How to check if text is written by AI
To check if text is written by AI, look at the writing process rather than the prose. Key signals include a large share of text arriving via paste, an unnaturally uniform typing rhythm (consistent with transcribing AI output), and a word count inconsistent with the time spent. These behavioural signals are concrete, explainable, and don't degrade as AI writing quality improves.
Why text-only AI checkers are unreliable
Tools that score how 'AI-like' text reads rely on statistical patterns in the prose. Those patterns erode as AI models improve, and they disproportionately flag non-native or simple writing. Several universities have stopped using text-based AI checkers for exactly this reason.
What to look for in the writing process
If you can observe how text was entered, look for: a large block appearing in one or two paste events; very little variation in keystroke speed; a submission time implausibly short for the word count; and no evidence of editing or revision. Multiple signals together are meaningful.
How to collect the process evidence
Text that has already been submitted as a document has lost its writing history. Tools like Learnaway capture behavioural telemetry as students write, preserving the evidence you need. For text you already have, process-based checking isn't possible; you'll need to rely on a conversation with the student.
Setting this up for a class, not one document
The practical version of this for a teacher is not checking one paste after the fact, it is setting the assignment somewhere that captures the writing process from the start. See how the detection actually works and, if the process points to AI use, how to raise it with a student without it turning into an accusation.
FAQ
- Can AI-detection tools check any text for AI?
- Text-based AI checkers can process any text, but their reliability is limited and well-documented. Process-based tools like Learnaway require the writing to be done through the tool, but are substantially more reliable.
- What's the difference between checking text vs checking the process?
- Checking text analyses what was written; checking the process analyses how it was written. How it was written is harder to fake, doesn't degrade as AI improves, and doesn't unfairly flag non-native writers.
- Does this work for text a student already submitted?
- Not reliably. Process signals only exist if the writing happened inside a tool that was watching; a document handed in after the fact has no writing history left to check.
Want to see this on real writing?