How to verify the authenticity of an image online

First question: authentic in relation to what?

Verifying the authenticity of an online image requires separating at least three questions: where the file came from, whether it was modified, and whether it is presented in the correct context. A real photograph can accompany a false story. An edited image may instead be an ordinary illustration that is clearly described as such.

The most useful check therefore starts with the claim being assessed. If a post says a photo shows an event that happened yesterday in a particular city, a valid signature on the file is not enough to prove the place and date. You need independent evidence and a reconstruction of its publication history.

The steps below offer a working method, not an infallible system for recognizing every deepfake. When evidence is insufficient, the correct conclusion may be “unverified.”

1. Find the source and preserve the context

Start with the page containing the image: who published it, with what description, and with which references? If the content is a reshare, look for a link to the earlier publication. A page that cites another page is not necessarily the original source.

Whenever possible, retrieve the file from the source that distributed it. Record the URL, date of your check, and accompanying text. Avoid working only from a screenshot, which can remove useful information and make comparison harder. At the same time, do not confuse the earliest copy you can find with definitive proof of who created the image.

A reverse image search can help find earlier copies. Treat each result as a lead to examine: a date in search results does not necessarily establish the date the photo was taken.

2. Check for Content Credentials

Content Credentials provide a technical channel for examining provenance claims. If you have a compatible file, open it in the C2PA verifier and read the results for the signature, content binding, and available history.

Do not stop at the icon. Look for the declared actions and the version to which the credential applies. If the file includes ingredients or references to earlier content, check which are actually accessible. Do not infer a complete history from an indication that the history is incomplete.

The C2PA specification limits this check to provenance information and its validation; it does not assign a universal truth rating to the scene.

3. Compare the image, caption, and evidence

Examine details that could confirm or contradict the story: signs, the geometry of locations, clothing, environmental conditions, and other material from the same event. The goal is not to collect impressions, but to find a verifiable match between the image and the claim.

Consider this scenario: a photograph of a lake illustrates a recent emergency. The photo may contain no manipulation, but that does not prove it was taken during the emergency. An earlier publication using the same file could instead disprove the date implied by the post.

EXIF information, when present, can provide useful details but is not automatically reliable or complete. Compare it with the source and other available evidence. Do not fill a missing field with an assumption.

4. Use AI image detectors carefully

A statistical detector and a provenance system answer different questions. The first looks for features associated with particular classes of images; the second examines claims linked to the content. A tool's score should not be translated into absolute certainty about every file.

Visual anomalies such as unusual hands, inconsistent shadows, or distorted writing deserve attention, but perspective, compression, and other processes can also cause them. Similarly, an image without obvious flaws is not thereby proven to be a photograph of a real event.

Treat automated results as supporting signals. Do not use them alone to accuse someone of falsifying content or to attribute a production method you do not know with certainty.

5. Record a conclusion proportionate to the evidence

A useful verification should be explainable to a colleague or reader. Separate what you observed from what you inferred, and state which checks were not performed.

To retain a reference to the exact file analyzed, calculate a SHA-256 hash. A hash can identify the same byte sequence later, but it adds no information about the truth of the scene by itself.

For publishers and businesses: make checks easier upstream

Publishers can reduce uncertainty by providing an accessible official source, identifiable versions, and provenance information. A press room should distinguish the original, crop, and new edition rather than silently overwriting every file.

SCE Content Credentials and verification paths can become part of this work. The technical layer complements a clear editorial description and gives the public a better way to check what they are viewing.

Frequently asked questions

If the concern is the description attached to a photo, explore how to verify images taken out of context. To decide whether to forward a story, use the checklist for preventing the spread of fake news.

To examine dates and capture details, learn how to read EXIF metadata. If you publish synthetic content, the guide to C2PA and AI images covers claims and captions from an editorial perspective.

Is an image without C2PA generated by AI?

No. It may be an ordinary photograph without credentials or a copy that lost its metadata. Their absence does not prove the method of creation.

Does a screenshot preserve the proof from the original image?

Not necessarily. It is a new file and may lose information associated with the original asset. For a technical check, look for the file distributed by the source when available.

Can a single tool give me certainty?

In general, that is not a careful approach. First define which claim you are checking, then combine checks suited to that question.