How to spot a deepfake video or voice clone
A deepfake is video or audio made with AI to show someone saying or doing something they never did. Some are easy to spot. The best ones are not. This guide shows what to check, and what to do when you cannot be sure.
1. Start with the source
Before you study faces and voices, ask where the clip came from. This step settles most cases.
- Is it on the person's own verified account or website?
- Have established news organisations reported it, with their own sources?
- Did it first appear on an anonymous account, or one that was created recently?
- Does it ask you to do something urgently, such as invest, donate, vote a certain way or click a link?
Fake videos of well-known people are often used to sell scams, especially investment and cryptocurrency schemes. If a celebrity appears to endorse a get-rich-quick offer, assume it is fake until proven otherwise.
2. Check it is not an old clip
Many misleading videos are not deepfakes at all. They are real clips from another time or place, shared with a new caption.
Take a screenshot of a few clear moments and run a reverse image search with TinEye, Google Lens or Bing Visual Search. If the same footage appeared years earlier, or somewhere else, you have your answer. RELTY calls this misleading context.
3. Watch for visual clues
Play the clip at the highest quality available. Watch it again slowed down, and pause on frames where the person moves.
- The mouth. Lip movements that do not quite match the words, or teeth that blur into one shape.
- Face edges. Blurring or flickering where the face meets the hair, ears or neck, especially when the head turns.
- Things in front of the face. Glitches when a hand, microphone or pair of glasses passes across it.
- Lighting. A face lit differently from the room, or skin tone that changes between frames.
- Movement. A head that looks pasted onto a body, stiff posture, or blinking and expressions that feel mechanical.
- Details. Earrings, collars and hairlines that change shape from moment to moment.
These are clues, not proof. AI video improves quickly and many of these mistakes are disappearing. Ordinary video can also show glitches from poor connections, compression or beauty filters.
4. Listen for audio clues
- A flat or unusual rhythm, with odd pauses or stress on the wrong words.
- Little or no breathing, or breaths in strange places.
- Background sound that cuts in and out, or is unnaturally clean.
- Phrases the person would not normally use.
Cloned voices can sound convincing after only a short sample of someone's real speech. Do not rely on your ear alone.
5. Voice-clone calls and messages
The most direct risk for most people is a call or voice note that sounds like a family member, friend or boss. It usually asks for money or information, and it is usually urgent. The caller may say they are in trouble and cannot talk for long.
If a familiar voice asks for money or passwords:
- Hang up, then call the person back on a number you already know.
- Ask a question only the real person could answer.
- Agree a family code word in advance, and ask for it.
- Never pay by gift card, cryptocurrency or a bank transfer you cannot reverse because someone is rushing you.
If you think you have been targeted, contact your bank straight away, then report it to the police or your national fraud reporting service.
6. Detection tools
Specialist services such as Reality Defender, Hive and Sightengine analyse video and audio for signs of AI manipulation. They are mostly sold to businesses through an API, and some offer limited free checks.
Like image detectors, they can be wrong. They work best on original, high-quality files. Re-recorded, cropped or heavily compressed clips are harder to judge. If you try more than one tool, read each result on its own. Disagreement means the answer is unclear.
A video can also carry Content Credentials that record how it was made. If it has them, Content Credentials Verify can read them. If it does not, that proves nothing either way.
Also search Google Fact Check Explorer. Viral deepfakes of public figures are often checked within hours.
7. Decide what you can say
- Verified. The clip comes from a reliable source and independent reports back it up.
- AI-generated or altered. The person, the creator, credentials or strong technical evidence show it was faked.
- Misleading context. The footage is real but is being used to tell a false story.
- Unverified. There is not enough evidence to be sure.
If you cannot verify a clip, say so, and do not share it as fact. With voice calls, you do not need to prove anything. If you are unsure, end the call and check through another route.
In the news
Recent reports from RELTY's AI news that show why this check matters.
- 13 November 2026 · DW
US seizes servers behind a North Korean deepfake video-call scam
DW reports the FBI says the scam used fake video meetings and deepfakes to steal more than $400 million in cryptocurrency. A face on a live call can be faked too, so check requests for money through another route.
- 10 November 2026 · Full Fact
Videos of tradesmen backing a van driver over an England flag were AI-made
Full Fact spotted vans with identical number plates, distorted letters and a foot that merges into a leg. Small details in the background can show a video is fake.
- 7 November 2026 · Newtral
Video of a woman crying over the Madrid camp eviction was made with AI (in Spanish)
Newtral says Instagram labels the clip as AI, the microphone carries a fake La Sexta logo, and the shop behind her does not match Sol. Details around the person can give a fake away as much as the face.
- 6 November 2026 · Maldita.es
Viral video of an immigrant woman and a Vox vote was made with AI (in Spanish)
Maldita.es cites an AI label, metallic-sounding voices and a voice that carries on after the microphone is taken away. Listening closely can be as useful as watching.
- 27 October 2026 · Full Fact
Video of people collecting iPhone boxes in Hormuz was made with AI
Full Fact found a SynthID watermark and visual glitches. One man appears barefoot, then in shoes, and boxes appear in his arms that he never picked up. Watching closely can reveal mistakes that a quick look misses.
Sources
The pages this guide relies on, checked on 25 October 2026.
- Criminals use generative artificial intelligence to facilitate financial fraud (public service announcement). FBI Internet Crime Complaint Center, December 2024.
- Content Credentials Verify. Content Authenticity Initiative.
- Fact Check Explorer. Google.
- Reality Defender. Reality Defender.
- Hive AI detection. Hive.
- Sightengine. Sightengine.
How to spot a voice-clone scam call
How to tell if text was written by AI
What to do if someone makes a deepfake of you
Why AI detectors disagree, and how to read their results
Browse the RELTY directory of tools
Links to other organisations are for reference only. RELTY has no partnership with, or endorsement from, any of them.
