43% of dentists use AI, but not for notes

Every dental AI pitch you have sat through in the last two years opened with a radiograph. A caries lesion the software found, a bone level it measured, a slide showing case acceptance climbing once patients see the overlay. The ADA Health Policy Institute has now published what dentists actually bought, and imaging is indeed the leader: 22.8% of responding dentists use AI for imaging and diagnostics (ADA HPI, HPI Perspectives, July 2026).

The number underneath it is the one worth your afternoon. Charting and note taking is the single most-wanted future application in the survey, at 34.8% of dentists who are not using it today. It does not appear anywhere in HPI's reported current-use figures. The task dentists most want AI to take off them is the task almost none of them have bought it for.

What the ADA's own panel found

The headline first. More than two in five responding dentists, 43.3%, use AI for at least one type of task in their practice, and another 26.4% plan to (ADA HPI, July 2026). The data comes from HPI's quarterly panel of US dentists, the same instrument behind the State of the U.S. Dental Economy series. For the Q2 2026 wave, invitations went to 2,432 panel dentists on 15 June 2026, 589 responded for an adjusted response rate of 24.3%, and 552 of those work in private practice including group and DSO settings (ADA HPI, State of the U.S. Dental Economy, 2nd Quarter 2026 Update).

Broken out by task, current use runs: imaging and diagnostics 22.8%, insurance verification 13.6%, explaining clinical findings to patients 13.2%, social media 10.7%, business analytics 10.1%, reception or front-desk check-in 10.1%, and patient treatment recommendations under 5%.

Stated future intent runs differently. Among dentists not using AI for a given task today, 34.8% plan to adopt it for charting and note taking, 32.6% for insurance verification, and 25.4% for imaging and diagnostics.

So what for you: the survey your vendors will quote at you is the 43.3%. The survey line that should shape your next purchase order is the ordering of that second list, because it is the only published signal of what your peers have concluded after living with the first list.

The demand-to-installed-base ratio, and how to read it

Put current use and stated intent side by side and divide one by the other. It is a crude ratio, and it is the most useful thing in the dataset.

Insurance verification: 32.6 divided by 13.6 gives 2.4. For every dentist running AI on insurance verification today, roughly two and a half more say they intend to.

Imaging and diagnostics: 25.4 divided by 22.8 gives 1.1. Intent is barely ahead of installed base. The category that took the marketing budget for three years is close to exhausting its own remaining demand pool.

Charting and note taking cannot be expressed as a ratio at all, because there is no reported current-use figure to divide into. On the strength of 34.8% stated intent, that is the widest demand gap in dentistry right now.

One caution on the arithmetic. These are two different question sets, current use and planned use, not a longitudinal measure of the same dentists, and HPI reports planned use among those not already using each task. The ratio is a directional read on where demand sits relative to supply, not a forecast of anybody's purchase date.

So what for you: if you are choosing between two proposals this quarter, the imaging one is competing in a category where your peers have largely already decided, and the admin one is competing in a category where they have decided and have not yet acted. Second movers get better pricing and fewer surprises in the first category. In the second, you are early enough for it to matter.

Broad, but roughly two tools deep

Sum the named current-use figures and you get about 85 percentage points of task usage spread across the 43.3% of dentists who use any AI at all. That works out at roughly two named tasks per adopting practice (85.1 divided by 43.3 equals 1.97).

Treat that as a floor rather than a precise count, for two reasons. HPI reports the treatment recommendations figure as "less than 5%" rather than a point estimate, so the numerator carries an error bar. And the published task list may not be exhaustive of everything the survey asked. Even allowing for both, the picture is clear enough: the typical dental AI adopter in mid-2026 is running an imaging tool and one administrative tool, not an integrated stack.

That matters more than it sounds. Two tools purchased separately, from separate vendors, on separate contracts, integrated into the practice management system by separate means, is not a small amount of governance for a practice with no IT function. It is the point at which the ADA's own submission to HHS becomes relevant: the Association told the Assistant Secretary for Technology Policy in February 2026 that adoption decisions require coordination between practice owners, IT, compliance and legal, and that this coordination burden falls hardest on small and rural offices.

So what for you: your second AI purchase costs more than your first, and not in license fees. Budget the integration and the oversight, or you will discover them.

The clinical ceiling almost nobody prices in

Here is the finding that should change how you read every dental AI proposal you see for the rest of this year. Four out of five responding dentists, 82.6%, do not plan to use AI for treatment recommendations at any point. Two-thirds, 68.3%, do not plan to use it to explain clinical findings to patients (ADA HPI, July 2026).

