What is Chassly and how does AI vehicle damage assessment work?
AI vehicle damage assessment is a relatively new way to triage car damage without driving anywhere. You take photos of the damaged area on your phone, upload them, and within about 60 seconds you get back a structured report: which parts are damaged, how severely, what it'll likely cost to fix, and whether the job is realistic to DIY or needs a professional. Chassly is built specifically for that workflow: no chat thread to navigate, no general-purpose model that hedges its answers, just a focused tool that does one thing well.
How the Chassly assessment pipeline works
When you submit photos, Chassly first runs a fast pre-screen using a purpose-built photo vetting layer. Its only job is to confirm each photo actually shows a vehicle. If you accidentally uploaded a photo of your dog or a blank wall, the system rejects it before doing the expensive analysis. This pre-screen takes about 8 seconds and saves you from waiting a full minute only to find out the photos didn't work.
Once your photos pass the pre-screen, the main analysis runs on the deep inspection layer, the most capable vision model our AI provider offers as of 2026. This layer identifies every damaged part visible across all your photos, classifies the damage type (dent, scratch, crack, etc.), rates severity on a 4-point scale (minor / moderate / major / severe), and decides whether each part can be repaired or needs full replacement.
The model is given vehicle context too: your year, make, and model. This matters because a 6-inch dent on a 2018 Toyota Tacoma's steel fender is repairable with paintless dent repair (PDR), while the same dent on a 2022 Tesla Model Y's aluminum panel almost certainly needs replacement. Aluminum work-hardens when struck, and large dents can't be massaged out without cracking the paint or weakening the metal.
If the deep inspection layer times out or hits an error (rare, but possible during AI provider incidents), Chassly falls back automatically to the damage detection AI. You don't see this happen; the report still arrives, just from a slightly faster, slightly less detailed model.
What the AI is actually looking at in your photos
Vision models work by identifying patterns across millions of training images. For damage assessment, the patterns include surface deformation (dents), paint disruption (scratches, scuffs), material fracture (cracks in glass or plastic), missing components (a torn-off mirror cover), and structural distortion (a bent fender that no longer aligns with adjacent panels).
Confidence scores matter. Every detection Chassly returns has an internal confidence value between 0 and 1. Detections under 0.55 are filtered out; they're more likely to be lighting artifacts, water droplets, or shadows than real damage. Detections between 0.6 and 0.8 typically come with a 'Low confidence detection' note in the report.
What the AI does NOT do: speculate about hidden damage. If a fender is dented but the wheel-well behind it is hidden from view, Chassly won't say 'possible suspension damage.' It will report only what's actually visible. The system prompt explicitly guards against this kind of hallucination because confident-sounding speculation is worse than honest uncertainty.
Cost estimates: how Chassly arrives at the numbers
Each damaged part the AI identifies is run through a cost estimator that combines a base part cost (informed by repair industry pricing data) with severity and vehicle-class multipliers. A bumper repair on a base-trim economy car is multiplied by 0.8x; the same repair on a luxury vehicle gets a 1.6x multiplier because parts and labor both cost more. Severity rolls into the same calculation: a 'major' bumper dent is multiplied by 1.5x relative to 'moderate.'
The result is a min/max range, not a single number. The range reflects real-world variance: body shop pricing varies by 30-50% depending on geography, shop reputation, and whether OEM or aftermarket parts are used. The range gives you anchoring data to evaluate quotes you receive from real shops.
You should treat the estimate as a starting point. Always get one or two real quotes from local body shops before committing to a repair. Chassly's range tells you whether a quote is reasonable. If a shop quotes $2,400 to repair damage Chassly estimates at $400-600, that's a signal to get a second opinion.
What you get in the final report
Every report contains the same backbone: a list of damaged parts with severity and recommended repair action, a total cost range, an overall severity assessment, and a top-level repair recommendation (DIY, professional, or either depending on your skill level).
Driver-tier and Business-tier subscriptions unlock additional sections. Driver gets DIY guidance (step-by-step instructions for minor cosmetic repairs you can attempt yourself) and insurance claim analysis (whether to file a claim and what to document). Business adds a downloadable PDF report and a per-part breakdown suitable for fleet managers.
On the assessment result page, Chassly's AI also suggests follow-up reminders: for example, 'Re-photograph this dent in 14 days to check for rust' or 'Schedule a body shop quote within 7 days for the major fender damage.' These suggestions appear as drafts in your /reminders inbox; you accept the ones that matter and dismiss the rest.
Where Chassly's accuracy lives, and where it doesn't
AI assessment is most accurate for visible body damage in the panels, lights, glass, and trim: exactly the kind of damage most owners want to triage after a parking lot incident or minor collision. The model has been validated against thousands of real assessments and consistently identifies the affected parts and severity within a body shop's estimate range.
It is less reliable for damage that's hard to see from the outside: frame integrity, suspension components, mechanical issues from impacts, hidden rust under panels, or alignment problems. For those, Chassly will sometimes flag 'professional inspection recommended' but cannot replace the inspection itself.
The AI also can't smell or hear your car, which matters for diagnostics that go beyond visual damage. If something doesn't feel right with your vehicle after an incident (a vibration, a pull to one side, a new noise), get it checked in person regardless of what Chassly's report says about the cosmetic damage.
Frequently asked questions
How accurate is Chassly compared to a real body shop estimate?
Most quotes from local shops fall within Chassly's estimated range. The range exists specifically to capture geographic and shop-to-shop variance. If a quote is far outside the range (either much higher or much lower), that's worth investigating with a second quote.
Does Chassly work for any vehicle type?
Chassly handles all consumer passenger vehicles: sedans, SUVs, pickups, vans, motorcycles aren't currently supported. The AI was trained predominantly on consumer cars and light trucks. Heavy commercial vehicles and specialty vehicles (RVs, boats, etc.) aren't in scope.
How long does an assessment take?
Typically 60-90 seconds end-to-end. The pre-screen takes about 8 seconds, photo upload depends on your connection (usually 5-10 seconds), and the main AI analysis runs in 30-60 seconds depending on whether the primary model or fallback is used.
What happens to my photos after the assessment?
Photos are encrypted in transit and stored on Cloudflare R2 with a public read URL gated by a custom domain. Basic-tier images are deleted after 12 months as part of automated retention; Driver and Business tier images are kept indefinitely. You can request full deletion of all your data via the GDPR export endpoint in Settings.
Can I run Chassly on damage that's older than the incident date?
Yes. Many users run Chassly months after an incident as part of pre-purchase inspections, insurance claim disputes, or before scheduling repair. The AI doesn't need to know when the damage occurred; it just analyzes what's visible in the photos.