No single trick can tell you for certain whether an image or video is AI-generated. Today’s tools make fakes that fool the eye, so the reliable method is to combine several checks: look at the source, look for provenance marks, look for visual and audio flaws, and trace where the file came from. This guide shows you how, step by step.
- Eyes alone are not enough. Obvious flaws such as extra fingers are becoming rarer, so do not rely on them.
- Use the SPOT routine: Source, Provenance, Oddities, Trace.
- Provenance marks help, but their absence proves nothing. Screenshots and many social apps remove them.
- AI detector tools give a clue, not a verdict. They make mistakes in both directions.
- When money or reputation is at stake, verify through a channel you already trust, such as a call-back to a saved number.
Fake images and videos are now part of everyday fraud
In December 2024, the FBI issued a public warning that criminals use generative AI to commit fraud on a larger scale. It described fake social media profiles and photos, fake identity documents such as driver’s licenses and badges, cloned voices of relatives and public figures, and staged live video chats in which scammers pose as company executives or officials.
The numbers show the trend. The FBI’s 2025 Internet Crime Report included an artificial intelligence section for the first time, with more than 22,000 complaints involving AI and losses of about $893 million (roughly AED 3.3 billion). In November 2025, the UAE Cybersecurity Council warned the public to be vigilant against deepfake videos and audio clips, saying the technology can replicate real people’s voices and appearances with high accuracy.
You do not need to become a forensic expert. You need to know when to be suspicious, how to run a few quick checks, and when to stop and verify before acting or sharing.
The SPOT routine: four checks that work together
Each check can be wrong on its own. Together they give you a much more reliable picture.
S: Source. Who posted it, and where did it first appear?
Look at the account that shared it. Is it new, anonymous, full of reposts or focused on one emotional topic? Check whether trusted news organizations, official accounts or the people shown have shared the same image or video. If a dramatic image appears only on obscure accounts, be careful.
P: Provenance. Does it carry credentials or a watermark?
Some tools attach signed information (Content Credentials) or invisible watermarks to what they create. Checking for them can confirm that a file was AI-made or edited. A missing mark proves nothing, so treat it as a clue only.
O: Oddities. Do the details look wrong?
Look closely at hands, teeth, text, lighting, reflections and backgrounds. In video, watch lips, blinking and face edges, and listen to the voice. Flaws are a strong sign when you find them, but their absence does not prove authenticity.
T: Trace. Can you find the original or a matching version?
Run a reverse image search, search key words from the claim, and look for fact-checks. A real event usually leaves a trail of independent coverage. A fake often has none, or it traces back to a different image.
If two or more checks raise doubts, do not share it, do not send money and do not act on it until you have verified it another way.
Clues in images
These signs appear often in AI-generated pictures. Newer tools fix many of them, so treat each as a prompt to look harder, not as proof.
Hands, fingers and teeth
Extra, missing or fused fingers, odd joints, too many teeth or teeth that blend together. Check hands holding objects.
Text and logos
Street signs, shirts, posters and labels with garbled, misspelled or half-formed letters. Real brand logos may look slightly wrong.
Light and shadows
Shadows that point in different directions, a face lit differently from its surroundings, or glare with no source.
Backgrounds
Warped straight lines, repeating patterns, melting objects, people whose faces blur or merge in a crowd.
Skin and hair
Skin that is too smooth or plastic, hair strands that merge with the background, earrings or glasses that do not match on both sides.
Reflections and details
Mirrors, windows and sunglasses that do not reflect the scene correctly. Clothing patterns that change across the body.
Too perfect
Everything composed like a poster: perfect symmetry, flawless faces, dramatic lighting. Real photos have clutter.
Context mismatch
Clothing, vehicles, plates, flags or landscapes that do not match the place or time claimed.
| Clue | What it may mean | How much to trust it |
|---|---|---|
| Garbled text on signs | Common in AI images | Medium. Newer tools are better. |
| Odd hands | Classic AI flaw | Medium, and fading |
| Inconsistent shadows | Edited or generated | Medium to high |
| Perfect, plastic skin | May be AI or heavy retouching | Low. Many real photos are retouched. |
| No clues at all | Could still be AI | Not proof of authenticity |
Clues in video and audio
Video fakes are harder, but they often slip in the details that are difficult to keep consistent from frame to frame.
