Newsletter Subscribe
Enter your email address below and subscribe to our newsletter
By Chris Clark, founder of Senior Tech Cafe
Spot AI-Generated Images and Videos: A Verification Guide
Spot AI-generated images and videos by checking three things. Where the media came from, whether anyone trustworthy is showing the same event, and whether the file’s details match its story.
Strange hands, garbled signs, stiff movement and audio that doesn’t quite fit are all clues. They’re also the clues newer tools have learned to avoid.
So treat every clue as a reason to investigate rather than a verdict.
And never send money or personal information because of one image, one video, or one voice message.
Your eyes aren’t the test anymore. The source is.
What AI-Generated Images and Deepfakes Actually Are
An AI image generator has studied millions of real photos. Ask it for a picture and it builds something new from what it learned, not a copy of any single photo, but a convincing original.
Apply that same idea to video and audio, and you get a deepfake: existing footage or audio altered so a real person appears to say or do something they never did.
AI-generated images are built from scratch from a text description. Deepfakes typically swap or alter a real person’s face or voice inside footage that was originally real.
Both exist for plenty of legitimate reasons, from film effects to accessibility tools. The problem isn’t the technology. It’s how it gets used against you.
How to Spot AI-Generated Images and Videos
Start with the source, not the pixels. Who posted this first, when, and with what evidence?
Then search a distinctive frame, a quoted phrase, the person, or the event in a separate tab.
Check whether established news outlets, official agencies, or the person’s own verified account are showing the same material.
Only then look at the details. Lighting, reflections, shadows, anatomy, background objects, signage, lip movement, whether the audio runs continuously.
One oddity proves nothing. Several inconsistencies alongside a source you can’t verify is enough to withhold both belief and a share.
And if a huge story exists only in one post, that’s your answer.
Image Clues That Deserve a Closer Look
Zoom in on the places AI still struggles. Hands, ears, eyeglasses, jewelry, repeating patterns, reflections, labels, and the faces in the background of your picture.
Look for objects melting into one another, lighting that changes direction halfway across the frame, or text that doesn’t match the country it’s supposedly in.
Then run a reverse-image search (searching by the picture itself) to see whether the image is older than claimed, cropped, or lifted from a different event entirely.
That last one catches more fakes than any visual detail does, and it’s exactly how you’d catch a stolen profile photo used to recognize romance scam profiles.
And remember that compression, editing and ordinary camera errors produce strange details too. A weird hand is a hint, not a verdict.
Video and Audio Checks
Watch it more than once, with the sound on and then with the sound off. You notice different things each way.
Look for abrupt cuts, reflections that don’t match, objects that vanish between frames, unnatural edges around a face, and speech that doesn’t fit the mouth or the room.
Then go looking for a longer version. A completely real video can still mislead when it’s clipped short or given a false caption, and that’s more common than outright fakery.
And if a familiar person appears to be asking you for money, end the call. Learning to spot AI voice scams helps here too. Contact them on a number you already have saved, and use a family code word for urgent requests.
A real emergency survives a phone call to a number you already have.
How to Spot AI-Generated Text Without Trusting a Detector
AI-written text can sound polished, repetitive, vague or oddly certain. So can plenty of human writing, which is the problem.
So don’t judge the style. Check the facts around it instead.
Who’s the author? When was it published? Are sources named? Do the quotations and links go anywhere real?
Then search the important claims yourself and check whether the cited sources actually support them, the same way you’d verify claims and fake news. Often they don’t, and that’s the real tell.
One warning about AI detectors. They get it wrong in both directions, and a score from one is not proof that a person did or didn’t write something.
Judge the evidence and where it came from, not the prose you’re reading. Plenty of humans write like a form letter.
Pause, Verify, and Report Harmful AI Fakes
Pressure to act right now is the signal. Everything else is detail.
If a photo, video or voice message is pushing you to send money, reveal a code, invest, or do anything immediately, stop there.
Contact the person or the organization through a number you already have, or an official site you found yourself.
Save the post address, the account name, the date and screenshots before you report it to the platform. Things get deleted.
Report fraud losses or attempts to the Federal Trade Commission, and internet-enabled crime to the FBI’s Internet Crime Complaint Center.
Send a correction privately to whoever forwarded it to you, but don’t repost the original in order to argue with it. That just spreads it further.
You’re allowed to take your time.
For the wider picture, our online safety tips for seniors guide covers scams, privacy and security together in one place.
Frequently Asked Questions
Can visual clues prove that an image is AI-generated?
No single visual clue proves it. Check the original source, the context, reverse-image results, independent coverage, and whether several technical details are wrong together. Odd hands and garbled text used to be reliable tells, and newer tools have fixed them, so where a picture came from matters far more than how it looks.
Are AI-content detectors reliable?
Not reliably. They produce false positives and false negatives, calling real writing fake and fake writing real. Use one as a single signal and never as proof. Verify the author, the sources, the context and the supporting evidence instead. A detector’s score is a starting point, not a verdict.
What should I do if a deepfake asks me for money?
Stop the conversation and contact the person on a number you already have saved, or another channel you trust. Don’t use any contact details the message gave you, and agree a family code word for urgent requests. Hanging up to check isn’t rude, and a real relative will understand why you did.
