AI video is improving so fast that obvious tells, such as six-fingered hands and melting faces, are disappearing. Still, most AI clips leave clues, and more importantly, there are reliable ways to check where a video came from. Whether you are a creator who wants to understand your medium or a viewer who wants to avoid being fooled, this checklist will help.
Visual clues
Hands and fingers
Still a common weak point. Watch for wrong finger counts, fingers merging, hands changing shape while moving, or objects melting into palms.
Faces and teeth
Look for faces that subtly change identity during a turn, overly smooth skin, mismatched earrings, teeth that blur into a single shape, or eyes with inconsistent reflections.
Text and signs
Background text is often garbled: shop signs with nonsense letters, inconsistent logos or writing that changes between frames.
Background details
Background people may have distorted features, walk strangely or appear and vanish. Patterns such as tiles, fences and windows may warp or fail to line up.
Lighting and shadows
Shadows that point in different directions, reflections that do not match the scene, or light that changes without a source can reveal generation.
Motion and physics
- objects passing through each other;
- liquids and fabric behaving oddly;
- people moving with a floaty, weightless quality;
- objects or limbs morphing into other objects;
- unnatural speed changes or motion that loops subtly.
Duration and editing
Many AI videos are built from short clips of a few seconds each, often with frequent cuts. Long continuous shots with complex action remain harder for current tools. A dramatic event shown only in a series of very short, perfectly framed clips deserves a closer look.
Audio clues
- voices with flat emotion or odd emphasis;
- breathing patterns that do not match speech;
- lip movements slightly out of sync;
- background sound that does not match the setting, or no ambient sound at all.
Context checks: the most reliable method
Visual clues are getting weaker every year. Context checks remain powerful:
- Who posted it first? Search for the original source. Reputable news organisations, official accounts or the actual creator?
- Is it reported elsewhere? A real major event will be covered by multiple independent sources.
- Is there an AI label? Check platform labels and descriptions.
- Reverse search key frames. Screenshot frames and use reverse image search to find earlier versions or the creator’s original post.
- Check the account. New accounts posting shocking clips without context are a red flag.
- Look for the creator’s statement. Many AI artists clearly state their work is AI; reposts often remove that.
Provenance and content credentials
Some cameras, editing tools and AI generators attach provenance information, known as content credentials, to files. Platforms and verification tools can read it to show whether content was created or edited with AI. The absence of credentials does not prove a video is real, since metadata can be stripped, but their presence offers useful evidence.
Detection tools: use with caution
Various tools claim to detect AI-generated video. They can help but are not reliable on their own: they produce false positives and false negatives, especially on compressed or edited clips. Treat their results as one signal among many, never as proof.
Why this matters
Being able to spot AI video protects you from scams that use fake videos of executives or relatives, from misinformation during elections or crises, and from manipulated “evidence”. It also protects real footage: when people know how to verify, genuine videos are less likely to be dismissed as fake.
For creators
Understanding these clues helps you make better work, since many tells are exactly the artifacts you want to avoid, and it reminds you why labelling matters. Clear labels and provenance metadata help viewers trust your work and stop your creative pieces from being misused as fake evidence. See Labelling AI-Generated Video.
A quick checklist
- Check hands, faces, teeth and text.
- Watch the background and physics.
- Listen to voice and ambience.
- Find the original source and creator.
- Look for coverage by reliable outlets.
- Check for labels and content credentials.
- Reverse search key frames.
- Treat detection tools as hints, not verdicts.
- When in doubt, do not share.
Teaching others
Share these checks with family members, especially older relatives and children, who may be more likely to trust a convincing clip. Scams using fake videos of familiar people are becoming more common, and a simple family rule, such as calling back on a known number before sending money, can prevent real harm. Schools and workplaces increasingly include media-verification skills in training, and creators who understand AI video are well placed to explain them.
Practice makes it easier
Spotting AI video is a skill that improves with practice. Watch openly labelled AI videos and look for the clues listed here; you will start to notice typical motion, lighting and texture patterns quickly. Compare them with real footage of similar scenes. Over time you build an instinct for when something feels off, which prompts you to verify before believing or sharing. Combine that instinct with source checks, and you will be far harder to fool than most viewers.
The verification mindset
The most important habit is not technical. It is pausing before reacting. Videos designed to deceive are usually designed to provoke strong emotion: outrage, fear or amazement. When a clip makes you feel an urgent need to share it, that is exactly the moment to check it first. A minute of verification is worth more than any detection trick.
Frequently asked questions
Can AI video be detected with certainty?
Not always by looking. Source verification and provenance data are more reliable than visual inspection alone.
Are watermarks proof of AI?
A genuine provider watermark or credential is strong evidence, but watermarks can also be faked, cropped or removed.
What should I do with a suspicious video?
Pause. Do not share it until you have checked the source. Report clearly deceptive content to the platform right away, and if it targets a real person, let them know if you can.
