Dental AI Is Changing Diagnosis, Case Acceptance, DSOs and Dental Insurance
Dental AI Is Changing Diagnosis, Case Acceptance, DSOs and Dental Insurance
A patient sits in the operatory while a bitewing appears on the monitor. A colored box suddenly outlines a suspicious area. To the patient, the image can feel more authoritative than a dentist pointing at a gray shadow with an explorer in hand. To the dentist, it can be a useful second set of eyes. To a DSO, the same technology can become a source of clinical and operational data. To an insurer, artificial intelligence can help evaluate whether submitted evidence satisfies coverage rules. Dental AI is no longer just software that circles cavities. It is becoming part of the infrastructure connecting diagnosis, treatment presentation, management and reimbursement.
That evolution is why the most important question in dental AI is no longer whether the technology works. In many narrow radiographic tasks, it clearly does. The better question is what happens after the software finds something. Who decides whether the finding represents disease, whether the disease needs treatment now, how it is presented to the patient, whether the dentist is measured against the algorithm, and whether the insurer applies a different threshold when deciding what to pay.
The FDA record is a useful place to begin because it strips away some of the marketing language. Videa Dental AI received FDA 510k clearance in September 2025 for computer assisted detection of multiple findings on dental radiographs, including caries, periapical radiolucencies, calculus and several other features. The FDA documentation is equally clear about the limit of that clearance. The system is intended as an adjunct, it should not replace the dentist’s review of the image, and it should not be used instead of a full patient evaluation or relied upon alone to make or confirm a diagnosis. That distinction matters. FDA clearance can involve meaningful performance testing, but it does not mean the FDA has declared every highlighted lesion diseased, every treatment recommendation appropriate, or every downstream marketing claim proven.
Independent clinical evidence reinforces that more restrained view. A 2026 JADA external validation study evaluated two FDA cleared commercial AI systems using records from 90 patients at the Veterans Affairs Greater Los Angeles Health Care System. For caries, the systems showed negative predictive values of about 97 percent. For clinically relevant periodontal bone loss, specificity was about 94 percent. The authors concluded that the systems were especially useful as adjunctive screening tools, with strong value in ruling out disease and reducing false positive findings, while positive findings still required clinician confirmation. In other words, one of AI’s most important contributions may not be finding more dentistry to perform. It may be helping dentists become more confident when treatment is not indicated.
That is a crucial correction to the way dental AI is sometimes discussed commercially. A finding is not a diagnosis. A diagnosis is not automatically a treatment recommendation. A treatment recommendation is not automatically treatment necessity. And treatment is not automatically a better patient outcome. The software may identify a radiographic feature accurately while the real clinical decision still depends on symptoms, history, disease activity, caries risk, periodontal probing, attachment levels, previous images, restorative history, patient preferences and the dentist’s judgment.
Patient psychology makes this distinction even more important. AI visualizations are persuasive because they make an invisible problem visible. That can be a genuine advance in communication. Patients who never understood a conventional radiograph may finally see what their dentist is describing. But persuasive presentation can also amplify error. Research published in NEJM AI found that nonexperts could place substantial trust in AI generated medical advice even when the advice was inaccurate. A polished answer, confident language or brightly colored annotation can create an impression of certainty that exceeds the underlying evidence. In the dental operatory, the phrase, the AI found this, may carry more psychological weight than most dentists realize.
That helps explain why vendors emphasize case acceptance. Videa reported that GPS Dental saw a 19 percent increase in gross production per patient for restorative and periodontal care and 32 percent growth in crown production after adopting its platform. Aspen Dental reported a 12 percent increase in acceptance of recommended care in pilot offices before Videa was rolled out across more than 1,100 locations. These are company reported results, not independent clinical outcome trials. They may represent previously missed disease being detected, better patient understanding and more appropriate care being completed. They may also reflect other changes in workflow, presentation, staffing or treatment thresholds. The production numbers alone cannot tell us which explanation is correct.
This is where dentists need a better scoreboard. Case acceptance is a business metric. Production is a business metric. Neither is inherently good or bad, but neither should be confused with a clinical endpoint. If AI helps a patient understand a legitimate untreated lesion and accept appropriate care, higher case acceptance is a win. If the same persuasive technology pushes an equivocal enamel lesion toward an unnecessary restoration, higher case acceptance is not a win. The meaningful measures are diagnostic accuracy, disease progression, retreatment, tooth survival, complications, patient reported outcomes and long term oral health.
The stakes rise when AI moves from a single operatory into a large organization. Enterprise platforms can standardize radiographic review, reduce unwanted variation and identify clinicians who may be missing disease. That can improve quality. The same data layer can also compare providers, locations, treatment patterns, conversion and production. There is no evidence in the material reviewed that Aspen Dental, GPS Dental, Heartland Dental, Videa or Overjet is imposing quotas requiring dentists to agree with AI. That claim should not be made without evidence. But the technical ability to measure provider variation increasingly exists, which means governance matters before a useful coaching tool quietly becomes a production pressure tool.
