Dental AI reality check · Reviewed July 2026

Overjet vs Pearl vs Videa.

All three can assist radiograph review. None replaces examination, history, clinical judgment, or a practice’s responsibility to decide whether an AI mark should change treatment.

Product posture

The meaningful differences are workflow and incentives.

Overjet

Provider and payer platform

Regulatory scope: FDA-cleared assistive detection and measurement products for findings including caries, bone levels, calculus, and periapical radiolucency.
Business footprint: Markets to practices, DSOs, schools, and insurers. Acquired American Dental Examiners in 2021 to add licensed claims review and payer decision-support services.
What to examine: A practice evaluating the clinical product should separately examine payer relationships, data separation, review workflows, and how disagreements are documented.

Pearl Second Opinion

Chairside detection and communication

Regulatory scope: FDA-cleared assistive detection products spanning multiple radiographic findings, with newer clearances for contouring caries and measuring bone levels.
Business footprint: Primarily presented as a provider-facing radiograph review and patient-communication system, with insurance-verification and practice workflow products.
What to examine: Confirm exactly which findings, image types, ages, and product version are enabled. A broad product demo can exceed one specific FDA indication.

VideaAI

Provider, DSO, and practice-software integration

Regulatory scope: FDA-cleared assistive products for dental radiograph findings, beginning with Videa Caries Assist and expanding through later clearances.
Business footprint: Integrated through partnerships including Heartland Dental and Henry Schein One’s Dentrix Detect AI.
What to examine: Evaluate local performance by sensor, case mix, pediatric use, and workflow. DSO adoption is evidence of scale, not proof of diagnostic superiority.

FDA study context

The clearance summaries prove imperfection, not equivalence.

Overjet Caries Assist

76.6% bitewing sensitivity

Its FDA summary reports 99.1% bitewing surface-level specificity. Secondary caries sensitivity was lower than primary caries sensitivity. These are standalone results on the submitted dataset.

Pearl Second Opinion

76.39–89.77% sensitivity range

The original FDA summary reports results across four findings and says aided readers improved overall. Performance varied by finding, and the false-positive rate was not zero.

Videa Caries Assist

70.8% image-level sensitivity

Its FDA summary reports 59.5% image-level positive predictive value on the submitted adult dataset. Positive overlays still required clinician confirmation.

Do not compare these figures directly. Overjet reported surface-level measures, Videa reported image- and lesion-level measures, and Pearl evaluated multiple findings under a different study design.

Claim check

What the Reddit post gets right, and where it overreaches.

Verified

The ADA invested in Overjet and Pearl.

The ADA says both investments came from reserves, were financial rather than endorsements, and were recommended through its Innovation Advisory Committee.

Verified

Overjet acquired American Dental Examiners.

Overjet publicly describes the 2021 acquisition as combining AI with licensed nationwide claims review, claim-selection, and payment-integrity services for payers.

Unverified

Overjet caused denials by the named insurers.

The public sources reviewed do not establish that Overjet or ADE caused specific Delta, MetLife, United Concordia, Humana, or Principal denials.

Unsupported

The ADA executive director left because of these investments.

The ADA announced Raymond Cohlmia’s resignation without connecting it to Overjet, Pearl, or the investment program. The causal claim should not be repeated as fact.

Clinical risk

A colored box is not a treatment plan.

FDA-cleared dental AI products are assistive readers. Pearl’s clearance explicitly says its product is not a replacement for complete dentist review, patient history, or in-vivo assessment. Independent literature reaches the same practical conclusion: AI can improve sensitivity, especially for early findings, while also increasing treatment recommendations. Real-world deployment evidence remains thinner than marketing suggests.

False positives are prevalence-sensitive

A system can show attractive sensitivity and specificity yet produce a burdensome number of false alerts in a low-prevalence population.

Detection is not disease activity

Radiographic appearance alone does not determine lesion activity, restorability, symptoms, patient risk, or whether intervention is warranted.

Standardization cuts both ways

Consistent overlays can reduce missed findings, but organization-wide thresholds can also normalize overcalling if governance is weak.

Procurement protocol

Run a local validation before signing.

Define the intended job
Caries screening, bone-level measurement, patient communication, chart review, or claims documentation are different jobs. Do not buy one vague promise called “AI.”
Use your own historical cases
Build a representative, de-identified sample across sensors, clinicians, ages, restorations, lesion depths, and image quality. Include normal cases.
Blind the comparison
Have calibrated dentists establish a reference before exposing the AI output. Adjudicate disagreements instead of letting the software become ground truth.
Measure the errors that matter
Track sensitivity, specificity, positive predictive value, negative predictive value, false positives per image, missed disease, and changes in treatment recommendations.
Audit workflow and governance
Review edit controls, audit logs, model updates, data retention, secondary use, BAA terms, exportability, and separation between provider and payer products.
Set a stop rule
Pause the pilot if positive overlays increase irreversible treatment recommendations without stronger clinical confirmation or if staff begin treating AI marks as diagnoses.

Demo questions

Make the vendor answer these in writing.

Which exact FDA-cleared product and version is producing each overlay?
Which ages, image types, sensors, findings, and exclusions are within the intended use?
Can thresholds be calibrated, and can clinicians reject or edit every result?
How are model updates validated and communicated?
Are practice images or decisions used for training, payer review, benchmarking, or another product?
Are provider-facing and payer-facing data, teams, models, and contracts segregated?
Can the practice export original images, overlays, audit logs, and disagreement records?
What evidence exists from independent, multisite, real-world deployment rather than company-sponsored validation?

Primary evidence

Read the source documents, not the sales deck.

FDA: Overjet Caries Assist 510(k) summary Indications, dataset, reader study, sensitivity, and specificity. FDA: Pearl Second Opinion 510(k) summary Indications, standalone performance, and reader-study results. FDA: Videa Caries Assist 510(k) summary Indications and standalone adult-dataset performance. Overjet: American Dental Examiners acquisition The company’s description of its payer claims-review business. ADA: investments in Pearl and Overjet The ADA says the investments are financial and not product endorsements. 2025 review: concerns in dental AI deployment Why validation metrics alone do not establish real-world clinical usefulness. 2026 independent Overjet validation Supports using AI as a screening adjunct within multimodal diagnosis. Peer discussion that prompted this review Useful questions and allegations, not a verified evidence source.

OnlyDentists has no financial relationship with Overjet, Pearl, Videa, the ADA, or the insurers named above. This page is educational and does not make a diagnosis, endorse a product, or determine insurance coverage.