A well-made fake account is almost impossible to spot on its own. A credible photo, well-written posts, varied interests: nothing looks off.
But a fake account rarely works alone. And a group, unlike an individual, always ends up repeating itself.
As the campaigns revealed by the New York Times show, networks of fake accounts are now run by AI. The principle for unmasking them remains the same, though: don't look at the person, look at the crowd.
1. Accounts born at the same time
When a discussion heats up around a topic, go and look at the creation date of the most active accounts. If many of them appeared in the same few weeks, and talk about nothing but that topic, that is a strong signal. A real crowd forms over years. A fake one is manufactured all at once.
2. The same ideas, on a loop
An AI can rephrase a sentence in a thousand ways. It has more trouble varying the arguments. If twenty different accounts make exactly the same points, in the same order, with the same links or the same keywords, be suspicious.
3. Impossible hours
Humans sleep, work, go away for the weekend. An account that posts day and night, that replies within seconds at four in the morning and never takes a break looks more like a machine than a person.
4. A closed circle
Look at who likes, shares and comments on the account's posts. If it is always the same accounts, also recently created, also obsessed with a single topic, you may have found the network. These accounts amplify each other to appear more numerous than they are.
5. The fake local
Campaigns often pass themselves off as ordinary residents of a town or a region. That is also their weak point. A real resident knows their street, their neighbourhood, its shops, the small local stories. A fake account stays vague, or makes mistakes no neighbour would make.
Faced with a post that irritates you, the reflex is to reply, or to share it in order to call it out. That is exactly what the campaign is hoping for. Social networks push posts that provoke a reaction, whether we approve of them or not. Every indignant reply therefore pushes the fake account towards more people. The most effective move is also the least satisfying: don't amplify it, and report it.
What not to do
None of these signs is enough on its own. Plenty of real people post late at night or talk about nothing but their passion. It is the combination of several signals that should alert you.
And even then, don't publicly accuse someone of being a fake account: you could be wrong about a real person. Report it to the platform, which holds information you don't, such as connection addresses or the devices used.
What to take away
Posts generated by AI are becoming impossible to tell apart from real ones. The behaviour of a network, however, remains visible: accounts that are born together, that talk alike and that talk only to each other.
For images, also bear in mind our five reflexes when faced with a dubious photo. And when in doubt, remember that the best way to weaken a fake crowd is not to lend it your voice.