How to Spot AI-Generated Fakes: A Practical Checklist for Images, Video, Audio and Text

Start with the source, not the pixels

• Who posted it first, and do they have a track record?
• Is a reputable news organisation reporting the same event with its own footage?
• Does the claim fit what you know, or does it conveniently trigger anger or fear?
• Search for the story or image before you react. If nobody credible mentions it, be cautious.

Images: what to look at

• Text and signs that look garbled, or logos that are slightly wrong.
• Hands, teeth, jewellery, glasses and ears that do not quite make sense.
• Shadows and reflections that do not match the light source.
• Backgrounds that melt, repeat or contain impossible details.
• Skin and fabric that look unnaturally smooth or glossy.

Video: extra clues

• Lip movements that drift out of sync with the words.
• Blinking, head turns or facial expressions that feel mechanical.
• Blurring or flicker around the edge of the face, hair or neck.
• Very short, tightly cropped clips with no wider context.

Audio and voice clones

Cloned voices can sound convincing on a short phone call. Be especially careful when a call or voice note asks for money, passwords or urgent secrecy, even if it sounds like someone you know.

• Hang up and call back on a number you already have.
• Agree a family or team code word for emergencies.
• Be suspicious of pressure, urgency and unusual payment methods.

Text: harder than it looks

AI detectors for text are unreliable and can wrongly accuse real writers. Instead of guessing who wrote something, check whether the facts, quotes and links are real. Fabricated citations and statistics are a more dependable warning sign than writing style.

Tools that can help

• Reverse image search to find the original or earlier versions of a picture.
• Fact-checking websites run by news organisations and independent groups.
• Content credentials or provenance labels, where a platform or camera provides them. Their absence does not prove something is fake.
• Detection tools can give a hint, but treat their scores as one input, not proof.

Before you share

• Pause: strong emotion is a signal to slow down.
• Verify with at least one trustworthy source.
• If you cannot confirm it, do not share it, or say clearly that it is unverified.
• If you are targeted by a fake of yourself or your business, save the evidence and report it to the platform.