Checking potentially AI-generated media
Recycled images are only part of the verification problem. Synthetic pictures, videos and voices can depict events that never occurred. Do not build the decision on whether the media looks convincing: use a repeatable procedure that investigates origin, context and independent evidence.
The essentials
Visual defects are not a dependable foundation for verification. Unusual hands, strange backgrounds or an awkward blink may prompt a question, but the absence of a visible flaw proves nothing. Start with a procedure that can be applied equally to material that looks suspicious and material that appears completely ordinary.
Ask four questions in order. Where did it come from? Trace the earliest publication, account and history. Who else has it? Look for independent recordings or reliable evidence of the event. Does the context hold? Check location, date, public schedule and other verifiable details. What have established fact-checkers found? Consult their work before sharing.
A dramatic clip with an unclear origin or no corroborating trace requires investigation, not immediate publication. Missing evidence should be described accurately rather than converted automatically into proof of fabrication. Urgency can favour deception: take the time required to document the decision and explain the limits of what you can establish.
Take it to the field
Use the same procedure for convincing and suspicious media: origin, other traces, context and corroboration. The absence of a visible flaw is not proof. Base the decision on evidence you can document.
Your turn
Apply the four questions to three videos circulating this week. Record the earliest trace, independent sources, contextual checks and fact-checkers consulted. Time the process and note what remains unresolved.