Is there a reliable way to spot a deepfake by eye?

No, and relying on a visual clue is risky both ways: you can think you've spotted a fake in a genuine video, and the reverse. Provenance is a sturdier criterion than the image, because it doesn't depend on how good the fabrication is.

A well-known figure recommends an investment in a video. What should I do?

Do none of what it asks. This is a recurring use of the technique, and the decisive point isn't the video but the fact that no legitimate investment is ever offered this way. Check the financial actor against the regulator's register before anything else.

Deepfake video

In plain terms

A video where someone says or does something they never said or did. The face and voice are a real person's; the scene is not.

Definition

A deepfake video is a sequence fabricated or altered using machine learning, in which a real person's appearance and voice are reproduced to make them appear to say or do things that never happened.

How it works

Making one takes material: images and audio of the target, supplied by public videos, interviews, social media. Public figures are therefore the easiest targets, and their fame is exactly what gives the fake its value — an investment tip attributed to a well-known figure, a political statement, a business announcement. What changes for the viewer isn't that doubt becomes impossible, but that the usual clues disappear: the face matches, the voice matches, the setting is plausible. Doubt has to shift from the image itself to its provenance — who is posting it, from what account, and does this statement exist anywhere else.

Warning signs

An important statement circulating without being picked up by any identifiable source
Video posted by a recent account, or cropped to hide its origin
Inconsistencies at the seams: face outline, hair, teeth, reflections in the eyes
Imperfect lip-sync, especially on consonants
Short clip, cut before any large head movement happens
Content that serves an obvious interest: a financial pitch, discrediting someone, a call for donations

How to verify

Look for the same statement elsewhere rather than examining the image: a genuine public statement from a well-known person leaves multiple, consistent traces. Trace the video back to its originating account and check its creation date. Visual analysis comes last, and isn't conclusive on its own.

What to do

Don't share before finding a second, independent source — sharing is what gives the fake its reach. Treat any video urging a financial investment or a payment as suspicious by default, no matter who appears to be speaking.

If it already happened

If you are the person depicted: keep the links and time-stamped screenshots, report the content to each platform as impersonation, and file a police report. Publicly alert your professional contacts — that limits the reach faster than any takedown. If you sent money after seeing such a video, contact your bank immediately.

Frequently asked questions

Is there a reliable way to spot a deepfake by eye?: No, and relying on a visual clue is risky both ways: you can think you've spotted a fake in a genuine video, and the reverse. Provenance is a sturdier criterion than the image, because it doesn't depend on how good the fabrication is.
A well-known figure recommends an investment in a video. What should I do?: Do none of what it asks. This is a recurring use of the technique, and the decisive point isn't the video but the fact that no legitimate investment is ever offered this way. Check the financial actor against the regulator's register before anything else.

Official sources

Cybermalveillance.gouv.fr — Assistance and reporting (French government cybersecurity assistance service)

Also known as: synthetic video, video deepfake, AI-generated impersonation video.

This article is part of the AI and synthetic content family. Last updated: 2026-09-03.