Check the review against the product
An AI-writing score cannot tell you whether a reviewer bought a product or was paid to promote it. A real customer may use editing or translation help, and a person can write a false review. Start with details you can compare with the listing and other accounts of using the product.
- Check which item the review describes. Compare its model, size and features with the current listing, especially when several variants share one review page.
- Read the description of use. Look for a task the buyer attempted, a result they observed or a limitation that matters to your purchase. Specific details are leads to check. They do not prove a real purchase.
- Inspect the reviewer's public history where it is available. Note repeated passages and reviews posted close together, while allowing for ordinary explanations such as a product launch.
- Compare accounts that disagree. A complaint about a feature you need deserves a closer look even when the average rating is high.
Why joining reviews does not validate a scan
Cobalynx refuses text shorter than 150 words. Combining several short reviews to pass that minimum does not establish who wrote each one. Reviews on one product page may have different authors and different amounts of editing, and a single result cannot separate them.
We have not published a separate accuracy measurement for product reviews or combined review batches. The rates on the evidence page should not be read as validation of a fake-review test. You can try the text scanner on one longer English review, but its result cannot establish a purchase, payment or deception. The methodology explains what the estimate supports.
What you can learn from a listing image
If you have the image file, the file check can look for Content Credentials. Credentials that verify, record an AI generator and have a signer we recognise are strong evidence about the image's origin. They do not establish that the seller is fraudulent or that a product does not exist.
No credentials tells you nothing either way about whether the image is real. An unverified signer is different from verified credentials, so read the result rather than treating any manifest as confirmation. The Content Credentials guide explains these outcomes.
How to document a concern
Save the listing URL and the individual review links where available. Record the date and the precise issue, such as identical wording across named accounts or reviews describing a different item. Screenshots can help preserve what you saw if the listing changes.
Use the platform's review or listing report control and describe those observations. A report that identifies the material to inspect gives the reviewer something to follow up. Do not present an AI score as proof of paid reviews. The decision about what to trust is still yours to make, using the product evidence and the terms of the purchase as well as the reviews.
Common questions
How many reviews do you need before the scan will work?
There is no validated number of reviews that makes a batch score reliable. Combining short reviews from different authors changes the input and does not establish who wrote any one review. Compare their details, dates and reviewer histories directly instead.
Does a flag mean the reviews were paid for?
No. A score does not prove payment, fabrication or a real purchase. AI assistance and review authenticity are different questions. Report specific observable concerns, such as copied details about the wrong product, rather than presenting a detector score as proof.
Could I check the product photos as well?
Yes. The file check can read Content Credentials. An AI-generator record is strong evidence of origin only when the credentials verify and the signer is recognised. An untrusted or invalid signature cannot support that conclusion. Missing credentials leave the origin unknown, and even verified credentials do not prove that a seller owns or will deliver the item.
Where would I send a report?
When we were looking at how the platforms handle this, what we found was that you do it on the platform where you found the reviews. Amazon lets you report one review at a time, and the link is called “Report abuse”. Google Maps calls it “Report review”. Etsy lets you report a listing or a whole shop, and the app stores will take a report on one rating. When you did it, you would tell them what you saw, which might be that the same phrasing turned up in several reviews or that 20 of them went up in the same week, since if all you said was that it looked fake there would be nothing for the person reading it to go on.