A repair photo that came from someone else’s website: what a reverse image search finds, and what it cannot

Most of what gets written about fake repair photos is about AI. The older trick needs no software at all. You find a picture of a finished water heater, a patched ceiling or a new roof on the internet, save it, and attach it to the invoice.

The people who look for this in insurance claims say it happens. Writing in Claims Journal in May 2023, Shane Riedman of Verisk, which sells anti-fraud tools to insurers, said that some “independent adjusters, repair contractors, and auto body shops might re-use images or download photos from the Internet to inflate bills or cut corners on providing estimates.” In one carrier’s claims his team found “a single property appraiser used the same photo in 170 different claims over two years.” In another case, a claimant “submitted hand burn photos she had downloaded from a Colorado hospital website.”

Those are insurance cases, reported by a company with products to sell, and they do not tell you how often this happens on a property manager’s invoices. What they do show is that the check for this is different from the check for AI. So here is how it works, and where it stops.

Two kinds of recycled photo, two different checks

• A photo from your own files. The contractor used the same picture on an earlier invoice, for another unit or another owner. We covered this in the invoice article: put last month’s photos in the same batch as this month’s, and our checker marks any two files that are identical to the byte.
• A photo from someone else’s website. It was never in your files, so no comparison with your files can find it. You need a search engine that has already seen that picture online.

What a reverse image search does

You give a search engine a picture instead of words, and it looks for the same picture on the pages it has visited. Two free ones are worth knowing, and they work differently.

TinEye looks for copies. Its help pages say it finds “exact and altered copies” of your picture, “including those that have been cropped, color adjusted, resized, heavily edited or slightly rotated.” It also says it “cannot recognize the contents of any image”, so it will not find a different photo of a similar water heater. It does not use the file name or the file’s metadata, and it says “we never add your image to the TinEye index.”

Google’s search by image casts a wider net. Its help page says the results can include “similar images” and “websites with the image or a similar image.” That finds more, and it also returns more pictures that just look alike. A page of similar bathrooms is not a match. What you are looking for is your picture, the same tiles and the same scuff on the same door frame, on a page dated before the job.

A real hit is strong evidence. If the “after” photo on a March invoice sits on a plumbing company’s gallery page from 2022, that question has been answered, and nobody needed a percentage to answer it.

Why no match proves nothing

This is the part that gets forgotten. A search engine can only find what it has already indexed.

• A photo from a private album, a closed social media account or another client’s job was never on a public page, so no search will find it.
• A photo the contractor took themselves at a different property is a real photo. It just isn’t of your unit, and no search engine knows that.
• A picture generated for the occasion has never been anywhere, so there is nothing to find.

So the result of a reverse image search is either “found, and here is where” or “not found”. “Not found” is not “original”. It is the same rule as everywhere else on this site: absence of evidence is not proof.

What the file itself says about a copied photo

A photo saved off a website often arrives with nothing left in it. Many sites rebuild the pictures people upload, and a rebuilt file usually loses its camera data along the way. Our checker reports that as “Can’t Verify”. That fits a copied photo, but it fits a screenshot or a picture sent by text just as well, so on its own it tells you nothing about where the picture came from.

Some sites keep the camera data. Then the checker shows the camera and the time on its clock, and that can be the most useful line on the page. A camera clock that reads four years before the job, or a camera model the contractor has never owned, is a specific question the contractor has to answer. Camera clocks can be set wrong and camera data can be edited, so it is a question, not a verdict.

What to do with an invoice you are unsure about

1. Ask for the original files, attached to an email or uploaded, not screenshotted and not sent through a messaging app. Here is how to ask so the camera data survives.
2. Check them in one batch with earlier invoices from the same vendor. That catches the same file sent twice.
3. Run the “after” photos through a reverse image search. It is free and takes a minute. Before you upload a photo of the inside of someone’s home, read the search engine’s terms on what it keeps.
4. If nothing turns up, go and look. A photo you cannot place is a reason to see the work, not a reason to accuse anyone. For work you cannot easily visit, ask for photos that could only come from your unit: the work-order number on a piece of tape in the frame, or the unit number on the door in the same shot.

The limits, stated plainly

• Our checker does not search the internet. It reads what is inside the files you give it, and it compares only those files with each other. It uploads nothing, and it stores nothing.
• It gives no confidence score and it does not judge a picture by looking at it. “Can’t Verify” is not a clearance and it is not an accusation.
• A reverse image search finds only what has been published and indexed. No match proves nothing.
• The cases quoted above are from insurance claims. We have not found published figures for copied photos on property maintenance invoices, and we will not invent one.

---

ImposterShield reads the evidence inside image and video files. It runs in your browser and uploads nothing. If a file has been stripped, it says so. That is the product working, not failing.