- Lips and speech. Mouth movements that do not match the words, especially on sounds like “b,” “m” and “p.”
- Blinking and eyes. Too little or too regular blinking, or eyes that do not follow the movement of the head.
- Face edges. Flickering or blurring around the jaw, hairline, ears or neck, especially when the head turns.
- Face and body mismatch. Skin tone, lighting or age that differs between the face and the neck or hands.
- Teeth and tongue. Blurred, uniform teeth, or a tongue that barely moves.
- Hands and objects. Fingers that blur when they pass in front of the face, objects that morph in the hands.
- Voice. A flat or oddly even tone, unnatural pauses, odd breathing, or a voice that is too clean with no background sound.
- Word choice. The FBI advises listening closely to tone and word choice, which can reveal a cloned voice.
Low resolution, heavy compression and “bad connection” excuses make fakes harder to judge. Scammers often use them on purpose. Be extra careful with grainy videos that ask you to act.
Provenance: Content Credentials and watermarks
Provenance means the record of where a file came from. Two systems matter for everyday users.
Content Credentials (C2PA)
Signed information stored with a file that can record the tool that made it, the date and any edits. A free viewer such as the C2PA Viewer can read it. If the signature is valid, the file has not changed since it was signed.
SynthID
An invisible watermark that Google DeepMind embeds in the pixels of content from SynthID-enabled generators. It survives screenshots, cropping and compression better than metadata. Google offers checks through its Gemini app and a SynthID detector.
- Screenshots and many social media uploads strip Content Credentials, so a missing label means little.
- Only some generators add watermarks, so an unmarked image can still be AI-made.
- As one provenance tool provider puts it, the absence of provenance signals is not proof that an image is authentic or human-made.
Use these systems as a quick positive check. If you find a valid credential or watermark, you have strong evidence about how the file was made. If you find nothing, move on to the other checks.
Trace the source
Tracing is often the most powerful check, because a real event leaves traces in many places.
Reverse image search
Upload the image or paste its address into Google Images or TinEye. Look for earlier versions, the original context and whether the same image is used for a different story. Our guide on checking whether a website is legit shows more verification tools.
Search the video by keyframes
Take screenshots of several clear frames from a video and reverse-search each one. Also search the key words or event in the claim, with the date and place.
Look for independent coverage
A real event usually appears in several unrelated outlets, official statements or eyewitness posts. If only one account shows it, treat it as unconfirmed.
Check fact-checkers
Search the claim on fact-checking sites such as Snopes. Many viral fakes have already been reviewed.
Check the details in the file
Where available, look at the file’s creation details. Social media apps remove most of this information, so a lack of data is not a red flag by itself.
What about AI detector tools?
Many websites claim to tell you whether an image or video is AI-generated, usually by giving a percentage. They can be a useful extra signal, but they have clear limits.
- They make mistakes both ways. They can label a real image as AI, and miss a fake that was made by a newer tool.
- They age quickly. A detector trained on older generators may fail against new ones.
- Editing breaks them. Cropping, filters, screenshots and compression can change the result.
- A score is not a verdict. Never accuse someone of faking something based on one detector result.
- Check privacy. Uploading an image sends it to a third party. Avoid uploading private or sensitive photos.
The best use is to run one or two detectors as part of the SPOT routine, and only trust a conclusion when other checks agree.
How fakes are used in scams
The celebrity or expert video
A video shows a well-known person recommending a trading platform. The face and voice are generated. Go to the person’s official website or verified accounts to see whether they said it, and read our guide on how to spot a fake investment platform.
The boss on a video call
A “manager” appears on video and asks for an urgent payment or a change of bank details. Verify by calling the person on a known number, and follow your company’s approval process whatever the pressure.
The voice or video of a relative
A caller sounds like your child or parent and asks for money. Hang up and call them directly. Use a family safe word that scammers cannot guess, as the FBI recommends.
The perfect profile photos
Fake profiles use AI portraits or stolen photos. Reverse-search the images, ask for an unplanned video call and be wary of anyone who refuses and moves quickly to money.
IDs, receipts and proof of payment
The FBI notes that AI can create fake identity documents. Never accept an image of an ID, receipt or bank transfer as proof. Confirm payments in your own bank account.
