How to check if something was written by AI
To check if something was written by AI, examine the writing process rather than the finished text. The clearest behavioural indicators are text arriving via paste in one or two events covering 60–90% of the final word count, inter-keystroke gaps below 150ms across a sustained typing run (consistent with transcribing AI output rather than composing), and a total writing session under 8 minutes for a 1,000-word piece. Two or more of these process signals together are more reliable and fairer than any text-based AI style analyser.
Why text-based AI checkers are unreliable
Tools that read prose to estimate AI origin work by scoring how statistically predictable each word is. As AI writing quality improves those patterns erode, and they systematically flag simple or non-native writing — regardless of who actually wrote it. Several universities have restricted their use in formal disciplinary processes because false-positive rates were unsustainably high. A text checker can't tell you how a piece was written; it can only guess about the style.
The three behavioural signals that actually reveal AI use
If you can observe how text was entered, look for three concrete signals. First, paste dominance: a large block of text arriving in one or two paste events that accounts for 60–90% of the final word count. Second, uniform keystroke timing: inter-keystroke gaps consistently below 150ms suggest the writer is transcribing existing text rather than composing original thought. Third, implausible speed: a 1,000-word piece submitted in under 8 minutes is a meaningful outlier for most writers. Any one signal alone is weak; two or more together are worth a closer look.
What to do when you only have the finished text
If the piece has already been submitted as a document, the writing-process evidence is gone. You can run it through a text-based AI checker, but treat the output as weak circumstantial evidence rather than fact. The more productive path is to ask the author to walk you through their process — a person who wrote the piece genuinely can describe specific decisions; someone who generated and pasted it typically cannot. For future submissions, use a tool like Learnaway that captures the writing timeline as the piece is composed.
FAQ
- Is there a free way to check if something was written by AI?
- Free text-based AI checkers are widely available, but their reliability is limited and well-documented — false positive rates are high enough that several universities no longer rely on them in disciplinary proceedings. The most reliable check is process-based: how was the piece written? Learnaway offers a free tier for teachers who want to capture that process evidence for future submissions.
- Can someone edit AI-written text to fool a checker?
- Yes — text-based checkers are relatively easy to fool by paraphrasing or light editing. Process-based signals are harder to fake: a student who manually re-types AI output can suppress paste signals, but unusually uniform inter-keystroke timing (gaps consistently below 150ms) can still hint at transcription rather than original composition.
- What's the difference between checking AI writing and checking for plagiarism?
- Plagiarism checkers compare submitted text against existing sources to find copied passages. AI writing has no source to match — a ChatGPT essay can score 0% similarity on a plagiarism checker while being entirely machine-generated. That's why behavioural process signals, not source-matching, are the appropriate tool for detecting AI-assisted work.
Want to see this on real writing?