Most tips about recognizing fakes come down to finding visual defects. Count your fingers, look at the shadows, check if the text floats on the signs. The problem with this approach is that it has an expiration date. Each new generation of generators removes the defects that were taught to look for last year. The six-finger tips are out of date in about a year and a half.

Therefore, this material is built around a different principle. Instead of analyzing the picture itself, it is more reliable to check its origin. This method does not depend on the quality of generation and will work the same in five years.

The main rule

A perfectly generated video should still appear somewhere for the first time. It has a place of publication, an author and a time of appearance. It's these things that are verifiable, while the pixels themselves become more and more convincing over time. Practically, this means three questions before believing or spreading.

Who posted it? Verified media or anonymous channel in social networks? See what other materials appear on this page, whether there are signs of systematic involvement. A page that regularly posts videos of the same type with loud emotional headlines is not trustworthy regardless of the specific material.

Who else writes about it? If the event actually happened, more than one anonymous channel will report it. The lack of mention in reliable media is a strong signal.

Where did it come from first? The most important question is often missed. Material that walks the network without the specified primary source almost always has reliability problems.

Reverse image search

The most useful tool in the entire topic, and at the same time the least known to the average user.

A reverse search shows where else on the Internet this image was found and when it first appeared. Works through Google Images, TinEye and other services.

The photo, which is presented as a photo of yesterday's event, may turn out to be a five-year-old shot from another country. This is the most common type of manipulation, much more frequent than generation from scratch, because it is cheaper and does not require any tools. For video, there is an InVID and WeVerify extension that breaks the video into key frames and allows you to search for each one separately. The check takes about a minute and closes most of the cases.

Emotion as a marker

Materials created for manipulation almost always cause a strong emotional reaction. Outrage, fear, panic, anger at a specific group of people. No one reposts a calm, balanced text, and the one that causes rage flies away in hours. Therefore, self-reaction is a diagnostic tool. If after reading you felt a strong impulse to share immediately, this is a reason to stop and check. The danger of fakes lies in the cumulative effect, when a large number of messages gradually form the desired narrative. A single message may seem like a trifle, and a hundred similar ones are already changing the perception of reality. An illustrative example of such work. Propaganda networks distributed a fake video allegedly about the mass mobilization of women and the ban on leaving, attributed to the deputy prime minister, in order to sow panic. No such decisions existed, and Ukrainian fact-checkers denied this.

What else works with the visual, with an honest caveat

What should be looked at in people's images: facial features look puppet-like, the skin is unnaturally smoothed, the look seems glassy and unfocused, facial expressions and especially lip movements do not coincide with what was said. In audio, the voice sounds artificial, in video there are no extraneous sounds and background noises, which are always in real recording. Common markers in images: disproportionate body parts, blurred movements, strange behavior of shadows and prints, unreadable text on signs and documents.

The caveat is that the absence of these signs proves nothing. High-quality generation does not have them. The defect found is proof of forgery, and its absence is not proof of authenticity.

There are also specialized services for analysis, such as Sensity AI for video and Resemble Detect for audio. They are useful, but they also do not give a 100% answer, so it is not worth perceiving their verdict as final.

Separately about voice fakes

The scheme looks like this: you get a phone call, and the voice sounds like the voice of a son, daughter or grandson. A person is in a panic, in trouble, urgently needs money. A few seconds of recording from social networks is enough to create such a voice. Agree with your relatives about a control question, the answer to which only you know. Or about a simple rule: hang up and call back to the known number yourself.

The second part of the same rule applies to money. Any urgent requirement to transfer funds must be verified through an independent channel, regardless of whose voice you hear. More about the mechanics of such pressure in the material on fraud recognition.

When confirmation is exactly a fake

The advice to check a person through a search has worked for years, but now has exceptions. Search engines index what is written on the Internet, including pages that a person has created about himself. AI summary at the top of the output synthesizes a ready-made statement from this material without a visible hierarchy of sources. A regular list of links allows you to see that the information comes from a personal account on the social network. The finished answer presents the same as the established fact.

I analyzed a real case of such mechanics in a material about a man who for years pretended to be a professional football player and used AI response screenshots as proof of his legend.

The conclusion is practical. Check the statement where it is officially recorded. A doctor in the register of medical workers, a company in the register of legal entities, a charitable foundation in the register of non-profit organizations.

A satire that is mistaken for news

A separate category that is not malicious intent but spreads just as quickly.

Satirical materials regularly lose the genre mark during retelling. Someone takes a funny post, retells it in their own words without pointing to the source, and in a few steps the joke turns into news. Signs to recognize: the story is too well made, it has a ready-made punchline ending, the hero does not exist in any independent source, and the details are physically impossible to verify. The check is the same as always: search by hero's name or event name in trusted sources. If there is nothing but the traditions of the same post, there was no case.

Where to check

Several specialized fact-checking projects are operating in Ukraine. Among the most famous are StopFake, VoxCheck, «On the other side of news» and «Liesless». The Institute of Mass Information maintains a white list of media, which can be used to navigate the reliability of a specific publication. To quickly check a specific statement, it is usually enough to enter it in the search together with the word «fake» or «refutation». If the material has already been disassembled by fact-checkers, it will be found in seconds.

If you have already spread the fake

This happens to everyone, and the main mistake is to leave everything as it is because of the inconvenience. Delete the post. Write a separate post with a rebuttal, because some people have already seen the initial material. If you shared it in a family chat, tell us directly.

Do not spread the fake further even for the purpose of indignation or exposure. Additional views enhance its effect regardless of your comment, and some of the audience will remember the content, not the objection.


Sources: Institute of Mass Information, StopFake, ICTV Facts, Osvitoria