Is this photo, video, or recording real?

This is the question everything else in this section eventually comes down to. The honest answer is that no tool settles it — but a handful of ordinary checks will resolve most of what you'll actually run into.

Start with where it came from, not what it looks like

Almost everyone approaches this backwards — staring at the image for flaws. Staring is the least reliable method available, and it gets worse every year. Where something came from is far more informative than how it looks, and it doesn't degrade as the technology improves.

Find the earliest version you can. Search a distinctive phrase from the caption, or use a reverse image search — right-click an image in most browsers and look for the option to search it. If a genuinely dramatic event happened, more than one source will carry it. If everything traces back to a single anonymous account posted an hour ago, that's your answer.
Check whether anyone with something to lose has reported it. Not as an appeal to authority — simply that a wire agency or a large newsroom has staff, competitors, and a reason to avoid being wrong.
Ask who benefits. Material designed to make you furious, frightened, or certain is worth twice the scrutiny, because that reaction is usually the product.
Look for the mundane confirmations. Weather, shadows, signage, licence plates, the language on a shopfront. Fabrications are built to be watched, not investigated, and they rarely survive contact with dull details.

Why "look closely" stopped working

The visible flaws people learned to check for — hands with too many fingers, garbled text on signs, unnatural blinking — were artefacts of a particular generation of tools, not permanent features. Many are already gone. Anything you learn about how fakes look will expire; how to trace a source will not.

Content Credentials, and what they honestly prove

There's a genuine effort underway to attach a signed record to images and video at the moment they're created, showing what device or program made them and what was done to them since. You may see it called Content Credentials, or C2PA. A growing number of cameras, phones, and editing programs support it, and verification is beginning to appear in mainstream places rather than only in specialist tools.

It's a real improvement and worth knowing about. But be precise about what a credential establishes: that a file was signed by a particular device or piece of software and hasn't been altered since. It does not establish that the camera was pointed at what the picture appears to show. Those are very different claims.

Two practical limits matter more than the theory. The first is that missing credentials mean nothing at all — most cameras and phones don't sign anything, and credentials are routinely stripped when a file is uploaded, screenshotted, or recompressed, so their absence is the normal case rather than a red flag. The second is that a valid signature isn't proof the content is honest: one camera maker's signing feature was suspended in 2025 after a researcher showed a manipulated image could be fed to the signer and come out with a signature that passed every standard checking tool. The cryptography worked exactly as designed. It was simply signing something that wasn't true.

Detection tools

Websites offering to tell you whether an image or recording is AI-generated do exist, and they can be a useful data point. Treat their output as a weak opinion rather than a verdict. They produce confident wrong answers in both directions, and they perform worst on exactly the material you'd most want to check — heavily compressed clips, screenshots of screenshots, short audio. If a tool says "87% likely AI," the number is doing more work than the analysis behind it.

The check that beats every tool

For anything that involves you personally — a message, a call, a video of someone you know — don't analyse the artefact at all. Contact the person through a channel you already had and ask. This resolves the question completely and instantly, while every technical method only ever produces a probability.

Living with the uncertainty

Two failure modes are worth avoiding. Believing everything is the obvious one. Believing nothing is the subtler one, and it's arguably more corrosive: once someone decides that anything could be fake, real evidence stops counting too, and that turns out to be very convenient for people caught doing real things.

The middle position is a habit rather than a skill. Ask where something came from before deciding how you feel about it. Confirm anything that touches you personally through a channel the other side didn't choose. That covers nearly everything, and it doesn't require you to become an expert in anything.