How to tell if something was made by AI

AI detection is not a single test — it is a combination of things you notice about text, images, or video that suggest a machine made it rather than a person. No detector catches everything, and AI tools improve faster than detection methods do. The most reliable approach is learning what AI tends to do wrong: repetitive phrasing, images with warped hands or teeth, video that moves unnaturally, or text that sounds confident but vague.

For text, read it out loud or paste it into a detector tool — but treat the result as a clue, not proof. For images, zoom in on hands, faces, and text within the image. For video, watch for flickering, objects that phase through each other, or movement that does not follow physics. None of these alone proves AI made something, but several together usually do.

Key Takeaways

  • AI text often repeats sentence structures, uses filler phrases, and avoids specific names, dates, or numbers.
  • AI images frequently show distorted hands, misaligned teeth, impossible anatomy, or text that reads backwards or is misspelled.
  • Detector tools like GPTZero and Turnitin flag AI text with varying accuracy, and no tool catches all AI or avoids false positives.
  • Video made by AI often shows objects that flicker, backgrounds that warp, or movement that violates basic physics.
  • The most reliable method is combining multiple signs — one odd thing might be a mistake, but several together usually means AI.

What to look for in AI-written text

AI text has recognizable patterns because it is trained on patterns in human writing. It tends to repeat the same sentence structures, especially at the start of paragraphs. It uses filler phrases far more often than a person would. It avoids specific details — actual names, dates, dollar amounts, or step-by-step instructions — in favor of general statements that could apply to anything.

Read the text aloud. AI often sounds smooth but hollow — it flows without saying much. A person writing about something they know tends to get specific, contradict themselves, or use their own phrasing. AI smooths everything into a middle tone. If a paragraph could be moved to a different article and still make sense, it was probably written by AI.

Look for hedging language: "may," "could," "might," "often," "typically." AI uses these words constantly because it is trained to avoid making claims it cannot verify. A person writing from experience tends to be more direct.

What to look for in AI-generated images

Hands are the easiest tell. AI struggles with fingers — they merge together, bend at wrong angles, or appear in the wrong number. Zoom in on any close-up of a hand. If the fingers look even slightly off, it is probably AI. Teeth are another weak point: AI often renders them as a blurry smear or makes them misaligned in impossible ways.

Look at text within the image. AI frequently misspells words, reverses letters, or renders text that does not quite read as English. If there is a sign, label, or caption in the image, read it carefully. Faces can be harder to spot because AI has improved at faces, but look for asymmetry — one eye slightly larger, ears at different heights, or skin texture that shifts unnaturally across the face.

Check the background and edges. AI often blurs or warps the space around the main subject. Objects in the background may be incomplete or phase into each other. Reflections rarely make sense — a mirror might reflect something that is not in the scene, or not reflect something that is.

Using AI detection tools

Tools like GPTZero, Turnitin, and Copyleaks scan text and return a score indicating how likely it is that AI wrote it. They work by analyzing patterns in word choice and sentence structure. None of them are perfect. They sometimes flag human writing as AI, and they sometimes miss AI that is well-written or mixed with human text.

If you paste text into a detector and it returns a high AI score, that is worth noting — but it is not proof. If it returns a low score, that does not mean the text is definitely human. The tool is most useful as one piece of evidence, not the final word. Detectors also struggle with text that is partly human and partly AI, which is increasingly common.

Free detectors exist but often have limits on how much text you can check at once. Paid versions used by schools and publishers (like Turnitin) are more thorough but not available to the general public.

What to look for in AI-generated video

AI video often shows movement that does not follow physics. Objects may flicker or phase in and out of existence. Backgrounds sometimes warp or shift in ways that do not match the foreground movement. People's mouths may not sync with audio, or their movements may be jerky rather than smooth.

Watch for repeated patterns. If a person blinks or moves their hand the same way multiple times in a short clip, it might be AI reusing a learned motion. Real video shows variation — people do not repeat gestures identically. Also watch for objects that pass through each other or people whose proportions shift between frames.

Lighting is another clue. AI sometimes struggles with consistent shadows and reflections. A shadow might not match the light source, or it might disappear and reappear as the camera moves.

Why detection is not foolproof

AI tools improve constantly, and so do detection methods — but detection always lags behind. A new version of an AI model can fool detectors that worked on the previous version. Humans also get better at prompting AI to sound more natural, which makes detection harder. Someone who knows what detectors look for can write AI text that avoids those patterns, at least partially.

False positives are common. A human writer who uses simple sentences, repeats words, or writes in a formal tone might score high on an AI detector. A student writing an essay about a technical topic might sound repetitive because the topic itself is repetitive. Detectors cannot read intent — they only see patterns.

The best approach is combining multiple methods: read the text yourself, check for specific details, look at images closely, and use a detector tool as one input among several. No single method is reliable on its own.

Frequently Asked Questions

Can AI detection tools tell the difference between ChatGPT and other AI models?

Most detectors cannot reliably distinguish between different AI models. They can flag text as likely AI-generated, but not identify which tool made it. Some research tools are being developed for this, but they are not widely available to the public.

What if text is written by a human but sounds like AI?

This happens often, especially with formal or technical writing. A detector might flag it as AI even though a person wrote it. This is why reading the text yourself matters — if it contains specific examples, personal experience, or unusual phrasing, it is probably human even if a detector says otherwise.

Is there a way to check video for AI without special software?

Yes. Watch for the signs listed above: flickering, unnatural movement, objects phasing through each other, and lighting that does not make sense. Slow down the video if possible. Real video, even low-quality video, usually obeys physics. AI video often does not.

Can AI make images that pass as human-made?

Yes, and it is getting better. Simple images with few details are harder to spot. Images with hands, text, or complex backgrounds are easier to catch. As AI improves, detection becomes harder, which is why combining multiple checks — hands, teeth, text, background — works better than relying on one sign.

What should I do if I find AI-generated content?

That depends on the context. If it is on a website claiming to be human-written, report it to the site owner or moderator. If it is in an academic setting, report it to your instructor. If it is in a news article, contact the publication. If it is on social media, you can flag it as misleading. There is no universal rule — context matters.