AI Tells Detector

AI content detectors: what they can and cannot tell you

No tool can tell you for certain who wrote a text. This one shows the patterns people link to AI writing, so you can judge the page for yourself.

Try a sample:

By using this checker you agree to the terms.

Pasted text is checked in your browser and never sent anywhere. A web address is read once by our server and not stored.

A pattern count, not a probability. People write like this too, and AI can write without these. It cannot tell you who wrote the text.

What most detectors measure

Most AI content detectors return one percentage. Behind it is a statistical model that estimates how predictable the wording looks. Predictable wording is common in machine output, and it is also common in plain, careful, formulaic or second-language writing.

Why a score can mislead

In a 2023 study, Weixin Liang and colleagues ran several detectors on essays by native and non-native English writers. The detectors consistently misclassified the non-native writers’ essays as AI-generated while classifying the native writers’ essays correctly, and light rewording changed the results. The authors cautioned against using detectors to judge students. Read the paper, GPT detectors are biased against non-native English writers (published in Patterns).

Wikipedia’s editors, who clean up machine-written articles every day, give similar advice: do not rely only on detection tools, which have real error rates (Signs of AI writing).

What this checker does instead

It looks for 29 named patterns, marks each one in place and links to the full list of AI writing tells with a source for every entry. You see the count, the rate per 1,000 words and the exact words, and you decide what they mean. To clean up your own draft, the AI text checker and the guide to humanizing AI text are the faster route.

Never use it to accuse someone

Do not use this or any detector to grade a student, make a hiring or HR decision, or accuse a person of using AI. A pattern count cannot carry that weight. The terms say the same.

Can any tool prove a text was written by AI?

No. Detectors estimate how predictable or familiar a text looks. That is a guess about style, not a record of who typed it, and a 2023 study by Liang and colleagues found detectors misjudged essays by non-native English writers.

Is this tool an AI detector?

A pattern checker. It shows which known tells appear in the text, where they are and how to fix them, with no percentage and no guess about the author.

Why is there no percentage?

A percentage looks like a measurement of authorship, and none of these tells can measure that. A count and a list keep the result to what is actually on the page.

This site is not affiliated with any AI company or detector maker, or with the study’s authors.