Page language: Italiano English Français Deutsch

How to spot audio, photos and videos made with artificial intelligence

By the FAIND newsroom · Updated on 17 October 2026

A security scanner on a conveyor belt checking paintings, photos and music discs

AI-generated content is increasingly hard to tell apart by eye. But there are signs to look for and, above all, checks that work even when the eye is not enough.

In brief

Visible flaws decrease with every new version: the eye is not enough.
The most effective check is on the source, not on the image.
Automatic detectors help but make mistakes: they are not proof.

How to spot a photo made with AI

Look at the details: hands and fingers, teeth, earrings that do not match, unreadable writing in the background, inconsistent shadows and reflections, skin that is too smooth, objects that merge into one another. These are useful clues, but recent models make fewer and fewer of these mistakes.

How to spot a video

Watch the sync between lips and voice, the blinking, the edges of the face when the person turns their head, flickering in the hair and background, and body movements that do not add up. In short, low-resolution videos it is harder.

How to spot AI-generated audio

Cloned voices tend to be uniform: little variation in tone, unnatural pauses, no breathing, no background noise. This is the territory of phone scams using the voice of a family member. The best defence is practical: hang up and call back the number you know, or agree a safe word within the family.

The check that works best: the source

Who published that content first? Are reliable news outlets reporting it? With a reverse image search you can see where and when a photo appeared. If a sensational piece of content exists only on an unknown profile, doubt is in order.

AI detectors and labels

There are tools that estimate whether content is artificial, but they give false positives and false negatives: they are a clue, not proof. More promising are labels applied at the source: invisible watermarks and “content credentials” that record how a file was created. In Europe the AI Act requires deepfakes to be flagged.

Before you share

If a piece of content makes you angry or surprises you a great deal, stop for a moment: that is exactly the effect its maker is after. For the wider risks, see is artificial intelligence dangerous?