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

Brainstorm With AI Without Losing Your Scientific Thinking

Use a six-round AI brainstorming method to frame observations, challenge assumptions and develop testable hypotheses while keeping research decisions in your hands.

By NPT Research5 min read
The novelty toolkit07 / 09
Explore. Challenge. Refine.Six rounds that keep the researcher in control.
NPT Research

The Novelty Toolkit · Series 1 · Part 7 of 9

Research brainstorming should clarify uncertainty. This six-round tutorial helps PhD scholars and researchers develop a testable hypothesis through observation, criticism and literature comparison. The method is an educational framework, not a validated guarantee of better research ideas.

Imagine using AI like a whiteboard that talks back. A whiteboard does not decide which experiment is worth doing; it helps you put competing thoughts side by side. AI is more active than a whiteboard, but the danger is similar: a well-written explanation can seem like the answer before the question has been properly examined.

This tutorial provides a six-round dialogue. Each round has one task and one stop condition. The goal is not to finish with “the most innovative idea” but with a smaller, better research question.

Begin with what you actually know

The novelty toolkitOne task at a time. Six chances to reconsider.Stop each round when its question is answered, or record what remains missing.
  1. 01
    FrameSeparate your observation from the explanation you prefer.
  2. 02
    UncoverName the assumptions and identify a way to challenge them.
  3. 03
    DivergeConsider genuinely different mechanisms, including measurement error.
  4. 04
    ChallengeLook for counterexamples, confounders and missing controls.
  5. 05
    ConvergeSpecify a falsifiable hypothesis and a fair comparison.
  6. 06
    VerifyRead the closest prior work and document uncertainty.
Figure 1. Six rounds of researcher-led brainstorming, ending in verification.

Consider this fictional example: a low-cost water-quality sensor loses agreement with laboratory reference measurements after prolonged deployment. We do not yet know whether the cause is fouling, temperature, sensor ageing, improper handling or something else.

The novelty toolkitKeep three kinds of notes separate.A plausible explanation is not yet an observation.
  1. 01ObservedWhat was actually measured or witnessed?
  2. 02HypothesisedWhich mechanism might explain the observation?
  3. 03Still neededWhich source, control or measurement would distinguish explanations?
The pond-sensor example is fictional; these categories are research-note conventions.
Figure 2. Separate observations from suspicions and information needs.

Round 1: Frame the observation

Example prompt

Here are my observations: [facts]. Here are my assumptions: [assumptions]. Here are my constraints: [constraints]. Restate the scientific problem without recommending a solution. Identify information I have omitted that could change the interpretation.

Stop when the problem statement reflects what you have truly observed rather than what you wish to prove.

Round 2: Uncover hidden assumptions

Example prompt

Act as a skeptical research supervisor. List the assumptions behind my interpretation. Which could be false? For each assumption, propose one simple way to challenge it using observations or a control.

Suppose the readings drift. The hidden assumption may be that the sensor is deteriorating; alternatively, the reference sample may be handled inconsistently. Naming the assumption prevents the project from becoming a solution in search of a problem.

Round 3: Diverge into genuinely different explanations

Example prompt

Propose five non-overlapping explanations for this observation: one physical, one chemical or biological if relevant, one measurement-related, one contextual, and one based on an incorrect initial assumption. For each, show the observable pattern that would distinguish it from the others.

The objective is diversity, not a flood of synonyms. Research on particular brainstorming tasks has found that individually attractive AI suggestions can coexist with reduced diversity across the collected ideas. Asking for alternative mechanisms makes the comparison explicit; it does not guarantee that anchoring is prevented.

Round 4: Challenge the most promising explanation

Example prompt

My leading explanation is [X]. Make the strongest evidence-based case against it. Identify a plausible counterexample, missing control, confounder, and a previously published approach I should investigate. Do not flatter or agree by default.

A useful counterargument improves the study even when the original suspicion is correct.

Round 5: Converge on a falsifiable hypothesis

Example prompt

Convert the remaining explanation into a testable hypothesis. Specify independent variables, outcomes, comparison group or baseline, a possible falsifying result, major confounders, and the minimum information needed to plan sample size or test duration. Mark any design details that require domain expert judgment.

In the example, a narrow hypothesis could concern whether temperature adjustment improves agreement relative to a fixed calibration method during repeated field measurements. This is a possible experiment, not a declaration of novelty.

Round 6: Verify and document the open question

Example prompt

Search or help me search for the closest prior work to this precise hypothesis. Propose synonym-rich search strings for Google Scholar and appropriate subject databases. Build a comparison table with evidence, overlap, uncertainty and links. If a reference is unverified, mark it unverified. Give me three reasons the proposed claim may not be novel.

A short dialogue journal

After each round, recordExample entry
What changed?I stopped assuming the device itself is ageing
Which claim is still unknown?Temperature may account for part of the drift
Which source do I need?Field-validation studies of low-cost sensor calibration
What decision is mine?Select a feasible controlled comparison

Common Mistake

Do not ask AI to generate twenty “innovative” ideas and then choose the most impressive-sounding one. Choose using evidence, feasibility, meaningful questions and explicit prior-work comparison.

Brainstorming succeeds when the researcher ends the conversation understanding the uncertainty more clearly than when it began.

What the evidence does—and does not—show

The Nature Human Behaviour analysis by Meincke and colleagues reports reduced pooled idea diversity in the brainstorming tasks it examined. That is a reason to inspect alternatives, not proof that this six-round dialogue prevents anchoring. Si and colleagues’ ICLR study evaluates proposed research ideas; expert ratings of an idea should be distinguished from evidence after an experiment.

SciMON investigates literature-based inspiration and iterative comparison, while HARPA explicitly considers testability and literature grounding. These are particular research systems and evaluations. Our prompts borrow the questions worth asking—what is the closest work, and what can be tested—without claiming their experimental results for an ordinary chat.

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. Meincke, Nave & Terwiesch. ChatGPT decreases idea diversity in brainstorming. Nature Human Behaviour (2025), Matters Arising. Published 14 May 2025
  2. Si, Yang & Hashimoto. Can LLMs Generate Novel Research Ideas? ICLR (2025).
  3. Wang, Downey, Ji & Hope. SciMON: Scientific Inspiration Machines Optimized for Novelty. ACL (2024).
  4. Vasu et al. HARPA: A Testability-Driven, Literature-Grounded Framework for Research Ideation. IJCNLP-AACL (2026).