How to detect if someone has used ChatGPT?

Question

Accepted Answer

Run the text through an AI detector first, and then compare the result against the evidence provided by the text. Detection scores tell how closely the text resembles machine-generated writing but cannot provide information about who wrote the text and when.
Signals that should be checked include: history of drafting and revisions in Google Docs, Word, or LMS, because copied AI text usually appears as one or two big pieces; uniformity in sentence length across the whole text; claims that sound confident but either contain vague citations or lack citations completely; going from the author’s voice to totally different ones; using generic examples in a situation when specific examples would be expected. Each of the signals proves nothing by itself.
The only reliable way to do the analysis is to combine high scores with two or three independent signs, followed by talking to the writer instead of making accusation.