AI for Dentists: What Works in a Dental Practice Today

The short answer

Where AI helps a dental practice today

AI is useful in a dental office today in two kinds of work: reading X-rays alongside the dentist, and running the front office. On the clinical side, several X-ray analysis tools are FDA cleared to mark suspected caries, bone loss and other findings for the dentist to review, among them Pearl's Second Opinion (510(k) K210365) and Overjet's Caries Assist (K212519). On the business side, AI answers phones, works recall lists, verifies insurance, drafts clinical notes and handles patient texting.

Be careful in three places. No cleared tool diagnoses on its own: they are labeled as aids to the dentist. Patient-facing claims ("the AI found it") can overstate what the tool is cleared for. And any tool that touches patient information needs a Business Associate Agreement first, because a dental practice that bills insurance electronically is a HIPAA covered entity (45 CFR 160.103; 164.502(e)).

FDA and ADA records checked November 4, 2026. General information, not legal or clinical advice.

Area by area

AI tools for dentists, and what to check before you buy

Six places AI shows up in a dental office, what it actually does, and the question that matters most for each.

Area	What AI does	Check first	For 1 to 3 dentists
Phones and the front desk	Answers when the team is chairside, at lunch and after hours; takes new-patient and appointment requests; sends dental emergencies to the on-call dentist.	A BAA; whether it books or only takes requests; how a caller reaches a person; what it says it can't do.	Good first step
Recall and reactivation	Finds patients overdue for hygiene in the practice management system and sends texts or calls to bring them back.	That it reads your schedule correctly, honors opt-outs, and keeps treatment details out of the message.	Worth it, second
Insurance verification and claims	Checks eligibility and benefits before the visit, fills in breakdowns, drafts claims and flags missing attachments.	A BAA; how it stores your payer portal logins; who reviews a claim before it goes out.	Good first step
Clinical notes	Listens to the visit or to dictation and drafts the note, sometimes the perio chart.	A BAA; that the dentist reads and corrects every note before signing; whether audio is kept.	Worth it, second
Patient texting	Reminders, confirmations, missed-call text-backs and two-way conversations with the office.	No diagnoses or treatment details in texts; carrier registration for business texting; STOP handled.	Good first step
X-ray and image analysis	Marks suspected caries, bone loss, periapical radiolucencies and restorations on radiographs for the dentist to review.	The 510(k) number; that the clearance covers your image types (bitewing, periapical, panoramic, CBCT) and patient ages.	Only with checks

The last column is our opinion for a small office, explained below. Our existing guides go deeper on the dental AI receptionist, the dental answering service and dental patient texting.

Clinical AI

FDA-cleared AI for dental X-rays

A sample from FDA's 510(k) database, not a complete list and not a recommendation. Each number links to the FDA record.

Device (FDA name)	Company	510(k)	Decided	Scope
Second Opinion	Pearl	K210365	Mar 4, 2022	Bitewing and periapical, permanent teeth, age 12 and up, as a second reader
Second Opinion Pediatric	Pearl	K243893	May 5, 2025	Children
Second Opinion Panoramic	Pearl	K250525	Nov 14, 2025	Panoramic radiographs
Second Opinion 3D	Pearl	K243989	May 23, 2025	CBCT
Overjet Dental Assist	Overjet	K210187	May 19, 2021	Bone level measurement
Overjet Caries Assist	Overjet	K212519	May 10, 2022	Caries on bitewings, permanent teeth, age 18 and up
Overjet Caries Assist-Pediatric	Overjet	K233738	Mar 4, 2024	Children
Overjet Periapical Radiolucency Assist	Overjet	K231678	Sep 21, 2023	Periapical radiolucencies
Videa Caries Assist	VideaHealth	K213795	Apr 21, 2022	Caries

Cleared means "as an aid," in FDA's own words. Pearl's summary describes Second Opinion as software that marks regions "in relation to suspected dental findings" to help dental professionals review bitewing and periapical radiographs of permanent teeth in patients 12 and older "as a second reader" (K210365 summary). Overjet's says Caries Assist is "not intended as a replacement for a complete dentist's review or their clinical judgment," and that it was cleared for patients with permanent teeth who are at least 18 (K212519 summary). Both companies later got separate clearances for children, which tells you the age limits are real.

