35 ChatGPT Prompts for Research That Work

The value of artificial intelligence in research does not come from asking it questions — it comes from knowing which questions to ask, and what to forbid it from doing. The ChatGPT prompts for research below are built around that: each one specifies the output, the standard of evidence, and where the model must admit uncertainty rather than fill the gap.

There are 35 of them, covering scoping, literature review, methodology, analysis, market research and writing up. Paste any into ChatGPT, or run them in Chat Smith, a free AI chatbot that keeps your sources and threads in one place. For non-research tasks there is a wider collection of ChatGPT prompts as well.

Verify every citation. A model can produce a reference that is correctly formatted, plausibly titled and attributed to a real researcher in the right subfield — and entirely fictitious. Treat any specific paper, statistic or date as a lead to check in a database, never as a source. Check your institution's or publisher's policy on AI use too; many now require disclosure.

Require uncertainty flags. Instruct the model to mark what it cannot verify rather than smoothing over the gap. This single clause does more for research reliability than any other instruction.

Ask for the disagreement. Where sources conflict, what stays unresolved, what a sceptic would say. Consensus-shaped output hides the part of the evidence that is actually informative.

State the field and the standard. A synthesis for a supervisor, a client brief and a systematic review are three different documents. Say which one you need before you ask.

Work in one thread. A long-context model such as GPT-5.6 Sol can hold several papers and your own notes at once, which is what makes genuine cross-source work possible rather than approximate.

Six prompts for the stage that determines everything downstream. Most weak studies were weak at the question, and no amount of analysis rescues a project that was scoped badly.

1. The topic narrower

I want to research [broad topic] for [a thesis / a market report / a product decision]. It is too broad. Narrow it into five specific, researchable questions, each answerable within [timeframe] using [the resources you have]. For each, state what evidence would be needed and how hard it would be to obtain. Then tell me which is most feasible and which is most original.

2. The research question tester

Here is my research question: [paste it]. Pressure-test it: is it actually answerable with evidence, is it too broad or too narrow, does it contain an assumption I have not examined, and is it likely already answered? Describe what a good answer would look like, then rewrite the question in a sharper form.

3. The hypothesis builder

My research question is: [paste it]. Develop two or three testable hypotheses with their null hypotheses. For each, name the variables, how they would be operationalised, and what result would count as support versus refutation. Flag any hypothesis that is not falsifiable as currently worded.

4. The gap finder

Here is what I know about existing work on [topic]: [summarise it]. Help me identify where the gaps plausibly sit — unstudied populations, untested conditions, methodological limitations, or contradictory findings nobody has reconciled. For each, tell me how I would verify the gap is real rather than a gap in my own reading.

5. The feasibility check

Here is my proposed project: [describe the question, method and scale]. My constraints: [time, budget, access, skills]. Tell me honestly whether this is feasible, which component will take longer than I expect, what I should cut first if I fall behind, and what a scaled-down version that still produces a defensible finding would look like.

6. The prior art scan

I plan to research [question]. Before I invest time, tell me which fields have plausibly already addressed it — including adjacent disciplines that use different terminology for the same idea. List the search terms each field would use. Do not cite specific papers unless you are confident they exist; I will run the searches myself.

Six prompts for the reading pile, where most research time actually goes. Pair them with an AI PDF summarizer for the papers themselves and an AI article summarizer for the coverage around them — and note that prompt 11 explicitly forbids invented references.

7. The search strategy builder

Build a reproducible literature search strategy for [research question]. Give me concept blocks with synonyms, including older terminology and spelling variants; a Boolean string per block; which databases and grey literature to search and why; and explicit inclusion and exclusion criteria for date, study type, population and language. Flag the terms likely to return high noise.

8. The paper summariser

Here is a paper: [paste it or the key sections]. Summarise the research question, method, sample, main findings and stated limitations. Then tell me what it does not address, how much weight its methodology can actually bear, and which part of my own question it is genuinely relevant to.

9. The thematic synthesiser

Here are [number] sources on [topic]: [paste summaries or excerpts]. Organise them thematically rather than one by one. For each theme, state what the sources agree on, where they contradict each other, and what remains unresolved. Flag any claim that rests on a single study.

