What are the AI Assistant’s limitations, and how should I disclose its use?
Overview
The Assistant can misread context, propose a quotation that does not exactly match the source (flagged as an invalid evidence anchor), miss meaningful passages, or produce different suggestions on repeated runs. It works from the scope you provide and general reasoning, not from your unstated intentions. Treat every suggestion as a draft that a researcher must verify against the source before it becomes coding, interpretation, or reported text.
What to do
- Verify every AI-suggested quotation against the original source before accepting it.
- Expect some variation in wording or emphasis if you rerun the same AI task.
- Do not treat an AI summary as evidence of full corpus coverage; open the source for prominent, rare, and contradictory cases.
- Disclose AI assistance in your methodology in line with your institution’s or journal’s policy, describing which tasks used it and how outputs were verified.