Guide · Updated November 2026
How to tell if a video is AI-generated
A practical, 5-minute checklist for checking any clip — from viral “caught on camera” moments to fake news segments.
Step 1: Check the file
Before you look at a single frame, check what the file says about itself. Run it through the detector. If it has Content Credentials that say it was AI-generated, you have your answer. If it has camera data from a phone (make, model, location), that’s strong evidence it was filmed — though it can still be edited or misleading.
Most clips you find on social media have had this data stripped. That’s normal, and it’s why the next steps matter.
Step 2: Look for the visual tells
- Hands, teeth and textCount fingers, watch hands grip objects, look at teeth when people talk, and read any signs or labels. Text that melts or changes between frames is a giveaway.
- Physics that doesn’t add upLiquids that pour wrong, objects passing through each other, hair or fabric that ignores wind, footsteps that don’t match the ground.
- Things that morphPause and scrub slowly. Background people, cars and jewellery often change shape, merge or vanish when the camera moves.
- Too-perfect lighting and skinGlossy, poreless skin and cinematic light in a supposedly casual phone video are common with Veo, Sora and Kling.
- Clip length and cutsMost generators make 4–10 second shots. A viral “single take” that hard-cuts every 8 seconds is suspicious.
- Watermarks and cropsLook for a Sora cloud logo, a small “Veo” mark bottom-right, or an oddly tight crop where a watermark may have been cut out.
- Audio that’s slightly offLip sync drifting, voices without room echo, or ambient sound that loops are common in generated clips with audio.
- No original sourceSearch for the earliest upload. Real events usually have multiple angles and witnesses; AI clips often trace back to one account.
Step 3: Check the context
- Find the first upload. Screenshot a frame and reverse search it. Check the uploader’s other posts — accounts that post only spectacular clips are a red flag.
- Look for other angles. Real events — especially public ones — get filmed by many people.
- Check the caption against reality. Weather, location, date and news reports should line up.
- Look for a label. TikTok, YouTube, Instagram and Facebook add “AI-generated” labels when they detect credentials or the creator discloses it.
Step 4: Weigh it up
No single test is conclusive. A confirmed Content Credential is close to proof; camera metadata plus multiple independent uploads is strong evidence of a real recording; everything else is a balance of probabilities. When it matters — before you share, report, or publish — say “unverified” rather than guess.
FAQ
What is the fastest way to tell if a video is AI?
Pause it and check hands, text and the background. Then run the file through a detector that reads Content Credentials. Together these catch most AI videos in under a minute.
Can AI videos fool experts?
Yes. The best clips from current models can fool trained eyes on a phone screen. That’s why provenance (where it came from) matters as much as how it looks.
Is reverse image search useful for videos?
Very. Take a screenshot of a key frame and run it through Google Lens, Bing or TinEye. Real events usually turn up from several sources; AI clips often lead back to one account or an AI community.