AI in the dental practice: what it does, what studies show and what to check

Artificial intelligence is now part of the dental conversation, and many owners are asked about it before they have a way to judge it. This guide explains what the term covers, what peer-reviewed studies show, how federal rules treat some of the tools, and what privacy questions to settle first. It is general education from Copeland Consulting Services. It names no products and gives no clinical advice. It belongs to the wider dental technology guide.

What does AI mean in dentistry?

In dentistry, artificial intelligence usually means software that finds patterns in data and produces a result, such as a marked area on an image or a draft of a note. A 2023 review in Frontiers in Dental Medicine defines AI as “the ability of machines to perform tasks that normally require human intelligence.” The same authors say the field became practical only in the past two decades, helped by more data, more computing power and better algorithms.

The ADA draws one more line that is worth knowing. In its news report on a standards white paper, it separates artificial intelligence from augmented intelligence. Artificial intelligence refers to intelligence shown by machines rather than humans. Augmented intelligence keeps elements of human intelligence in the process and acts as a tool that helps the clinician do the task. Many dental products are described the second way: the software points, and the dentist decides.

What are the main uses of AI in a dental practice?

Uses fall into two groups with different evidence and different rules: clinical uses that look at images and data, and office uses that handle words and tasks.

The ADA news report lists these uses, based on comments from the chair of its Council on Dental Practice:

Another category that owners ask about is the AI dental scribe, software that listens to a visit and drafts a note. It fits the “administrative support” idea. This guide found no peer-reviewed summary of scribe accuracy among its sources, so it makes no accuracy claim about them. Ask the seller for independent evidence.

A second review, in Medical Science Monitor in 2025, adds that AI has been studied in endodontics, periodontics, implantology, orthodontics, prosthodontics and oral and maxillofacial surgery. Research is wide. Daily use in most offices is narrower.

Where is the research strongest?

Research is strongest for tasks that look at images. The Frontiers review says “the majority of the AI applications in dentistry are for diagnosis based on radiographic or optical images.” It explains that other tasks are harder to study “mainly due to the constraints of data availability, data uniformity, and computational power for handling 3D data.”

This matters for buyers. If a seller claims a result for a task outside image analysis, ask what study supports it, because far fewer studies exist.

What does a meta-analysis of caries detection show?

One of the most studied tasks is finding caries on bitewing radiographs. A 2024 systematic review and meta-analysis in the Japanese Dental Science Review screened publications from after 2010. Fourteen of 935 references met its criteria.

Its main findings, as reported in the abstract:

• Dataset sizes ranged from 112 to 3,686 radiographs.
• 86 percent of the studies reported a model with an accuracy of 80 percent or more.
• Most studies had an unclear or high risk of bias.
• Three studies compared the model with dentists, and the models showed higher average sensitivity.
• Five studies were pooled. The summary sensitivity was 0.87 and the summary specificity was 0.89.
• For lesions in dentin, sensitivity was 0.84. For lesions in enamel, it was 0.71.

Sensitivity is the share of true cases the tool finds. Specificity is the share of healthy sites it correctly leaves alone. A tool with good numbers can still be wrong for a given patient.

The authors conclude that, despite the promising results, “the lack of high-quality, adequately reported, and externally validated studies highlight current challenges and future research needs.” External validation means testing a model on images it did not learn from, often from different clinics and machines. Without it, you cannot assume the result will hold in your office.

For a reader, the practical message has three parts. Results are encouraging. Study quality is uneven. The enamel result is weaker than the dentin result, which shows that performance changes with the task.

What limits and risks do the reviews name?

The 2025 narrative review in Medical Science Monitor lists challenges in plain terms: “inaccurate data annotation, limited capability for fine-grained feature expression, a lack of universally applicable models, potential biases in learning algorithms, and legal risks pertaining to medical malpractice and data privacy breaches.”

Put simply:

• Training data can be wrong. If the labels used to teach a model are inaccurate, the model learns the errors.
• Models may not travel. A model built in one setting may not fit patients, machines or image styles in another.
• Bias is possible. If the training set under-represents some groups, the tool may do worse for them.
• Legal and privacy risk is real. The authors name malpractice and data privacy breaches.