10. The contradiction resolver

These two sources disagree about [topic]: [paste both]. Determine whether they differ on the evidence itself, on the interpretation of the same evidence, on definitions, or on population and context. Then tell me what a careful reader could legitimately conclude from the disagreement, and what would be overreaching.

11. The citation integrity check

Here is a passage from my draft with citations: [paste it]. For each claim, tell me whether it needs a citation, whether the cited source plausibly supports that specific claim, and whether I am presenting a hypothesis as established fact. Do not invent references or claim to have verified anything you cannot actually see.

12. The note consolidator

Here are my notes from several sources on [topic]: [paste them]. Reorganise by theme rather than by source, keeping the source attached to every point. Merge duplicate points, flag contradictions rather than resolving them silently, and mark any point where my note is too vague to use as evidence.

Expand

Six prompts for decisions that are expensive to reverse once data collection starts. Use them to prepare for scrutiny rather than to replace it — methodology is where a supervisor or reviewer earns their place.

13. The design comparer

I want to investigate whether [variable] affects [outcome] in [population]. Constraints: [sample access, budget, timeframe, ethics]. Compare the three most appropriate designs with their trade-offs, recommend the strongest feasible one, and explain clearly what each design could and could not establish about causation.

14. The validity threat audit

Here is my planned study design: [describe it]. Identify the main threats to internal, external and construct validity. For each, tell me how likely it is in my specific design, how a reviewer would phrase the objection, and what mitigation is realistic given my constraints.

15. The sampling planner

My study needs a sample of [population] to test [hypothesis]. Recommend a sampling strategy, indicate roughly what sample size this design would need and what drives that number, and identify the biases my likely recruitment route would introduce. Be explicit about where you are offering a rule of thumb rather than a calculation I should run properly.

16. The instrument reviewer

Here is the survey or interview guide I plan to use: [paste it]. Review each item for leading language, double-barrelled questions, ambiguous wording, and response options that do not cover the range. Then tell me which items are measuring something different from what I intended them to measure.

17. The pilot planner

I am about to run [describe the study]. Design a small pilot: what to test, with how many participants, what specifically I am looking for, and which results would tell me to change the design rather than proceed. Name the single failure this pilot is most likely to catch.

18. The ethics preparation

Here is my proposed study: [describe the method, participants and data collected]. Identify the ethical considerations I need to address: consent, anonymity, data storage, potential harm, and any vulnerability in the population. Draft the questions an ethics committee would most likely ask me. Treat this as preparation only — my institution's approval process still applies.

Six prompts for turning results into claims you can defend. Language models are far better at framing an analysis than performing one, so run any arithmetic through an AI math calculator or proper statistical software rather than trusting a number that appears in a chat window.

19. The analysis planner

I have collected [describe your data: type, size, variables]. My hypothesis is [state it]. Recommend the appropriate analysis, explain why it fits this data and design, state its assumptions, and tell me what to check before running it. Flag any assumption my data is likely to violate.

20. The results interpreter

Here are my results: [paste the figures, including sample size, test used, effect size and confidence intervals]. Interpret them plainly: what can I claim, what can I not claim, and how strong is this really? Distinguish statistical significance from practical importance, and name the most likely alternative explanation.

21. The statistics sanity check

Here are statistical results reported in a paper I am reviewing: [paste them]. Assess whether the test suits the design, whether the effect size is meaningful or merely significant, whether confidence intervals are reported and how wide they are, and whether there are signs of multiple comparisons without correction or selectively reported outcomes.

22. The qualitative coder

Here are excerpts from my qualitative data: [paste them]. Propose an initial coding frame: the codes you see, what each captures, and where they overlap. Then apply it to three excerpts as a worked example. Flag anything you coded where a second coder would plausibly disagree.

23. The null result interpreter

My study did not find the effect I expected. Design: [describe it]. Result: [paste it]. Help me interpret this properly: is this evidence of absence or absence of evidence, was the study powered to detect the effect I was looking for, and what alternative explanations exist? Then tell me how to write it up honestly rather than apologetically.

24. The visualisation adviser

Here is the finding I need to communicate: [describe it and the underlying data]. Recommend the chart type that shows it most honestly, tell me what belongs on each axis, and warn me about the presentation choices — truncated axes, selective ranges, misleading scales — that would make the effect look larger than it is.

