Reverse image search is the cheapest forensic check that is almost never done. It takes 30 seconds per photo, costs nothing, and catches a specific fraud pattern that metadata checks miss entirely: the customer who grabs a stock photo of a scratched screen from Google Images and sends it as evidence. This guide covers which tool to use for which job, how to interpret results, and the one operational trick that makes this scale.
What Lift-and-Submit Fraud Looks Like
A customer orders a €120 pair of wireless headphones. Three days after delivery they open a ticket: "Arrived damaged, won't charge, please refund." Attached: a photo of a very convincingly scratched pair of headphones. The photo has no EXIF (the customer says they had to screenshot it because the app glitched). The photo looks real — because it is real. It was taken by someone else, uploaded to a review site two years ago, and your customer pasted it into your claim. The photo isn't manipulated. The fraud is that it is not of your shipment.
This pattern is why reverse image search matters. EXIF forensics catches doctored photos. Pixel forensics catches edited regions. Neither catches a photo that is 100 % authentic — just from someone else's headphones.
The Four Tools Ranked By Use Case
| Tool | Best at finding | Free tier | API for automation |
|---|---|---|---|
| TinEye | Re-uploads of the exact same image across indexed sites; early-web matches | Limited to 150 searches / month / IP | Yes (paid, from $50/month for 5 000 searches) |
| Google Lens | Shopping sites, news images, editorial photos; visual-similarity (not just exact) | Free, unlimited via browser; API limited | Partial via Google Cloud Vision |
| Yandex Images | Social media (VKontakte, OK, Telegram channels), older Russian-language stock sites | Free, unlimited via browser | No official public API |
| Bing Visual Search | Microsoft properties, some stock libraries; weaker than the others overall | Free, unlimited via browser | Yes via Bing Image Search API |
In our dataset, running all four in sequence catches ~80 % of lifted photos. Running only Google Lens catches ~55 %. Running only TinEye catches ~40 %. The overlap is small, so the four-in-sequence workflow has diminishing returns fast — three is usually enough.
A 30-Second Workflow
Reverse-image-check a damage claim photo
Run tools in this order. Stop if any single tool returns a hit older than the order date.
- Save the photo locally, right-clickMost browsers let you 'Search the web for this image' via right-click (Chrome, Edge). On other browsers, save the photo, then drag-drop into the search tool.
- TinEye — tineye.comDrag-drop the file. Wait 2–3 seconds. Results page shows hits ranked by how closely they match. Click 'oldest' to sort by earliest-seen date. Anything predating the order is a smoking gun.
- Google Lens — images.google.comClick the camera icon in the search bar, upload or drag-drop the file. Google surfaces visually-similar images plus sites that embed them. Scroll for shopping-site matches; they are the strongest tell.
- Yandex Images — yandex.com/imagesCamera icon in the search bar, drop the file. Yandex's index is visibly stronger on older web content and social media than Google's. Skim the top 12 results.
The workflow pays off because in 9 of 10 lift-and-submit cases, the image has been online for months or years under someone else's URL. You do not need forensic expertise — if the Google Images result says "first seen on reddit.com/r/audiophile in 2022", the photo is not of the customer's 2026 purchase.
Reading the Results
| Finding | Verdict | Next step |
|---|---|---|
| No hits across any engine | Inconclusive — photo is new or the index has not crawled it yet | Continue with other forensic checks (EXIF, ELA) |
| Hit on a stock-photo site (Shutterstock, Unsplash, Adobe Stock) | High probability of fraud — stock photos are not customer evidence | Reject under policy; document hit URL and first-seen date |
| Hit on a past eBay / Amazon listing by a different seller | Very high probability — lift from a competitor's listing | Reject under policy; screenshot the competitor listing |
| Hit on a review site, blog, or forum post predating the order | High probability — customer grabbed a review photo | Request original file; expect customer to drop |
| Hit on the customer's own public social media, predating the order | Low probability of fraud, high probability of double-dipping | Escalate to human reviewer; cross-reference with order history |
| Hit on a news article about product defects (class action, recall) | Medium probability — could be genuine customer of affected batch | Continue with other checks; do not auto-reject |
Pay close attention to the first-seen date. A photo first indexed in 2019 cannot be of a 2026 purchase regardless of how the customer explains it.
Scaling Beyond 100 Photos a Day
Manual reverse-image search caps at ~50–80 photos per reviewer per day. Above that volume three options:
- TinEye API. Paid tier with programmatic access, $0.01–0.02 per search at bulk rates. Easy integration in a support-ticket webhook — photo attached, API hit, result logged in the ticket.
- Google Cloud Vision Product Search. Purpose-built for matching photos against your own product catalogue, which is a slightly different problem (detecting item-swap fraud) but useful adjacent.
- Automated pipeline (e.g. Claimscan). Combines reverse-image search, perceptual hashing and cross-customer duplicate detection behind one API call. A new upload is checked against the open web, prior submissions across the network, and known fraud-photo databases — in one pass.
Option (3) also catches the pattern a single-engine search alone misses: the same photo submitted as a damage claim to three different stores in the same week, by three different customers. That cross-store reuse signal is decisive but only visible if you aggregate across stores.
Edge Cases That Look Like Fraud But Aren't
| Result | Why it might be legitimate |
|---|---|
| Photo appears on a Pinterest board | Pinterest users often re-pin product photos from merchant sites — including yours. Check if the original pin is from your own store's product page. |
| Photo is on a consumer-review site like Trustpilot | The customer may have posted a genuine review months ago and is now claiming damage on the same product. Cross-reference to their account if it exists on the site. |
| Photo is on an influencer's Instagram, but the account is the customer's | Not fraud in the lift sense; potentially influencer-wardrobing (see the [wardrobing guide](/en/blog/wardrobing-detection)). |
Rule: reverse-image hits are strong evidence, but verifying who uploaded the image first still matters. A photo on the customer's own old social-media post is not lift fraud; it is usually a forgotten re-post.