These are not fence-sitters yet to be convinced. They are recorded refusals. The free-text responses HPI published are consistent and specific: concern that dentists will lean on AI in place of their own diagnostic judgment, concern about overtreatment and misdiagnosis, and at least one respondent reporting a hit rate around 30% on radiographic reading in their own hands.

The ADA's response to this is not to dismiss it. Its Standards Program is developing a standard for AI model fact labels in dentistry, covering intended use, training data, validation method, performance characteristics and limitations. That is a direct institutional acknowledgement that dentists currently cannot tell, from what vendors publish, whether a clinical claim holds. Until those labels exist, any vendor telling you a diagnostic tool is validated should be asked to name the study, the population and the comparator, in writing.

So what for you: the clinical end of dental AI has a stated ceiling of roughly 17% for treatment recommendations, and you are being sold into it as though it were open field. Discount clinical-decision claims accordingly, and price the imaging tool on what it does for throughput and patient communication rather than on a diagnostic promise the profession has already declined.

Why this is an expansion question, not a technology question

The reason dentists want AI on charting and insurance rather than on diagnosis is not squeamishness about technology. It is where the pain is. Around one-third of dentists, 32.4%, report being overworked (ADA HPI, State of the U.S. Dental Economy, Q2 2026), and dentists are spending a growing share of the working week on non-clinical tasks (ADA HPI, Trends in Dentists' Income, Revenue and Hours Worked, 2026). The hygienist shortage running through the same HPI data means the labour you would normally throw at admin is the labour you cannot hire.

Now apply that to a third site. Imaging AI is priced and deployed per operatory, so its cost scales linearly with chairs and its benefit stays inside the surgery. The administrative layer prices per site or per seat, and its benefit compounds across sites, because insurance verification, recall and front-desk handling are the workflows that fragment first when a principal stops being physically present in every building. That fragmentation is the actual constraint on the third site, and it is the same standardization problem that multi-site groups solve with configuration rather than with more staff.

So what for you: the sequencing question is not which AI is better. It is which AI makes the third site cheaper to run than the second one was, and only one of these two categories is priced in a way that can.

The buy order the data supports

Frankly, if you own two or three sites and have budget for one AI purchase in the next two quarters, the ADA's own numbers say buy the administrative layer first and put imaging behind it.

The reasoning is arithmetic rather than preference. The admin categories carry the highest unmet demand among your peers (2.4 on insurance verification, no measurable installed base on charting against 34.8% intent). They address the pain the profession reports directly, that 32.4% overworked figure. They scale on the cost basis that helps a growing group rather than the one that penalizes it. And they sit entirely outside the clinical-decision zone that 82.6% of dentists have already refused, which means you are not buying into a category with a ceiling.

Three practical conditions on that call. Ask any charting or scribe vendor for its error profile in writing before signing, because the documented failure mode in ambient dental documentation is fabricated detail rather than missed detail, and a note that reads fluently is not the same as a note that is correct. Confirm the vendor will sign a business associate agreement and tell you where audio and transcripts are stored and for how long. And check whether your practice management system supports the integration natively or whether you are paying for a middleware layer nobody mentioned in the demo.

The practices that get this right are usually the ones that decided what their bottleneck was before they took the first vendor call. If you are not certain where yours sits, that is exactly what our AI Opportunity and Growth Assessment establishes: what is running, what it costs, where the time actually goes, and what to buy in what order. You can also book a 20-minute call to talk through the third-site question specifically.

Related: The hygienist shortage is a capacity problem, not a hiring problem  ·  Multi-site dental AI already runs across 1,900 sites  ·  What an AI scribe's hallucination rate actually measures

All adoption figures above are from the ADA Health Policy Institute's quarterly panel of US dentists as published in HPI Perspectives, July 2026, and from the State of the U.S. Dental Economy, 2nd Quarter 2026 Update. HPI notes that other published surveys report higher take-up rates than its own; those alternative figures come from trade publications and vendor-adjacent research and are not relied on here. One inconsistency in the source is worth flagging: the HPI Perspectives text gives both 34.8% and 30.6% for planned adoption of AI charting and note taking in adjacent sentences. The higher figure is used above, and the argument does not turn on which is correct, since both are the largest planned-use figure in the dataset. No verified figure for dental AI spending per practice was available at the time of writing, so no spending estimate is given. This article is for informational purposes only and does not constitute clinical, legal or financial advice.