Even the software’s sensitivity settings raise questions that were once theoretical. Videa’s current FDA cleared system allows users to switch between predefined high sensitivity and high specificity operating points for caries and periapical radiolucencies. That is not a free slider that lets every dentist train the algorithm to match a personal treatment philosophy. It is a choice between validated operating points. That may actually be safer. An unrestricted system could reproduce an aggressive dentist’s bias just as easily as a conservative dentist’s bias. The larger issue is transparency. Dentists should know which operating point is being used, organizations should preserve an audit trail when settings change, and business management should not be able to alter clinical thresholds invisibly.
Then there is the payer side. Overjet markets AI directly to dental insurers for utilization review and says its insurer customers include a majority of the ten largest United States dental insurers, collectively covering more than 120 million people. The company says its platform can automatically review and approve claims for hundreds of procedure codes and reduce administrative work by 90 percent. Those are Overjet’s own operational claims, but the direction of travel is clear. AI is becoming part of dental reimbursement infrastructure as well as clinical infrastructure.
This does not mean that an algorithm simply denies every claim it dislikes. Overjet has described a model in which AI analyzes submitted evidence against clinical guidelines supplied by the payer, automatically approves claims that satisfy the rules, and routes unresolved cases for human clinical review. Overjet’s legal officer has stated that its AI is not used to deny claims. The distinction is important because an objective measurement is not the same as an objective insurance policy. Two systems can agree perfectly that bone loss is 27 percent and still reach different coverage outcomes if the payer’s reimbursement criteria differ from the treating dentist’s clinical judgment.
Dentistry is therefore approaching a strange new reality. The dentist’s AI may help identify disease and persuade the patient to accept treatment. The insurer’s AI may analyze the same radiograph and documentation to determine whether the claim satisfies coverage criteria. The algorithms may even agree on the anatomy while the humans behind them disagree about what should happen next. The software is not necessarily the source of the conflict. The hidden variable may be the rule each organization asks the software to enforce.
Lawmakers have started to notice. The ADA reported that 20 states introduced bills in 2025 addressing artificial intelligence in dental claim payment adjudication, with Arizona and Maryland among the states enacting legislation. Arizona’s law, effective July 1, 2026, requires a medical director to individually review a proposed claim denial based on medical necessity, exercise independent medical judgment and not rely solely on recommendations from another source. Arizona’s statutory definition of health care insurer includes prepaid dental plan organizations and dental service corporations. The law does not ban AI. It places an accountable human above it when a medical necessity denial is being made.
That may be the regulatory model that survives. AI can measure, flag, triage, compare and recommend. A licensed professional still owns the consequential decision. The same principle belongs on the provider side. If the computer highlights a lesion, the dentist should be able to explain why it is clinically meaningful, why treatment is indicated now, and what reasonable alternatives exist. If the dentist disagrees with the AI, that disagreement should not be treated as a defect merely because the software appears more mathematical.
Periodontal AI provides a good example of why language matters. The CDC describes gingivitis as reversible, while periodontitis involves irreversible destruction that can be slowed and managed. The ADA likewise recognizes important associations between periodontal disease and systemic conditions but cautions that association does not establish causation and that claims that periodontal treatment prevents systemic disease can be misleading without stronger evidence. AI may measure bone loss accurately and help identify patients who need further evaluation. That does not justify turning a radiographic measurement into a promise about reversing periodontitis or preventing heart disease.
For the practicing dentist, the safest workflow is surprisingly simple. Read the original radiograph and examine the patient. Then use AI as a second reader, not the first authority. Investigate what it finds, especially when the finding would change treatment. Be willing to disagree with it. Use the overlay to educate the patient, but do not let the color become the diagnosis. Document the clinical evidence supporting treatment. Know which software version and operating point your practice uses. And when an insurer rejects a claim, separate the questions of clinical necessity and contractual coverage rather than assuming the payer’s algorithm has disproved your diagnosis.
The winning dental practices will not be the ones that reject AI, and they will not be the ones that surrender judgment to it. They will be the practices that understand exactly where AI is strong, where it fails, how it changes patient behavior, and how economic incentives can quietly shape the rules around it. Dental AI is likely to become a routine second reader, a workflow engine, a patient communication tool, an enterprise analytics layer and a major part of insurance administration. The technology may be new, but the professional obligation is not. The dentist remains responsible for deciding what the patient actually needs.
When the dentist, the patient, the DSO and the insurer are all looking at the same X ray through different algorithms, who should have the final say?
Join the Conversation!