How to spot AI generated text, images and video
Realistic ways to judge whether content is AI made: what actually gives text and images away, why AI detectors fail, and how provenance and watermarks work.
No test reliably tells you whether a piece of text or an image was made by a machine. What you have instead is judgment: a handful of tells that are genuinely worth something, a set of detection tools worth much less than they claim, and a provenance standard being built into cameras and editing software that may eventually answer the question properly.
Start with the signals in AI written text, which are more about structure and hedging than any single giveaway word, and what to look for in a generated image, where the tells have moved on from hands to lighting, text and background detail. Then read why a detector score is not evidence before you ever act on one, and how Content Credentials attach a history to a file.
The costly mistake is accusing a student, a colleague or a supplier on the strength of a percentage from a detector. If the thing you are judging also asks for money or pushes you to hurry, treat it as a scam question rather than a detection question.
Articles in this hub
- How to tell whether a piece of text was written by AIThe signals that suggest AI writing, why none of them are proof, and how context, drafts and specificity tell you more than any stylistic tell ever will.
- How to tell whether an image was generated by AIThe tells that still work on AI images, the ones that stopped working, and the provenance data that settles it when your eyes cannot.
- AI detectors: why the score is not evidenceHow AI text and image detectors actually work, why false accusations happen, what the error rates mean for a student or an employee, and what to do if you are accused.
- Content Credentials: the provenance label on a pictureWhat C2PA Content Credentials are, what the little label on an image actually proves, where the chain breaks, and how to check a file yourself.