How to do it
First you look, then you verify where it comes from. The second step matters more than the first.
- Look for physical inconsistencies, but treat them as a warning bell, not as proof. Look at hands and fingers (number, length, position), teeth, mismatched earrings, reflections in glasses and pupils, the background that warps at the edges. In a video, check whether the lips actually follow the words and whether the eyes blink naturally.
- Look for logical inconsistencies. Text on a sign or a T-shirt that becomes illegible up close, a shadow falling from the wrong side, jewelry or moles that appear and disappear from one frame to the next: AI takes care of the overall effect, not the consistency of the details.
- Verify the source, and this is the decisive step. Who first posted this image? An anonymous account created yesterday, or an outlet that puts its name on the line? Is the same news reported by independent sources you already know? If it exists only on an unknown account, be suspicious.
- Do a reverse image search. Save or copy the image and upload it to an image search engine: it tells you where else it appears and whether it's a recycled real photo, an old photo passed off as new, or a fake already debunked by others.
Check: if the same image appears across multiple reliable and independent sources, with a consistent date and context, it's almost certainly authentic. If you find it only where you saw it, or paired with a story that pushes you toward instant outrage, stop.
A concrete example
Carla receives in the family group chat a video of a politician saying something scandalous. It looks real: right voice, right face. Before getting outraged and forwarding it, she does two things. She watches the mouth: in a couple of spots the lip-sync is slightly off compared to the audio. Then she searches Google for the politician's name and the phrase: not a single outlet mentions it, only the same video circulating on anonymous accounts. That's enough. She forwards nothing and writes in the group that the video doesn't come from any reliable source. She has stopped the chain.
When it does NOT work (and how to fix it)
If the image passes every visual check
The most recent generators produce faces and scenes that are indistinguishable from the real thing to the naked eye: 2026 studies indicate that very few people can reliably tell generated content from real content. When the visual check isn't enough, shift everything onto provenance: who says it, where it comes from, who else confirms it. The right question isn't "does it look real?" but "where does it come from?".
If you need a more solid verification
For the cases that matter (a piece of news that could have consequences, a photo used in a dispute) there are dedicated detection tools that analyze the image frame by frame looking for the traces of generation. They aren't infallible and they change often, but they give you one more technical opinion. Search for them as "deepfake detector" or "AI image detector" and use them as a second opinion, not as the sole verdict.
If the content aims straight at your emotions
When an image or video instantly triggers anger, fear or tenderness in you, that's exactly the moment to slow down. Fakes built to deceive are designed to make you react before you reason. The remedy is a personal rule: faced with content that shakes you, wait ten minutes and verify the source before believing it or sharing it.
A tip from someone who actually uses it
Stop hunting for the wrong detail and start asking yourself why you're seeing that particular content. Most dangerous fakes don't reach you by chance: they arrive with a story built to make you react. Train the habit of verifying the source before the content. It's the only defense that doesn't grow old as the generators improve.
Frequently asked questions
Is there an app that tells me for certain whether a photo is AI-made?
No, and be wary of anyone promising otherwise. Detection tools give a probability, not a certainty, and generators improve faster than detectors. Use them as an additional technical opinion, but verifying the source remains the main defense.
If I learn to recognize today's fakes, will I be safe tomorrow?
Unfortunately no, and it's the point many don't want to hear. The visual clues that work today will stop working with the next generation of tools. The only skill that stays valid is verifying where a piece of content comes from and who confirms it: that doesn't depend on the quality of the fake.