AI for Market Research & Idea Discovery — Workflow & Prompt Bank

Why this workflow helps

If you want fast, focused market research and idea discovery without hiring an analyst, this workflow shows how to combine human judgment with AI to rapidly surface opportunities, generate testable hypotheses, and design low‑cost validation. The AI helps you scale synthesis and idea generation; you must still verify signals with primary research and experiments to avoid false or biased conclusions.

When to use

Use this workflow when you need to:

• Scan competitors, industry writing, and customer conversations quickly
• Synthesize large or messy notes into actionable insights
• Generate hypothesis statements and validation plans
• Draft interview guides and landing page test copy

High‑level steps

1. Collect & ingest sources. Gather competitor pages, product docs, reviews, forum threads, social posts, job ads, and any interview transcripts or notes.
2. Auto‑summarize & categorize. Use AI to produce concise source summaries and tag by theme, customer segment, or problem area.
3. Competitive scan & matrix. Extract features, positioning, pricing, strengths, and weaknesses into a structured matrix.
4. Synthesize jobs, pains & desired outcomes. Convert signals into jobs‑to‑be‑done statements and prioritized outcome metrics.
5. Generate hypotheses. Formulate clear, testable hypotheses about who will pay and why, and what the simplest offering could be.
6. Design validation. Create interview guides, landing page copy, and micro‑experiments to test demand and willingness to pay.
7. Verify & iterate. Run primary research, collect responses, and reconcile AI output with observed customer signals. Update the knowledge base and repeat.

Step details, actions, and prompt templates

1) Collect & ingest

Action: Save source URLs, transcripts, and raw text into a single folder or a single document per source. For long documents, split into 400–1,000 token chunks before ingesting.

Prompt (Ingest Summarizer):

"Summarize the following source in 3–5 bullets: origin (URL or source), target customer, main claims, supporting evidence, and any quoted customer pain. Keep each bullet 20 words or less."

2) Auto‑summarize & categorize

Action: Ask the model to tag summaries with theme labels (e.g., onboarding friction, pricing complaints, missing integrations) and cluster similar signals.

Prompt (Tagger & Cluster):

"Given these summaries, assign 1–3 theme tags to each and group similar items. Provide a short rationale for each cluster."

3) Competitive scan & matrix

Action: Build a simple competitor matrix (rows = competitor, columns = positioning, pricing, key features, target customer, major complaints, differentiation opportunities).

Prompt (Matrix Builder):

"Create a competitive matrix from these summaries. For each competitor, list: one‑sentence positioning, price tiers, notable features, common complaints, and 2 areas they under‑deliver that could be an opportunity."

4) Synthesize jobs, pains & desired outcomes

Action: Convert observed customer language into jobs‑to‑be‑done (JTBD) and measurable desired outcomes (e.g., reduce time to X by Y%). Prioritize outcomes by frequency and intensity across sources.

Prompt (JTBD Extractor):

"From these clustered signals, extract 6–10 'jobs to be done' in the format: When [situation], I want to [motivation], so I can [expected outcome]. Also propose 3 measurable success metrics per job."

5) Generate hypotheses

Action: Write crisp hypotheses that connect customer segment, job, proposed solution, and expected value.

Template (Hypothesis):

"We believe [customer segment] who struggle with [job/pain] will value [solution/feature] because it delivers [measurable benefit]. We'll know we're right if [success metric] changes by [amount] in [timeframe]."

6) Design validation experiments

Action: Select low‑cost tests: one‑page landing pages with email capture or paid ads, short qualitative interviews, micro‑sales offers, or concierge prototypes.

Prompt (Interview Guide Generator):

"Create a 12‑question interview guide to validate the hypothesis: avoid leading questions, start with context, surface goals and current behavior, end with willingness‑to‑pay probes."

Prompt (Landing Page Copy):

"Write a 4‑section landing page headline, subhead, three benefit bullets, a social proof placeholder, and a clear CTA to 'Get Early Access' that tests value proposition for [customer segment]."

Prompt bank (select and adapt)

• Ingest Summarizer (see above)
• Competitor Matrix Builder (see above)
• JTBD Extractor (see above)
• Hypothesis Generator: "Propose 5 testable hypotheses from these JTBD and competitor gaps."
• Interview Guide Generator (see above)
• Landing Page Copy Tester: "Draft two headline variants and 3 bullets for A/B test; include a single quantitative CTA."
• Uncertainty Detector: "List key assumptions and rank them by risk and impact, and suggest a first experiment for each assumption."

Avoiding the mal‑hunger: verify primary signals

Common risk: relying on AI summaries without checking customer reality. Use these guardrails:

• Triangulate: confirm each major insight with at least two independent primary signals (interviews, survey responses, usage data, or paid-test results).
• Record evidence: for every insight, save the original quote/source and a one‑line reason you trust it.
• Prefer behavior over opinion: willingness to pay, clickthroughs, signups, or actual purchases beat survey intentions.
• Spot‑check AI output: randomly sample 10% of AI summaries and compare them to the original source for fidelity.

Outputs to capture (minimum)

1. Source index (URL/file, date, short summary)
2. Competitive matrix (CSV or table)
3. JTBD list with metrics and prioritization
4. Top 3 hypotheses with success metrics
5. Interview transcripts and landing page results
6. Experiment outcomes and next decisions

Quick checklist before running experiments

• Do I have a clear hypothesis with measurable success criteria?
• Have I captured primary evidence backing the hypothesis?
• Is the test cheap and fast to run?
• Will the result reduce uncertainty for the riskiest assumption?

Next steps and suggested cadence

Run 3–5 lightweight experiments in parallel for 2–4 weeks, then regroup to update JTBD and hypotheses. Keep a living research folder and tag artifacts by hypothesis ID and experiment outcome so you can trace decisions back to evidence.

Where this can get better with platform capabilities

Suggested capability enhancements below describe how the workflow could become interactive, reproducible, and shareable inside The Hunger Engine.