How can you tell if content is AI-generated?

You usually cannot tell for certain, but you can build a good picture. Look for tell-tale flaws (odd hands, garbled text in images, empty but fluent writing, voices that sound too smooth), then check the source: who posted it first, and whether a trustworthy outlet carries the same thing. No single clue or tool is proof.

In this guide, you will learn the clues to look for in text, images, video and voice, and social media posts, a quick routine for checking a suspicious post, and why the source matters more than any clue. The technology is improving quickly, so surface clues fade over time, while the habit of checking where something came from keeps working.

Why does it matter whether something is AI-made?

Most AI-made content is perfectly innocent: a company’s marketing image, a summary drafted by a helper, a stock photo on a blog. The kind of AI that creates new text, images, and video from a request is called generative AI, and it is now part of everyday media.

The reason to pay attention is the small slice that is meant to mislead. A convincing fake image, audio clip, or video designed to look real is called a deepfake, and one can be used to put false words in a real person’s mouth or invent an event that never happened. Being able to pause and question what you see protects you and the people you share things with.

How can you tell if text is AI-generated?

AI writing is often clean and confident, which is part of what makes it hard to spot. Still, a few patterns turn up often:

• Fluent writing that says a lot without landing on anything specific
• Facts that sound authoritative but are vague on names, dates, or sources
• Details that turn out to be invented, stated with full confidence (AI making things up like this is called hallucination)
• Repetitive phrasing, tidy lists of three, and an oddly even, tireless tone
• No personal experience, opinion, or detail that only a real person would know

None of these prove anything by themselves. Plenty of humans write this way, and plenty of AI writing is excellent. Treat them as gentle prompts to look closer, not as a verdict.

How can you tell if an image is AI-generated?

Images are getting harder to judge, but details still trip AI up. Zoom in and check:

• Hands and fingers, teeth, ears, and jewellery
• Writing on signs, labels, and clothing, which is often garbled or nonsense
• Reflections, shadows, and light that do not match each other
• Backgrounds that bend, blur oddly, or repeat
• Skin and surfaces that look too smooth or glossy, like a painting of a photo

The newest image tools get many of these right, so a clean image is not proof that it is real. Treat a flaw as a reason to check further, and the absence of flaws as telling you very little.

How can you tell if a video or voice is AI-generated?

Video and audio fakes are the ones most often used in scams, so they are worth a careful look or listen:

• Mouth movements slightly out of step with the words
• Blinking, head turns, or hair that look unnatural or glitch at the edges
• Speech that is flat, oddly smooth, or missing natural pauses and breaths
• A voice that sounds right but uses words the person would never use
• A very short clip, often low quality, where flaws are easier to hide

If a well-known person appears to say something surprising and no reputable outlet is reporting it, that mismatch is itself the biggest clue. And if a familiar voice on the phone asks for money in a hurry, hang up and call them back on a number you already have.

How can you tell if a social media post is AI-generated?

On social media, the post around the picture often tells you more than the picture itself. Watch for:

• A new account, or one with little history, a stock-looking profile photo, or a sudden change of topic
• A dramatic image or story with no named place, date, or original source
• Posts designed to stir a strong feeling, such as outrage, pity, or amazement, with a request to share
• Replies that all sound alike, which can be automated accounts
• The same image appearing elsewhere with a different story attached

A post that ticks several of these boxes is not necessarily AI-made, but it has not earned your trust yet either.

Can a tool tell you for sure?

No, and it is worth knowing why. Tools that claim to detect AI writing guess from style, and they get it wrong often enough that you should never treat a score as proof. Our guide on whether AI can detect AI writing explains the problem in detail.

For images, video, and audio, there are two more promising ideas. Some AI makers hide an invisible pattern, called a watermark, in what their tools create, and offer a way to check for it. Google, for example, adds one called SynthID, and its Gemini app can check whether an image was made or edited with Google’s own AI. Separately, an industry scheme called Content Credentials attaches a digital label to a file recording how it was made and edited, which some cameras, editing apps, and AI tools support.

Both help when they are present, but neither is universal. A watermark check only recognises that company’s own tools, many AI tools add nothing at all, and labels are often stripped away when a file is uploaded or reshared on social media. So a missing label tells you nothing, which is why the checks below still matter.

A 60-second check for a suspicious post

When something stops you in your tracks, run through these steps before you believe it or share it:

1. Check who posted it. Look at the account: how old it is, what else it posts, and whether it is who it claims to be.
2. Do a reverse image search. Upload the picture, or a screenshot of a video frame, to a reverse image search such as Google Lens or TinEye to see where else it appears.
3. Look for the original. Find the earliest version you can. A photo credited to a real photographer or news agency is a good sign; one that first appeared on an anonymous account is not.
4. Check a second source. Search for the claim itself. If something big really happened, reputable outlets will be reporting it too.

If you cannot confirm it within a minute or so, the safest choice is not to share it yet.

Why is the source the best clue?

Whatever the medium, the strongest test is the same one journalists use. Ask where the content came from, whether a source you already trust is carrying it, and whether the account or website sharing it is genuine.

This matters because AI is increasingly trained and tested on material that was itself produced by other AI, known as synthetic data, and the polish keeps improving. As the surface clues fade, the trail back to a trustworthy origin becomes the thing you can actually rely on.

Try it yourself

You can use AI to help you think an image through, as long as you treat its opinion as one input rather than a ruling. Share the image with a chatbot and ask it to walk you through what to check:

I've come across the image below and I'm not sure whether it's real.
Please list the specific details I should check that might suggest it
was made or edited with AI, point out anything that looks convincing,
and suggest how I could find the original source. Be honest that you
can't tell for certain whether an image is AI-generated.

Then do the human part: run the 60-second check above and see whether a source you trust backs it up.

A calmer way to look

A common instinct is to feel you now have to doubt everything you see, which is exhausting and unnecessary. A better habit is a small pause before you believe or share something that stirs a strong reaction. Most content is exactly what it appears to be. Reserving your scrutiny for the surprising and the emotional is both easier and more effective.

Next steps

AI-made voices and videos are a favourite tool of scammers, so our guide on how to spot AI scams is a natural next read. If you would like to understand the bigger picture of how AI is built and the ethical questions around it, our conceptual guide on AI safety and ethics is a thoughtful place to go next.