How our AI works

When you submit photos and a short description, our system produces a preliminary triage: likely damage type, likely damage area, an approximate severity, a repairability read (likely repairable as cosmetic vs. may not be repairable in scope), and a routing decision — whether to match you to shops, ask for more photos, send to human review, or politely decline as out of scope.

• It is not a pricing engine. Pricing comes from the matched shops after they review your case.
• It is not a diagnosis. Final scope is confirmed by a human shop, often after an in-person look.
• It is not used for insurance, claims, fraud detection, lending, or underwriting.
• It does not assess collision, structural/frame, airbag, suspension, drivetrain, or mechanical damage.
• It does not identify people or perform facial recognition.

The AI flags its own confidence. When confidence is low, when photos are too few, too blurry, or too far away, or when the damage looks like it might be out of scope, the case is routed to a human reviewer or politely declined with an explanation. We would rather slow down than guess.

• Glare, harsh shadows, or wet paint can make scratches look worse or hide them entirely.
• Photos taken too far away or only from one angle limit what the model can see.
• Pearl, matte, and color-shifting finishes increase the chance the price range is off.
• Damage that overlaps panels or wraps around bumpers often needs in-person inspection.

We do not use your photos or case details to train third-party AI models. We may use de-identified examples internally to evaluate and improve the assessment quality. You can request that your case be excluded from any internal evaluation set by emailing trust@wellkeptauto.com.

You can ask for a human reviewer at any point. Reply to your case email or contact help@wellkeptauto.com.