It finds more, and asks you to rule out a little more. In Overjet's reader study, dentists' overall sensitivity for caries rose from 57.9% unassisted to 76.2% with the software; specificity went from 99.3% to 98.4% (K212519 summary). More true findings, and a few more marks the dentist has to look at and dismiss. That trade is worth knowing before the first patient asks why the screen shows a red box.

The ADA wants the accuracy claims tested independently. Its Technical Report No. 1109:2025 calls for an independent validation dataset held by a third party, not an AI manufacturer, and ANSI/ADA Standard No. 1110-1:2025 sets out how radiograph data for AI should be annotated and collected (ADA, AI in dentistry). When the report came out, the ADA's advice to dentists was to ask about a product's intended use and how well it performs (ADA News). To check a vendor yourself, search its name in FDA's 510(k) database or FDA's list of AI-enabled devices, which FDA says is not comprehensive.

Where to be careful

Three ways AI goes wrong in a dental office

Diagnosis without the dentist. A mark on a radiograph is a suggestion. The treatment plan should rest on the dentist's read and the clinical exam, and the chart should show the dentist confirmed it. Watch for workflow drift: a hygienist presenting treatment straight from the AI overlay before the dentist has looked is the failure the labels warn about.

Patient-facing claims. A line like "our AI detects every cavity" on a website or in the chair is a claim you have to be able to back up, and the cleared indications don't say that. Safer: "We use FDA-cleared software that helps the dentist review your X-rays." The same goes for phone AI: it should never tell a caller what's wrong with a tooth.

Tools without a BAA. Any vendor that creates, receives, maintains or transmits patient information for your practice is a business associate, and the BAA comes before the data (45 CFR 160.103; HHS guidance). That includes the AI note taker, the insurance tool holding your payer logins, and a staff member's personal ChatGPT. One detail for dentists: OpenAI's free ChatGPT for Clinicians, which comes with an individual BAA, currently lists physicians, nurse practitioners, physician assistants and pharmacists as eligible, not dentists (OpenAI). Our guide to whether ChatGPT is HIPAA compliant covers which plans can sign one.

Our take

Start at the front desk, not the X-ray

For a one to three dentist office, the first AI money should go to the phones or insurance verification, not to X-ray analysis. Front-office tools can be measured in a month against numbers you already have (missed calls, verification time), and a mistake costs a callback, not a wrong treatment plan. X-ray AI is a reasonable second step, chosen by the dentists, once the office has a written AI policy and the habit of checking what AI produces.

Here is how to prove us wrong in your own office. Pull last month's missed-call and after-hours call counts from your phone system, and the hours spent on insurance verification. If both are close to zero, the front office isn't your bottleneck and our order doesn't apply to you: go straight to the clinical tools your dentists want.

Disclosure: we make one of the tools in this order. Lydoh answers dental office phones and handles patient texting, so we benefit if you start with the phones. It takes requests for staff to confirm rather than booking into your practice management software, and it doesn't run recall or bulk campaigns; if those are what you need, a dental engagement platform is a better fit. VoiceboxMD, our sister product, drafts clinical notes (voiceboxmd.com). Hold both to the same checks as everyone else on this page.

For what a front-office pilot looks like in numbers, our case study follows a one-doctor medical practice, not a dental office, through its first month with AI on the phones: 18% of calls came outside office hours, and a third were forwarded to staff by design.

A plan

The first 90 days of AI in a 1 to 3 dentist office

One problem at a time, with a baseline before and a decision after.

Days 1 to 30

Write the rules and measure the baseline

List every AI tool already in the office, including features switched on inside your practice management, imaging and insurance software, and any personal ChatGPT accounts staff use for work.
Adopt a one-page AI policy (our free template is a starting point): approved tools, what never goes into an unapproved tool, who reviews AI output.
Get a signed BAA from every tool on the list that touches patient information, or stop using it for patient information.
Pick one problem and measure it for two weeks: missed and after-hours calls from your phone system's report, or the hours the front desk spends on insurance verification.

Days 31 to 60

Run one front-office pilot

Start the one tool that matches the problem you measured: AI phone answering after hours and on overflow, or insurance verification. One tool, not three.
Keep a person one step away: callers can reach the front desk during hours, and staff can override anything the tool did.
Once a week, read what the tool handed back to staff (requests, flags, errors), not only the vendor's dashboard of what it handled.
At day 60, compare with your baseline. Keep it, change it or cancel it. Avoid an annual contract for a pilot so this stays a real decision.

