Digital reference
Is That Image or Video AI-Generated? How to Investigate
Can a single AI detector prove that a viral photo or video is genuine?
No. A detection score is not a definitive authenticity test. Start with the original uploader, the earliest traceable copy and independent reporting. Check Content Credentials when available, but remember they document signed provenance claims, not whether every statement or depicted event is true. Credentials may be absent or lost during sharing.
Four complementary checks
| Check | What it can tell you | What it cannot prove |
|---|---|---|
| Original source | Who first posted it and what they claimed | Whether that claim is accurate by itself |
| Reverse image and keyframe search | Older copies or a different context | Whether every pixel is authentic |
| Content Credentials (C2PA) | Verifiable signed editing/origin assertions if present | Truth of the scene or authenticity of an unsigned file |
| AI detector | A model's estimate for a particular input | A definitive real/fake verdict |
A practical verification sequence
Save the claimed date, location and creator; look for earlier uploads or independent coverage. Compare details with credible external evidence. Search original video frames rather than relying only on the caption. If evidence conflicts or is missing, share as unverified rather than authentic.
- Check context, not just pixels.
- Avoid spreading personal accusations based on a score.
- For urgent or consequential claims, seek established reporting.
What are C2PA Content Credentials?
C2PA defines cryptographically signed provenance metadata that may record creation tools and edits. Valid signatures provide tamper-evident provenance data. Missing credentials are not proof that a file is fake: platforms or editing workflows can strip or separate metadata.
Can perfectly ordinary photos look suspicious?
Yes. Compression, filters, cropping, reflections and ordinary editing can resemble the artifacts people associate with generative images. Visual oddities are prompts to investigate, not conclusive proof.
Frequently asked questions
If a picture has no Content Credentials, is it AI?
No. Many real files have no embedded credentials, and metadata can be removed in ordinary sharing.
Do valid Content Credentials mean the pictured event happened?
No. They support checking signed provenance claims, not the independent truth of a scene or caption.
Can I prove a video is fake from strange hands or lips?
Not reliably. Such signs can be caused by editing or compression and should trigger broader verification.
Should I repost a suspected deepfake with the person's name?
Avoid amplifying unverified accusations or harmful imagery; preserve evidence and verify through trustworthy sources first.
Related tools and guides
Provenance standard and risk references
Sources last checked:
- C2PA — Frequently asked questions
- C2PA — Content Credentials explainer
- NIST — Generative AI Risk Management Profile
This independent AI literacy reference does not guarantee accuracy, privacy or earnings. Provider terms, local rules and sources may change. Verify primary sources before acting.