RESEARCH / BLOG

Research,
Explained.

Read the research question, follow the authors’ methods and explore the evidence behind the article.

Research Blog

Research methodology · Research insights

Prompt Like a Scientist: Seven Parts of a Useful AI Research Question

Write better AI research prompts with context, evidence, limits and verification. A practical seven-part guide for literature reviews, hypotheses and research methodology.

By NPT Research5 min read
The novelty toolkit06 / 09
Ask a question you can check.Context, evidence and a clear verification step.
NPT Research

The Novelty Toolkit · Series 1 · Part 6 of 9

Good AI research prompts make the next step easier to verify. This tutorial shows how to give an assistant a clear scientific brief for literature review, research methodology and hypothesis development, with practical prompt examples you can adapt to your own field.

“Suggest a new research idea.” This sentence feels efficient. It also leaves almost every important decision to the assistant: field, problem, evidence, constraints, meaning of novelty and level of difficulty. The answer may sound elegant while being almost impossible to evaluate.

Prompting for research is closer to writing a good laboratory instruction than searching for a clever phrase. Give enough context, define the task, require an inspectable output and specify how uncertainty should be handled.

The seven parts of a research-grade prompt

The novelty toolkitGive the assistant a brief it can follow.Seven elements make the request inspectable. They do not guarantee a correct answer.
  1. 01ProblemWhich observation or decision needs explaining?
  2. 02ContextWhich field, conditions and audience apply?
  3. 03EvidenceWhich measurements and papers can you supply?
  4. 04TaskExplain, compare, critique or propose hypotheses?
  5. 05LimitsWhat must remain uncertain or confidential?
  6. 06OutputWhich format will help you evaluate the answer?
  7. 07CheckWhich source, calculation or test will verify it?
Figure 1. Seven elements that make an AI research request useful and auditable.
ElementPractical question
ProblemWhat unresolved observation or decision is this about?
ContextWhat field, conditions, audience and background apply?
EvidenceWhat reliable observations and source papers do you already have?
TaskDo you want explanation, comparison, critique or hypothesis generation?
LimitsWhat must remain uncertain, excluded or confidential?
OutputTable, comparison, search strings, decision tree or experimental plan?
CheckWhat citations, calculations or tests would verify the answer?

Before and after: the same research idea

Unhelpful prompt: “Tell me the most novel use of AI in a water-quality sensor.” It asks for a superlative no search can establish and provides no local evidence.

Example prompt

I am investigating long-term drift in low-cost water-quality sensors used in remote ponds. I observed that readings gradually diverge from laboratory reference measurements after weeks of deployment. Help me identify possible physical and environmental causes. Separate established mechanisms from plausible hypotheses. For each mechanism, provide search keywords, the experimental observation that would support it, and a competing explanation. Do not claim a method is novel. If you cannot access live papers, tell me explicitly.

The novelty toolkitRefine the request before trusting the answer.Replace a demand for novelty with a question that can be investigated.
  1. 01
    Start with a topicA broad subject gives too little context to assess an answer.
  2. 02
    Supply a specific observationDescribe the evidence, constraints and competing explanations.
  3. 03
    Ask for a checkable outputRequest sources, uncertainty and a discriminating test.
This is a prompt-design example, not a measured performance comparison.
Figure 2. A specific, verifiable task gives the researcher more to check than a vague request for novelty.

Request evidence, not confidence

Instead of asking “Are you sure?”, ask “Which exact sentence or result in the cited paper supports this claim?” A more confident tone is not additional evidence. It cannot replace the full paper or the measurement.

Example prompt

For each factual claim, provide: (1) claim; (2) source title and stable link or DOI if verified; (3) what the source actually establishes; (4) uncertainty or boundary condition; and (5) what I need to check in the full text. If no source is available, label the item UNSOURCED rather than guessing.