If you have already acted on a fake, follow our first-24-hours checklist.
If you suspect a live video call is fake
Real-time face and voice swaps exist, which is why a call that “looks real” is not enough proof when money or secrets are involved. These actions can expose a fake or buy you time, though none is a guarantee.
- Ask for a movement. Ask the person to turn their head fully sideways, or pass a hand in front of their face. Real-time fakes sometimes glitch.
- Ask something only the real person knows. Not something that appears on social media.
- End the call and call back on a number you already have, or message on a different, trusted app.
- Use a code word agreed in advance for family or for sensitive company requests.
- Slow down. Pressure to act “right now” is the biggest warning sign of all.
UAE and legal notes
- Verify official content through official sources. The UAE Cybersecurity Council’s advice is to check content attributed to public figures or national institutions through official, trusted channels, and not to share material of doubtful authenticity.
- Sharing can have consequences. The Council has cited Federal Decree-Law No. 34 of 2021 on Countering Rumors and Cybercrimes, which it says imposes strict penalties for such offenses. For legal questions about creating or sharing a fake, speak to a licensed lawyer.
- Report cybercrime. In Dubai, call Dubai Police on 901 or use the eCrime platform. In Abu Dhabi, call the Aman service on 800 2626.
- If money was involved, call your bank immediately and see our checklist for the first 24 hours.
Laws and reporting channels can change. Confirm details on the official websites of the authorities.
United States and elsewhere
Report scams that use fake images, video or voice at ReportFraud.ftc.gov and ic3.gov, and use the report tools on the platform where you saw the content.
Protect yourself from being faked
Agree on a safe word
Choose a private word or phrase with family. Use it whenever someone claims to be in trouble and asks for money.
Limit public images and voice
Fewer public videos, voice notes and high-resolution photos means less material for a clone. Review privacy settings on social media.
Verify before you act
For any request that involves money, passwords or secrets, confirm through a second channel you already trust.
Secure your accounts
Strong, unique passwords and two-factor login make it harder for criminals to post as you. Read our guide on checking your email for a data breach.
Pause before sharing
If an image makes you angry or afraid, that is the moment to check it. Do not repost what you have not verified.
Label your own AI content
If you create AI images, say so. It builds trust and helps others avoid confusion.
For more on AI errors in text, see what an AI hallucination is and how to catch one.
Frequently asked questions
How can I tell if an image is AI-generated?
Combine several checks: look at the source, check for Content Credentials or watermarks, look for visual flaws such as odd hands, text and shadows, and trace the image with a reverse search. No single check is certain.
Can AI detectors be trusted?
They can be a useful extra clue, but they make mistakes in both directions. Never rely on one score, and confirm with other checks before drawing a conclusion.
What is a deepfake?
A deepfake is a video, image or audio clip created or altered with AI to make it look or sound like a real person said or did something they did not.
Does a missing watermark mean an image is real?
No. Only some tools add watermarks, and screenshots and social media often remove metadata. A missing mark is not proof of authenticity.
How do I check a suspicious video?
Take screenshots of key frames and run a reverse image search, search for the claim and the date, look for coverage by reliable outlets, and check the poster’s account. For money requests, verify through a call-back.
How can I tell if a video call is a deepfake?
Ask the person to turn their head sideways or pass a hand in front of their face, ask a private question, then end the call and contact them on a number you already know. Pressure to act immediately is a strong warning sign.
What should I do if someone makes a fake of me?
Save evidence, report the content to the platform, report to the police, and speak with a licensed lawyer if needed. In the UAE, report to Dubai Police (901) or the Aman service (800 2626).
Verify before you trust
Learn how to check websites, texts and offers before they cost you.
Sources
- TechTarget: FBI, Criminals using AI to commit fraud on a larger scale (December 2024)
- CyberScoop: FBI IC3 2025 annual report · ic3.gov
- Gulf News: UAE Cybersecurity Council warns against deepfakes (November 2025)
- C2PA Viewer: How to verify an AI-generated image with C2PA and SynthID
- ReportFraud.ftc.gov
Currency conversions use the fixed rate of about AED 3.67 per US dollar. Detection methods change as AI tools improve. This guide is for general information and is not legal advice. See our Disclaimer and Editorial Policy.