Stop paying refunds on AI-faked damage photos
Customers can now add a crack, a stain or a torn seam to a photo in seconds with free AI tools. HumanMeter Claims has them take the photo live through a link you send, then checks it for AI edits and photos of screens. Your team gets a clear risk level on every claim.
Or email support@humanmeter.app
AI-faked damage is now the fastest-growing kind of return abuse
- Fraud-prevention company Forter calls AI-generated damage claims the fastest-growing form of return abuse, and says it stopped an $800k ring using them (Retail Gazette, June 2026).
- DTC brands report seeing AI-altered claim photos daily to monthly, and some now verify damage over FaceTime calls (Modern Retail).
- Airbnb stopped accepting AI-generated or AI-enhanced photos as evidence in damage claims in April 2026 (The Host Report).
Why detecting fakes after the fact isn't enough
We tested our own detector on 36 product photos in September 2026: real photos, real damage, and damage we faked with Google's image editor.
| Photos | Result |
|---|---|
| 16 real photos, including 6 of real damage | 15 read as real, 1 unclear. No real photo flagged as AI. |
| 10 AI-faked damage photos, straight from the AI tool | 10 of 10 caught, from the file's own content credentials |
| The same 10 fakes after a screenshot or re-save | Only 2 of 10 caught |
One screenshot strips the evidence, and every detector faces the same problem. So we changed the question from "is this file fake?" to "was this photo taken live, of the real item, just now?"
How HumanMeter Claims works
- A customer files a damage claim. Your helpdesk or backend creates a HumanMeter link for that order with one API call.
- They take the photos live. The link opens their phone camera in the browser. There's no upload button, so there's no chance to swap in an edited file. Every photo is timestamped by our server.
- You get a risk level. Each claim comes back Low, Review or High, with the reasons and the photos. The customer never sees the verdict.
What we check on every photo
- Live capture: taken in the camera through your link, not uploaded from a gallery.
- Photos of screens or prints: the usual trick for getting a fake past a live camera.
- AI generation and editing signals, with plain-English reasons.
- Content credentials: the provenance data AI tools like Google's and OpenAI's attach to their images.
Pricing
| Plan | Price | Claim photos per month |
|---|---|---|
| Starter | $49/mo | 100 |
| Growth | $199/mo | 600 |
| Scale | $499/mo | 2,000 |
| Volume | Custom | Marketplaces, warranty and claims teams |
One avoided refund on a $60 order covers the Starter plan.
Questions
Detectors look at a finished file. A fraudster can edit a photo with AI, re-save it, and most detectors miss it: in our own test, re-saved AI fakes got past our detector 8 times out of 10. HumanMeter Claims changes when the photo is taken. The customer opens your link and takes it live with their phone camera, so there's no finished file to edit first. We still run AI and screen checks on every photo.
For every claim: the photos, when they were taken, and a risk level (Low, Review, or High) with plain-English reasons, such as 'Photo was taken of a screen' or 'AI generation signals in the damage area'.
No. The customer only sees that their photos were received. The verdict goes to you, so fraudsters can't learn what gets past it.
Treat High as 'ask for more': another angle, a short video, or the item back. Treat it as a reason to look closer, not as proof. Legitimate customers are unaffected, since they take real photos of real damage.
The link also works in a desktop browser with a webcam. Anything not taken live with a camera is marked for review so your team can decide.
Through a simple API call from your helpdesk or backend. For pilot customers we set up the integration with you.