure suggests those are the cases where the tool ends up switched off.

Where to start

1. Pick one job, not three. For most physio clinics that is treatment notes; for most dental offices it is insurance verification or after-hours calls. Measure how long it takes today, so you can tell in a month whether anything changed.
2. Before any trial, put four questions to the vendor in writing: where recordings are stored, how long they are kept, whether your data trains their models, and which practice-management system they write into. If the answers are vague, stop there.
3. Write a one-paragraph consent script and a note-review rule, such as "no AI note is signed unread", and run the trial for a month with one practitioner before rolling it out.

If you want a second pair of eyes on the vendor answers or the setup, you can book a call with Trimotech.