AI Fake News: How to Check a Claim, Image, or Screenshot
About this check
Saw a post that may be AI fake news? Start with two separate checks: is the image or video likely authentic, and is the claim itself supported by evidence? A useful finished answer should tell you whether the visual is likely authentic, likely AI-generated, or unverifiable—and whether the claim is true, false, or inconclusive. It should also show a confidence score and any sources the check found, so you can check the reasoning rather than rely on a label alone.
The manual method
1. Preserve what you saw
Save the image or screenshot, copy the exact wording of the claim, and note where it appeared. Keep the original post or page available if you can. Small changes in wording can change what needs to be checked.
If the post contains a screenshot, separate the text inside the image from the account or page that shared it. The screenshot may show a claim, but it does not by itself establish who wrote it, when it was written, or whether the surrounding context was preserved.
2. State the claim precisely
Turn the post into one checkable sentence. Avoid combining several claims into one question. For example, ask whether a particular event happened, whether a person made a particular statement, or whether an image depicts a particular place or moment.
A precise question makes it easier to compare the claim with relevant reporting, official information, and other available evidence.
3. Find the earliest available source
Look for the original post, image, video, statement, or report rather than relying only on reposts. Check the visible date, account, caption, and surrounding text. Compare versions of the same material for changes in cropping, captions, or wording.
This step can reveal that a real image is being used with a false caption, that an old event is being presented as new, or that a screenshot leaves out important context. It can also fail: the original may be deleted, private, unavailable, or impossible to identify from the copy you received.
4. Check independent sources
Search the exact claim and key details. Compare more than one relevant source, and distinguish reporting about an event from commentary about it. Give particular attention to primary material when it is available, such as a direct statement or an official record.
Do not treat search-result position, repetition, or a large number of matching posts as proof. Many posts can repeat the same unsupported claim. On the other hand, a lack of search results does not by itself prove that a claim is false.
5. Examine the image or video separately
Inspect faces, hands, text, shadows, reflections, edges, and objects that appear inconsistent. Read visible signs and captions carefully. Compare the visual with other images or footage of the same subject when those are available.
These observations can raise questions, but they do not always settle authenticity. A compressed file, cropped screenshot, low-resolution image, or short video may remove the details needed for a reliable conclusion. A polished or unusual appearance is not enough on its own to establish that something was AI-generated.
6. Keep the conclusion proportional to the evidence
Record what the evidence supports and what remains unknown. A claim may be true, false, or inconclusive. An image may be likely authentic, likely AI-generated, or unverifiable. If the evidence cannot settle the question, the honest result is uncertainty—not a forced yes or no.
Where manual checking stops working
Manual research is useful for understanding context, but it depends on what you can find and interpret. A deleted source, an altered screenshot, missing metadata, a recycled image, or conflicting reports can leave important gaps. Visual inspection can identify suspicious details without proving how a file was created. Search can surface relevant sources without proving that a claim is accurate.
That is why the answer should include the evidence behind it. Without the original context and sources, a confident-sounding conclusion can be harder to evaluate than an explicit “inconclusive.”
How VOM handles the check
VOM assesses an uploaded image as likely authentic, likely AI-generated, or unverifiable. It reads text inside a screenshot, so a forwarded image can be checked without retyping it.
For a checked claim, VOM provides a verdict, a confidence score, and any sources the check found. Its verdict can be true, false, or inconclusive; when the evidence cannot settle a claim, it reports inconclusive rather than forcing a yes-or-no answer. VOM verifies claims using real-time web search and shows any sources the check found.
VOM also offers local, world, and trending feeds of circulating claims, answers in the language a person is using across its eight supported languages, and supports natural back-and-forth conversations for everyday questions, learning, brainstorming, writing, and planning.