When an image looks suspicious, the goal is not to guess whether it is AI-generated. The goal is to establish whether you can trust the image, its source and the claim being made with it.
That difference matters. A genuine photograph can be shared with a false caption, an old image can be presented as breaking news, and an AI-generated image can sometimes be clearly labelled and harmless. Verification is about evidence, not instinct.
The 7-step image verification check
1. Stop and identify the claim
Ask what the image is being used to prove. Is it claiming that an event happened, that a public figure said something, that a product works, or that a person is who they say they are? You cannot verify an image properly without first identifying the claim attached to it.
2. Check the original source
Look for the earliest identifiable uploader, publisher, photographer or organisation. A screenshot of a screenshot is much harder to assess than the original file or original post. If the image is supposedly from a news organisation, company or public body, go to that organisation directly rather than relying on a repost.
3. Search for earlier versions
Use reverse image search or visual search tools to see where else the image has appeared. An image presented as a current UK event may turn out to be years old, from another country or taken from an unrelated story. Finding an older source can immediately change the meaning of what you are seeing.
4. Check the surrounding context
Compare the image with independent reporting, official statements, maps, weather, dates, landmarks and other available evidence. A technically genuine image can still be misleading when its context is false.
5. Look for provenance and Content Credentials
Where available, Content Credentials can provide tamper-evident information about how a file was created or edited. The C2PA standard describes Content Credentials as cryptographically signed provenance information that can record aspects of a file’s origin and history. Learn more from C2PA.
Content Credentials are useful evidence, but they are not a truth certificate. OpenAI notes that provenance signals can help identify origin but do not guarantee that content is accurate, unedited, legally owned or presented in the correct context. Read the provenance guidance.
6. Use forensic clues carefully
Visual inconsistencies, strange reflections, impossible geometry, repeated textures, lighting errors or unusual text can be useful warning signs. But these clues are becoming less reliable as generative models improve. Compression, editing and poor-quality cameras can also create artefacts in genuine images.
7. Treat detectors as one signal, not the answer
An AI-image detector can contribute evidence, but no detector should be treated as a universal truth machine. Results can change after compression, resizing, screenshots or editing, and detectors may perform differently on content from models they have never seen.
A stronger question than “Is this AI?”
Instead of asking only whether an image was generated by AI, ask: Who published it? Can I find the original? Does the context match? Is there provenance? Do independent sources support the claim?
This is the approach behind Community Smart Hub’s Verify & Trust pathway and our AI Trust Lab.
What Content Credentials can and cannot tell you
- They can: provide signed information about creation, editing and provenance when supported.
- They cannot: prove that a claim made with the image is true.
- They can: help reveal which tools or processes were involved.
- They cannot: guarantee that missing credentials mean an image is fake.
A simple rule to remember
LOOK → SOURCE → SEARCH → VERIFY → PROVENANCE → DETECT → DECIDE.
Do not stop at the first clue. Trust comes from combining evidence.
Continue learning: Verify & Trust · The Five-Layer Shield · Who Proves That a Picture Is Real?
