Research methodology · Research insights
From a Rough Idea to a Research Hypothesis: A Complete AI Dialogue
Follow a worked AI research dialogue from a vague topic to a testable hypothesis, literature search and novelty matrix. Every sensor example is hypothetical and clearly labelled.
The Novelty Toolkit · Series 1 · Part 9 of 9
How do you turn a broad topic into a defensible research hypothesis? This worked dialogue connects research methodology, a reproducible literature search and a novelty matrix. The prompt examples are adaptable; the fictional sensor project is not a published study or a claim of discovery.
You do not need a perfectly worded research idea before beginning. You need a real question, a willingness to revise your assumptions and a method for deciding what counts as evidence. This worked tutorial shows exactly how an AI conversation can help without taking ownership of the scientific thinking.
All project facts below are hypothetical. No publication, sensor result or novelty claim is presented as verified. The example deliberately stops before declaring any discovery.
Starting point | The vague idea
A student begins with “I want to build an innovative AI-based pond-monitoring system.” That is a topic, not yet a research question. The student has access to low-cost sensors, a small pond and a few weeks of preliminary measurements.
- 01Vague topicAn AI-based pond-monitoring system.
- 02Observed mismatchReadings gradually disagree with a laboratory reference.
- 03Competing explanationsDrift, fouling, temperature or reference-handling error?
- 04Controlled comparisonDefine measurements and comparisons that can separate these causes.
Move 1 | Supply an observation only the researcher knows
Example prompt
I collected repeated pond-sensor measurements and noticed increasing disagreement with a laboratory reference over several weeks. I have not established the cause. Help me formulate three precise research questions. Do not recommend a specific algorithm until we identify relevant competing mechanisms.
Researcher decision: choose a question about the cause and correctability of drift, not a generic “AI device.” This is the first meaningful contribution to the dialogue.
Move 2 | Ask AI to explain concepts you are unsure about
Example prompt
Explain sensor drift, biofouling, calibration transfer, temperature sensitivity and reference-measurement error in plain language. Use a household thermometer analogy and then give the engineering meaning. Separate observations, plausible mechanisms and established mechanisms.
Researcher decision: correct any misunderstanding and record which physical variables were actually measured. AI explanations are learning aids; technical facts still need appropriate sources.
Move 3 | Create competing hypotheses
Example prompt
Propose four distinct explanations for the field-reference disagreement. Make one explanation a measurement artifact. For each, identify the expected pattern in time-series data, a possible confounder and the simplest discriminating test.
Researcher decision: identify which hypotheses can be examined with current instruments. Do not let the assistant invent data that were never collected.
Move 4 | Convert one hypothesis into a comparison
Example prompt
Help me frame a controlled comparison between fixed calibration and a context-aware correction method. State input variables, target outcome, reference standard, validation split, likely leakage risks and limitations. Ask me questions wherever real data or domain judgment is required.
Researcher decision: define a meaningful success measure before model selection. For example, compare error and stability across held-out deployment periods, with uncertainty reported. The exact method depends on domain constraints.
Move 5 | Search for the closest previous solution
Example prompt
Create a reproducible prior-work search plan on long-term low-cost water-quality sensor drift correction. Include broader terms, synonyms, database-specific queries, conference and patent routes, and forward/backward citation searching. Do not claim the combination is new until closest papers have been compared in full.
Researcher decision: execute the queries in actual scholarly databases and read the closest articles. A bibliography assembled from AI alone is insufficient.
Move 6 | Build a novelty comparison matrix
| Feature / claim | Closest work A | Closest work B | Proposed work |
|---|---|---|---|
| Long-term field records | To verify | To verify | Planned |
| Correction approach | To verify | To verify | Context-aware candidate |
| Independent reference | To verify | To verify | Planned |
| Out-of-period validation | To verify | To verify | Planned |
| Key unresolved question | To verify | To verify | Effect on field reliability |
This matrix intentionally contains “To verify” instead of invented values. The scholar fills those cells only after inspecting the actual papers. The matrix might show the proposed method is already known, in which case the hypothesis must be revised.
Move 7 | Invite the toughest criticism
Example prompt
Assume this project will be rejected for insufficient novelty or invalid methods. List the five strongest reasons. For each, tell me exactly which paper, observation, control or analysis would help resolve the objection. Separate fatal flaws from fixable weaknesses.
Move 8 | Write an honest contribution statement
Example prompt
Based only on the studies and verified evidence I provide, draft a cautious two-sentence research contribution statement. It must name the closest prior work, specify my difference and state what experiment could demonstrate its significance. If the novelty evidence is incomplete, retain that uncertainty.
- 01Verify the bibliographyRead the indispensable sources and record the search trail.
- 02Compare honestlySeparate established prior findings from planned work.
- 03Challenge the methodCheck controls, reference measurements and validation leakage.
- 04Choose the next testIdentify evidence that could overturn the preferred explanation.
- 05State a bounded contributionRetain uncertainty wherever the evidence is incomplete.
The final 30-minute review checklist
- Can I state the problem without mentioning the words “novel” or “AI”?
- Can I identify the closest published method and explain its limitation fairly?
- Have I verified every indispensable bibliographic source?
- Have I proposed an experiment that could falsify my preferred explanation?
- Do I know which claims remain unverified or confidential?
- Can another researcher reproduce my search and comparison?
AI can make a researcher’s thinking more informed, but the decisive work is still the scholar’s: observing carefully, questioning assumptions, searching honestly and designing evidence that could change the conclusion.
What the evidence does—and does not—show
The dialogue uses the clarity and iteration encouraged by official prompting guidance. Its comparison steps are informed by literature-grounded novelty assessment and RINoBench, while HARPA illustrates why testability belongs alongside literature comparison. None of those publications verifies the hypothetical pond project.
PRISMA-S is a reporting guideline for literature searches in systematic reviews. Its emphasis on recording sources, search strategies and dates is useful when documenting this example’s search. Using those habits alone does not make a novelty audit a complete systematic review, or establish that a proposed method is new.
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.
Related articles
Sources and further reading
- OpenAI. Prompt engineering best practices for ChatGPT. Official guidance, accessed 22 November 2026.
- Shahid et al. Literature-Grounded Novelty Assessment of Scientific Ideas. Scholarly Document Processing workshop (2025).
- Schopf & Färber. Is This Idea Novel? An Automated Benchmark for Judgment of Research Ideas. LREC (2026).
- Vasu et al. HARPA: A Testability-Driven, Literature-Grounded Framework for Research Ideation. IJCNLP-AACL (2026).
- Rethlefsen et al. PRISMA-S: an extension to the PRISMA Statement for Reporting Literature Searches in Systematic Reviews. Systematic Reviews (2021).