Learnaway
Guide

How to tell if ChatGPT wrote an essay

The clearest way to tell if ChatGPT wrote an essay is to check how it was entered rather than how it reads: a single paste event covering 60-90% of the final word count, inter-keystroke gaps consistently below 150ms (a sign of transcribing rather than composing), and a 1,000-word essay finished in under 8 minutes are all objective, hard-to-fake indicators. Reading the prose for 'AI style' is far less reliable, since ChatGPT's phrasing changes with every model update and often flags non-native writers by mistake.

Why 'it reads like ChatGPT' isn't good evidence

Text-based AI checkers try to spot ChatGPT by scoring how statistically predictable the wording is. That signal erodes every time OpenAI ships a new model, and it regularly misfires on formal, simple, or non-native English, flagging human students as machine-generated. Several universities have restricted these tools in disciplinary proceedings specifically because of high false-positive rates.

The process signals ChatGPT essays actually leave behind

A student who generates an essay in ChatGPT and submits it typically pastes it in one or two actions, so 60-90% of the final word count arrives via paste rather than keystrokes. If any typing happens around the paste, it's often unusually uniform - inter-keystroke gaps under 150ms across a sustained run suggest transcription, not original thought. And the total session time is frequently implausible for the length: an 1,000-word essay wrapped up in under 8 minutes is a meaningful outlier for genuine drafting.

What to do when you spot these signals

No single signal proves ChatGPT was used - a student might paste from their own earlier notes, or draft in another app first. Treat two or more corroborating signals as a prompt for a conversation, not a verdict.

FAQ

Can a text-based AI detector reliably catch ChatGPT essays?
Not reliably. Text-based detectors estimate probability from prose style, and that accuracy degrades every time ChatGPT's underlying model changes. They also disproportionately flag non-native and simple writing as AI-generated, which is why several universities have limited their use in disciplinary decisions.
What's the single strongest sign an essay came from ChatGPT?
A large share of the final text - typically 60-90% - arriving in one or two paste events, combined with a total writing session far shorter than the word count would suggest for genuine drafting.
Can a student disguise a ChatGPT essay by retyping it?
Manually retyping suppresses the paste signal, but it often leaves unusually uniform keystroke timing behind - consistent gaps below 150ms are more typical of transcribing existing text than composing original thought.

Want to see this on real writing?