Researchers: what to do when you are accused of using AI

A reviewer, editor or integrity office has questioned a manuscript.

What is at stake

Publication, funding, and in serious cases an institutional investigation.

Your evidence position

Unusually strong: raw data, analysis code with commit history, lab notebooks, pre-registrations, co-author correspondence, and earlier conference versions.

Do this first

Supply the data and code trail first. It answers a different and harder question than detector output does.

The trap

Treating a journal's detector screen as a finding. Most publishers describe it as a triage signal, not a determination.

Related pages that will help

Researcher disputes usually turn on the same few things: what a detector actually claims about its own accuracy, which pieces of evidence survive scrutiny, and whether a contract put the risk on you. The pages below cover each.