Using ChatGPT Responsibly in Research Writing

Define the research work that remains yours
Using ChatGPT for research writing can help with organisation, clarity and questions to investigate. It does not transfer responsibility for the research question, ethical conduct, analysis or final claims. Begin with your institution’s rules and the target journal’s current author instructions. Requirements for disclosure, confidential data and acceptable assistance differ.

Separate three materials: your research records, the sources you have actually read, and AI-generated suggestions. A sentence that moves from the third category into the manuscript needs the same scrutiny as any other unsupported claim. Do not turn a plausible reference into a citation until you have located and checked the real publication.

Build an evidence map before drafting
Create a table with columns for the claim, supporting source, relevant passage, study context and limitation. “Training improved performance” is too broad if the source measured only participants’ satisfaction. Record the actual outcome and population. This table makes it harder for fluent writing to conceal a weak connection between evidence and conclusion.

Ask the assistant to identify gaps in a supplied evidence map, rather than asking it to invent a complete literature review. A useful instruction is: “Suggest questions this evidence does not answer. Do not add references or findings. Distinguish limitations of the studies from limitations of my proposed interpretation.” Verify the suggestions yourself.

Draft from verified notes, not from a topic alone
For an introduction, provide the problem, definitions and checked evidence. Request an outline that shows the purpose of each paragraph. For a methods section, work from the procedure actually followed; do not ask an assistant to improve the study by silently adding recruitment steps, instruments or approvals that did not exist.

For results, provide outputs whose calculations you understand. Ask for a description of the supplied numbers without causal claims or additional analyses. Compare every reported value with the analysis record. For discussion, distinguish what the findings support, what remains uncertain and what future work might investigate.

Example: a small training evaluation
Suppose a constructed teaching example contains pre- and post-training scores from the same learners. An assistant may suggest that the training “caused a dramatic improvement.” That statement goes beyond a simple before-and-after comparison without a suitable design addressing alternative explanations.

A better writing task is to explain the observed change, identify incomplete pairs, describe the assessment and state the design’s limits. You still need to select an appropriate analysis, inspect assumptions and report uncertainty. The writing assistant should not choose a favourable method merely because a desired conclusion has already been written.

Protect participants and confidential material
Do not upload participant identifiers, unpublished collaborator work or restricted records without permission and an appropriate data-handling arrangement. Removing names may leave identities visible through combinations of school, role, location or a distinctive quotation. Decide what information the task needs and share only an authorised version.

Keep records of substantive AI assistance so that disclosure can be accurate. Preserve drafts and analysis scripts where institutional policy requires them. Never claim that an automated process supplied human specialist review, verified originality against all published literature or established authorship eligibility.

Review the manuscript as a connected argument
Read the final version without the assistant open. Does each conclusion follow from the actual results? Do citations support the sentences where they appear? Are quotations exact and permitted? Are uncertainty and contradictory findings visible? Have co-authors reviewed their responsibilities?

Use AI assistance to expose a weakness, not to hide it under more elegant language. A shorter, accurate statement of what your study established is more valuable than an impressive paragraph that the research cannot support.

Editorial drafting used AI assistance. Examples labelled fictional or constructed are teaching examples, not client results. The original author record is retained; no independent specialist review is claimed.