Scam Detector: How to Check a Claim, Image, or Screenshot
About this check
If you are trying to check something you just saw, the useful answer is not simply “scam” or “not a scam.” A finished check should tell you what the evidence supports, how confident the answer is, and which sources were used. If the evidence cannot settle the question, the honest result is inconclusive.
You can do that check manually, or use VOM to return a verdict with a confidence score and cited sources. VOM’s verdict can be true, false, or inconclusive. It uses real-time web search and shows any sources the check found. For an uploaded image, it assesses whether the image is likely authentic, likely AI-generated, or unverifiable. It can also read text inside a screenshot, so the text can be checked without retyping it.
The manual scam-checking method
The following is practical checking guidance, not a guarantee that every claim can be resolved.
1. Isolate the exact claim
Write down the sentence you are checking. Separate the factual claim from the reaction around it. For example, distinguish “This image shows an event in a particular city” from “This proves the event happened.” The more precisely you define the claim, the easier it is to look for relevant evidence.
2. Find the earliest source you can
Look for the original post, article, document, video, or image rather than relying only on a forwarded copy. Record the date, author or publisher, and surrounding context. If you cannot find an original source, note that as a limit on your check rather than treating the repost as proof.
3. Search the wording and its variations
Search the exact wording in quotation marks, then try distinctive phrases without quotation marks. Add relevant names, places, dates, or organizations. Compare what you find with the original wording. A similar result is not automatically evidence for the exact claim, so check whether it actually addresses the same event or subject.
4. Compare evidence, not just search results
Read the underlying pages where possible. Check whether they identify their sources, show supporting documents, and match the date and context of the claim. Look for more than one relevant source, while remembering that several pages may repeat the same original report. If the available material conflicts, preserve that uncertainty in your conclusion.
5. Check images and screenshots separately
For an image, inspect the surrounding post and look for earlier versions or different captions. You can also try an image-search method to look for other appearances. For a screenshot, read the visible text closely and check names, dates, links, and cropping. Treat image context and image authenticity as separate questions: an authentic image can be used with a false caption, and an image may be impossible to verify from the available material.
6. Set a time limit and record the stopping point
As practical guidance, use a short initial pass—about five minutes—for a simple claim, then allow more time when the subject is disputed, the source is missing, or the image needs additional checking. Stop when you have a clear, supported answer, or record that the result remains unresolved. Manual checking may stop working when the original source is unavailable, the evidence is inaccessible, the wording is too vague, or the available sources contradict one another. In those cases, “inconclusive” is a more responsible result than a forced yes or no.
What VOM returns
VOM is built for the same first question: what does the available evidence say about this claim or image? Its verification response includes a verdict, a confidence score, and any sources the check found. Claims can be reported as true, false, or inconclusive rather than being forced into a binary answer.
For images, VOM can assess whether an upload is likely authentic, likely AI-generated, or unverifiable. It can read text inside a screenshot, allowing a forwarded image to be checked without retyping its contents. VOM also supports natural back-and-forth conversations for everyday questions, learning, brainstorming, writing, and planning. It answers in the language being used across its eight supported languages, and it offers local, world, and trending feeds of circulating claims.