How to tell if an image is AI generated
Seven checks, from hands and shadows to metadata, Content Credentials, reverse image search and the original source.
Guides
AI detection guides
Step-by-step checks for telling AI-generated images and text from human work, what each check can and cannot prove, and when a detector helps.
How to tell if something is AI generated
How to tell if text is AI generated
Six checks for essays, articles and emails, including version history, citations and how to read a detector score.
Do AI detectors flag non-native English writers?
What research found, our own test on 879 essays by English learners, and how teachers and students should read a result.
What are Content Credentials (C2PA)?
How signed provenance records label AI and camera images, how to check them, and what they can and cannot prove.
The quick version: four checks that work for anything
Whether it is a photo, an essay or a post, the same order of checks gives the most reliable answer. Detectors come last, not first.
01
Check the source
Who made or first shared it, and could they have? A credible source raises confidence; an anonymous new account lowers it.
02
Look for evidence in the file
Images can carry metadata and Content Credentials; documents can have version history. Missing evidence proves nothing.
03
Compare and verify
Search for earlier copies of an image, compare text with the writer's earlier work, and look up any facts and citations.
04
Use a detector as one more signal
A detector gives a model's estimate. Read where the score sits on the scale and treat a middle result as a real answer.
Which guide do you need?
| What you have | Start with |
|---|---|
| A photo or picture shared online | How to tell if an image is AI generated |
| An illustration, product shot or ad image | How to tell if an image is AI generated |
| An essay, report or homework | How to tell if text is AI generated |
| An article, review or email | How to tell if text is AI generated |
| An essay by someone writing in a second language | Do AI detectors flag non-native English writers? |
| An image file with a "cr" label or provenance data | What are Content Credentials (C2PA)? |
| A voice clip, video or song | Not covered yet: we do not check audio or video until a detector passes our own tests. |
How our checks work
Background for reading any result: the engines and scales we use, and where detection fails.
Questions about spotting AI content
Can I ever be certain something was made by AI?
Rarely. The strongest evidence is a verified Content Credential, version history or an invented source. Visual clues and detector scores are estimates, so combine several checks.
Should I run a detector first?
No. Check the source and the file first; they are often more telling and they help you read the detector result. Use a detector when your own checks are inconclusive.
Can I use these checks to accuse someone?
Use them to ask better questions, not to accuse. Talk to the person, ask for drafts or the original file, and never act on a detector score alone.