How to use ChatGPT for research
Used well, ChatGPT is a genuine force multiplier for a researcher. Used naively, it’s a fast way to fabricate citations and embarrass yourself in front of a supervisor. The whole game is knowing what it’s good at (the shape of your work) versus what it’s dangerous at (the truth of it) — and verifying everything in the second column.
What it’s good at — and dangerous at
| Use it for (the shape) | Never trust it for (the truth) |
|---|---|
| Brainstorming angles & counter-arguments | Real citations — it fabricates them |
| Outlining and restructuring | Current facts, figures, statistics |
| Explaining a method in plain language | Judging the quality of evidence |
| Rephrasing your own writing | Writing claims you present as verified |
| Writing & debugging analysis code | Summarising papers it hasn’t actually read |
A safe stage-by-stage workflow
- Frame — pressure-test your question and brainstorm angles. (Ideas, not facts.)
- Understand — ask it to explain a method or concept in plain English, then verify against a real source.
- Find literature — do this in a database, not the chatbot. Paste real abstracts in for help summarising.
- Analyse — have it write or debug your analysis code; you run it and check the output.
- Write — draft in your own words, then use it to tighten and restructure — never to invent content.
The good prompts vs the bad prompts
Bad: “Give me five sources on X.” (invented references). Good: “Here are five abstracts I found — help me group them by theme.” The difference is whether you supply the facts and ChatGPT works on them, or you ask it to supply the facts. Master that distinction with the prompting guide.
Where it must never be used
Don’t paste confidential or unpublished data into a public model; don’t present AI-written text as your own where that’s prohibited; and disclose AI use wherever your journal or institution requires it. Permitted, transparent, verified use is a productivity tool — the trouble only starts when one of those three is missing.
Get the free AI-in-Research toolkit
A vetted prompt library, an AI-use disclosure template, and a verification checklist from The AI-Powered Scholar. We’ll email you the download link.
Frequently asked questions
Is it OK to use ChatGPT for research?
Yes — for the right tasks (brainstorming, outlining, explaining, rephrasing your own text, code) and with disclosure where required. Never as a source of facts or citations.
What's it good and bad at?
Good at the shape of your work (ideas, structure, code); dangerous at the truth (citations, current facts, judging evidence). Verify everything in the second column.
Can it find real sources?
Not reliably — a general chatbot invents references. Use a database or a source-grounded tool, and verify every reference exists.
Will it get me in trouble?
Only if you misuse it — undisclosed use where prohibited, fabricated citations, or passing AI text as your own. Transparent, verified use is fine.