AI tools are part of research practice. Using them is not misconduct. Concealing their use, or letting them replace the intellectual contribution being claimed, is.
Fine
- Literature searching and summarising, with the sources verified.
- Drafting and editing language you then check and own.
- Code assistance you read, understand and test.
- Explaining a method to yourself.
Not fine
- Presenting AI-generated text as an expert's own analysis.
- Concealing AI use where the client, institution, funder or publisher requires disclosure.
- Citing references you have not verified. AI systems produce plausible references that do not exist. Submitting one is fabrication regardless of where it came from. This is the single most common way honest people commit misconduct.
- Pasting confidential, unpublished or participant data into a third-party AI tool without authorisation.
Disclosure
If the client, their institution, their funder or their target journal requires AI use to be disclosed, disclose it. Clients should state any such requirement in the brief; experts should ask if it is not stated.
Screening
Deliverables are screened for AI-generated content and the result is shown to both parties. It is a signal, not a verdict: legitimate writing flags, and generated text sometimes does not. Read it as one input among several.