← Back to blog

How to tell if a text was written by AI

2 min read

AI writing tools have become so good that many people can no longer tell the difference at a glance. Still, AI-generated text tends to leave patterns behind. Knowing them helps you read more critically — as long as you remember that patterns are hints, not proof.

Uniform sentence length and rhythm

Human writers speed up and slow down. We write a long, winding sentence full of side remarks, and then a short one. Like this. Language models, on the other hand, often produce sentences of very similar length and structure, paragraph after paragraph. The result reads smoothly but strangely flat.

Try reading a passage out loud. If every sentence has the same cadence and every paragraph is roughly the same size, that is one signal worth noting.

Generic phrasing and safe conclusions

AI text often leans on stock phrases: "in today's fast-paced world", "it is important to note", "plays a crucial role", "a wide range of". It tends to summarise both sides of an argument and land on a balanced, non-committal conclusion.

None of these phrases is wrong on its own — people use them too. But when a text is full of them and never takes a clear position, it is worth looking closer.

Missing personal detail

People write from experience. They mention a specific colleague, a mistake they made, the name of a street, a number they remember. AI models have no lived experience, so their examples are often hypothetical or vague: "a small business owner", "many students", "in some cases".

Ask yourself: could this text have been written by anyone, about anything similar? If there are no concrete, checkable details, that is another signal.

Why no detector is 100 percent accurate

AI detectors, including ours, estimate probabilities based on patterns in the text. They do not have access to how the text was actually produced. That means they can be wrong in both directions.

A careful human writer with a formal style can be flagged as AI. A text generated by AI and then edited by a person can look human. Short texts give the detector very little to work with. Non-native writers, who often use simpler and more predictable phrasing, are at particular risk of false positives.

Anyone who promises perfect accuracy is overselling. Treat any score as an estimate with uncertainty attached.

Use a detector as one signal among several

The most reliable approach is to combine evidence. Look at the text itself for the signs above. Compare it with other writing by the same person. Check whether facts and sources are real — AI models sometimes invent citations. If possible, ask the writer about their process or to explain a passage in their own words.

A detector score is useful as a starting point: it tells you where to look more carefully. A good detector also shows you which passages drove the score, so you can judge for yourself whether the reasoning holds up.

See what the signals look like in your text

Paste a short text into our free check — no sign-up needed — and get an AI-probability estimate in seconds.

Check your text free