Expand

Six prompts for commercial research, where the standard of evidence is lower than in academia but the cost of a wrong conclusion arrives faster. Each asks the model to separate what it was given from what it is inferring.

25. The competitive landscape mapper

Map the competitive landscape for [product or category] in [market]. Here is what I have gathered: [paste it]. Identify the main positions, where competitors genuinely differ versus where they only market differently, and the position nobody currently occupies. Mark clearly where you are inferring rather than working from what I supplied.

26. The customer interview guide

I am interviewing [describe the customer type] about [topic]. Write a discussion guide: an opening that gets them talking, questions that surface actual behaviour rather than stated opinion, and probes for when answers stay abstract. Avoid anything that leads them toward my product. Include the one question most likely to disconfirm my assumption.

27. The survey designer

I need a survey to understand [research goal] from [target respondents]. Draft it: screening questions, main items with response scales, and one open-ended question actually worth analysing. Keep it under [number] questions. Then tell me which items would produce data I cannot act on, and cut them.

28. The feedback synthesiser

Here is customer feedback from [source]: [paste it]. Extract the recurring themes, distinguish product problems from expectation problems, and identify the single fix that would resolve the most complaints. Separate clearly what customers said from what you are inferring, and note the likely bias in who chose to leave feedback at all.

29. The trend evidence check

I keep seeing the claim that [state the trend] in my market. Here is the evidence I have found: [paste it]. Assess how well supported it actually is, who benefits from the claim being believed, what would falsify it, and what a sceptical analyst would say. Rate your confidence and state what you cannot verify.

30. The persona pressure test

Here is a customer persona my team uses: [paste it]. Test it against this evidence: [paste the data or research you have]. Tell me which attributes are grounded in evidence, which are assumption dressed as fact, and which ones would change our decisions if they turned out to be wrong.

Five prompts for the last stage. Keep citation placeholders rather than generated references throughout, and save the AI paraphrasing tool and AI grammar checker for the final pass, once the argument is settled.

31. The structure planner

I am writing up research on [topic] for [venue or audience]. My findings: [list them]. Propose a structure with what each section argues, roughly how long each should be, and the logical link between them. Flag any finding that has no natural home — that usually means an appendix or a separate paper.

32. The discussion drafter

Help me draft the discussion section. My findings: [list them]. What is already established in the literature: [summarise it]. Interpret each finding against that, using citation placeholders rather than invented references, explain the practical implications for [audience], acknowledge the three most significant limitations honestly, and propose two specific directions for future work.

33. The abstract compressor

Here is my paper or report: [paste it or a detailed summary]. Write an abstract under [word count] covering purpose, method, key findings with numbers, and what they mean. No adjectives doing work the numbers should do, and nothing in the abstract that does not appear in the paper itself.

34. The reviewer simulator

Act as a peer reviewer for [venue]. Here is my draft: [paste it]. Assess the clarity of the question, the appropriateness of the method, whether the conclusions actually follow from the results, and how honest the limitations section is. Give me the three objections most likely to come back, and the one that would be hardest for me to answer.

35. The reviewer response drafter

I received these peer review comments: [paste them]. Here is the relevant part of my manuscript: [paste it]. For each comment, tell me whether the reviewer has a point or has misread something, the minimum change that would satisfy it, and whether it is worth contesting. Then draft a response letter addressing every comment in order — conceding fairly, defending with reasons rather than assertion, neutral in tone.

Research is the use case where continuity matters most. Scoping, synthesis, methodology and write-up each want a different prompt, and the extraction ones get run on every new batch of sources. Chat Smith lets you save them as reusable templates, group them by project, and keep the whole thread in one place rather than starting a fresh chat each session.

It also runs several AI models side by side — ChatGPT, Claude, Gemini, DeepSeek and Grok. That has a specific use here: when two models summarise the same paper and disagree about what it found, one of them has misread it, and discovering that before you cite it is worth considerably more than a single confident answer.

Start with whichever prompt matches your current stage — the search strategy builder if a project is beginning, the citation integrity check if you are close to submitting. And keep the verification habit: the prompts here are designed to accelerate your judgement, not to substitute for it.