Days 61 to 90

Decide on clinical AI with the dentists

If the dentists want X-ray AI, ask each vendor for its 510(k) numbers and read the indications on FDA's database: image types, patient ages, findings covered.
Trial it as a second reader on real cases. Note how often it marks something the dentist rules out, and how often it catches something worth a second look.
Decide how you'll talk to patients about it: the images help explain findings, but the diagnosis is the dentist's, and the chart should say so.
If you trial AI notes, the dentist edits and signs every note during the trial, and you check a sample for anything the AI added that didn't happen.

Tools for the first month: our free AI policy template and vendor checklist (written for medical practices; it works for a dental office with the word "doctor" changed). For the same question in other practices, see AI for doctors, AI in veterinary practice and AI in healthcare examples.

FAQ

AI in dentistry questions

How are dentists using AI today?

In five places: software that marks suspected findings on X-rays for the dentist to review, AI that answers the phone and takes appointment requests, tools that check insurance eligibility and help with claims, AI that drafts clinical notes from the visit, and texting that sends reminders and handles replies. The X-ray tools are FDA-regulated medical devices; the rest are office software, where the main questions are privacy and accuracy.

Is dental AI FDA approved?

Some X-ray analysis software is FDA cleared, which is different from approved. Pearl's Second Opinion (510(k) K210365, March 2022), Overjet's Caries Assist (K212519, May 2022) and VideaHealth's Videa Caries Assist (K213795, April 2022) are examples in FDA's 510(k) database. A 510(k) clearance means FDA found the device substantially equivalent to one already on the market for a stated use. Phone, insurance, notes and texting tools are not medical devices and are not FDA cleared.

Can AI diagnose cavities?

No. The cleared tools are cleared as aids. Overjet's FDA summary says Caries Assist provides additional information for the dentist and is not a replacement for the dentist's review or clinical judgment. Pearl's says Second Opinion works as a second reader. The dentist looks at the marked area, examines the patient and makes the diagnosis.

Does AI X-ray software find more cavities?

In Overjet's reader study in its FDA summary, dentists' sensitivity for caries on bitewings rose from 57.9% without the software to 76.2% with it, while specificity dipped from 99.3% to 98.4%. That means more true findings and a few more false ones to rule out. Ask any vendor for the equivalent numbers in its own FDA summary, for the image types and ages you treat.

What does the ADA say about AI in dentistry?

The ADA has published ANSI/ADA Standard No. 1110-1:2025, the first US standard on AI in dentistry approved by ANSI, on how 2D radiograph data should be annotated and collected; Technical Report No. 1109:2025, which calls for an independent validation dataset held by a third party rather than an AI manufacturer; and White Paper No. 1106:2022, an overview of AI uses including claims processing. The ADA's advice when it released the technical report was to ask about a product's intended use and how well it performs.

Can my dental office use ChatGPT with patient information?

Not with a personal or consumer account: those come with no Business Associate Agreement, and a dental practice that bills insurance electronically is a HIPAA covered entity. OpenAI's free ChatGPT for Clinicians, which includes an individual BAA, currently lists physicians, nurse practitioners, physician assistants and pharmacists as eligible; dentists are not on that list. Our guide to ChatGPT and HIPAA covers which plans can sign a BAA.

What AI should a small dental office start with?

Our view: start with the phones or insurance verification, because both are easy to measure in 30 days and neither touches a diagnosis. Add X-ray AI when the dentists want a second reader and have checked that the clearance covers the image types and patient ages they see. The 90-day plan on this page lays out the order.

Can AI book appointments for a dental office?

Some systems write directly into practice management software; others, including Lydoh (which we make), take the request and leave the booking to staff. Writing straight into the schedule saves a step but needs a tight integration and rules about which slots it may use. Either way, ask what happens when a caller wants something the AI can't do: there should be a person one step away.

Is a dentist responsible for what the AI gets wrong?

Treat it that way. The cleared X-ray tools are labeled as aids to the dentist, and AI-drafted notes become the dentist's notes once signed. On the business side, a vendor that handles patient information for your practice is your business associate, and the practice is the one that must have a BAA in place before sharing it. This page is general information, not legal advice.

AI for dentists: what works in a dental practice today