The ADA’s commentary adds the human side. Its council chair is quoted: “While the integration of AI and augmented intelligence in dentistry is progressing, it’s important to note that human expertise and clinical judgment remain essential.” The report also lists patient safety, data privacy and ethics, including transparency, informed consent and bias, as reasons standards are needed.

How does the FDA treat dental AI?

Some AI products are regulated as medical devices, and some are not. The rule depends on what the software is for.

Federal regulation 21 CFR 892.2070 addresses medical image analyzers. It says these devices, “including computer-assisted/aided detection (CADe) devices for mammography breast cancer, ultrasound breast lesions, radiograph lung nodules, and radiograph dental caries detection,” are “a prescription device.” They are meant to direct the clinician’s attention to parts of a radiology image that may show abnormalities. In plain terms, software that flags possible caries on dental radiographs is a named example in this regulation. The regulation describes the device as an aid to the clinician reading the image.

The FDA also keeps a public list of AI-enabled medical devices, with links to its database entries. It warns that the list “is not a comprehensive resource of AI-enabled medical devices,” because it was built mainly from AI-related terms in the summary documents. So the absence of a product from the list is not proof that it lacks clearance, and the presence of a product tells you only what it was cleared to do.

Use this table when a seller says a product is “FDA cleared.”

A statement from a seller is not legal advice. Ask your attorney how the product’s status affects your duties.

What are the HIPAA risks?

Any AI tool that sees patient information raises privacy questions. The first is whether the vendor is a business associate. HHS explains that a covered entity may disclose protected health information to a business associate “if the covered entity obtains satisfactory assurances, in the form of a contract or other written arrangement” (a business associate agreement) that the associate will safeguard the information.

For cloud-based tools, HHS says that when a covered entity uses a cloud provider to “create, receive, maintain, or transmit ePHI,” the provider “is a business associate under HIPAA,” and the parties must sign a HIPAA-compliant agreement. Many AI tools run in the cloud. The HIPAA Security Rule and your own cybersecurity practices still apply. If a vendor will not sign an agreement, do not send it patient information.

Questions to ask any AI vendor that touches patient data:

1. Will you sign a business associate agreement?
2. What patient data goes to your system, and what stays in our office?
3. Do you keep our data? For how long? Where?
4. Do you use our data to train or improve your models? Can we opt out?
5. Who at your company can see our patient information?
6. What is your process if there is a data breach, and how fast will we hear about it?
7. Can we get our data back, and is it deleted when we leave?

The last two matter more than they seem. Settle them before you sign, not after.

Do not paste patient details into a general-purpose chat tool that has no agreement with your practice. If your team might do this on its own, put a clear written policy in your handbook. Your attorney or a compliance adviser should check the policy and the contract terms.

How should a practice try an AI tool?

A short, controlled trial is a safer first step than a full rollout. Try this plan:

1. Pick one question. For example, “Does this tool change how often our providers agree on bitewing findings?” or “Does it shorten the time to finish notes?”
2. Take a “before” measure. Time the current process or count the current disagreements.
3. Keep the dentist in charge. Every finding is reviewed, and every clinical decision belongs to a licensed dentist.
4. Tell patients when it matters. The ADA report names transparency and informed consent among the ethical questions in AI. Ask your attorney how to describe AI use in your forms and in the room.
5. Log what happens. Note what the tool flagged and what the dentist decided. Write down the cases where the tool was wrong.
6. Set a review date. After the trial, compare the result with the “before” measure and decide whether to stop, adjust or widen use.

The same cost and training questions that apply to any purchase apply here too. See the guide on dental technology ROI for a way to count the costs.

What should a dental team lead remember?

Four points hold across the sources:

• AI is a support tool. The ADA and the reviewed articles describe it as help for clinicians.
• Evidence is uneven. Image tasks have the most research, and even those studies often carry a risk of bias.
• Regulation depends on use. Some tools are prescription devices. Ask for the specific clearance.
• Privacy comes first. Settle the agreement before you share any patient data.

Standards are also changing. The ADA says it is working on a technical report and a standard for image analysis systems and on the U.S. position for an international standard. Check the ADA’s technology pages again before you decide.

Next step

For the wider topic, read the dental technology pillar guide, and see the technology page. To raise a question with Dr. Jon Copeland, DDS, a general dentist in Wildwood, Missouri, contact Dr. Copeland about consulting.