Ask for alternatives before choosing a direction

Premature convergence is dangerous. A single convincing answer can anchor the entire project. In an early brainstorming stage, ask for competing hypotheses and substantially different methods—not just paraphrases of the first idea.

Example prompt

Generate four scientifically distinct explanations for [observation]. One should be a measurement artifact, one a physical mechanism, one a confounding environmental factor, and one a genuine model limitation. For each, propose a discriminating experiment and one reason it might be wrong.

Use the assistant to ask you questions

If you do not yet know what information is missing, reverse the conversation. Ask AI to interview you like a cautious methodology supervisor, one or two questions at a time.

Example prompt

Before proposing a research direction, ask me the most important missing questions about observations, target population or system, available data, equipment, ethical constraints, comparison method, and outcome. Avoid solutions until the problem has been sufficiently defined.

Build a transparent output

A table with columns for claim, evidence, uncertainty and action is usually more useful than a persuasive paragraph. It lets you verify the source, reject a weak claim and maintain a research record.

AI outputHow the researcher uses it
ExplanationChecks whether the concept is understood
Search stringRuns the query in a real literature database
Candidate referenceOpens the original publication
Proposed experimentReviews feasibility, controls and ethics
Novelty suggestionCompares with closest prior work

Iterate, rather than searching for a magic prompt

A useful prompt may develop through several revisions of a modest first attempt. Review what AI misunderstood, supply missing details, correct assumptions and repeat. Official prompt guidance recommends clear requests, context and iteration; it does not promise that a perfect prompt can guarantee a correct answer.

  1. Version 1: explain the problem in plain language.
  2. Version 2: add actual observations, constraints and terminology.
  3. Version 3: request alternatives, counterexamples and disconfirming evidence.
  4. Version 4: request traceable sources and a comparison table.
  5. Version 5: check the sources and revise the question yourself.

The best research prompt is not the one that makes AI sound brilliant. It is the one that helps you understand the problem more accurately and identify what evidence is still missing.

What the evidence does—and does not—show

OpenAI’s ChatGPT prompting guidance and official prompting documentation recommend clarity, context and iteration. They do not validate this exact seven-part template or promise factual correctness. The Nature Perspective on illusions of understanding explains why a plausible explanation still deserves scrutiny.

Shahid and colleagues’ literature-grounded novelty assessment work studies assessment with research evidence. It supports the importance of inspecting the literature in that setting; it does not make a general chatbot response a verified review. For a practical reading sequence, use the closest-prior-research search guide and record each verified comparison in the novelty matrix.

The examples and prompt frameworks are educational illustrations. Verify sources, use domain expertise and follow research ethics. Do not upload confidential or patentable material to an AI service without authorisation.

Continue the research novelty series

Series 1: originality, evidence and scientific judgment

Begin with what makes research novel, then find the closest prior research and build a novelty matrix. Before using an assistant, examine the limits of AI novelty judgments and why the researcher owns the question. Develop the conversation with scientific prompting and six-round brainstorming, then challenge the idea against evidence. The complete worked research dialogue connects those decisions.

Series 2: from a problem to a defensible proposal

First separate a research gap from a research problem and identify a genuine problem worth investigating. Next search published literature for the gap and align the question, objectives and hypothesis. For AI-assisted work, clarify the problem without outsourcing judgment, verify the literature search and its references and compare papers in an evidence matrix. Then stress-test the research hypothesis before following the complete observation-to-proposal workflow.

Sources and further reading

  1. OpenAI. Prompt engineering best practices for ChatGPT. Official guidance, accessed 22 November 2026.
  2. OpenAI. Prompting guide. Official documentation, accessed 22 November 2026.
  3. Messeri & Crockett. Artificial intelligence and illusions of understanding in scientific research. Nature (2024), Perspective. Published 6 Mar 2024
  4. Shahid et al. Literature-Grounded Novelty Assessment of Scientific Ideas. Scholarly Document Processing workshop (2025).