AI Prompts for Market Research: A Verifiable Prompt Framework

AI prompts for market research are structured instructions that help a researcher investigate a market, compare competitors, analyze customer evidence, and test assumptions. The best prompts specify the decision, geography, time period, supplied data, required output, and verification rules. Use AI to organize evidence and generate hypotheses; use cited sources and real participants to establish findings.

Which 25 prompts cover the research workflow?

These 25 AI market research prompts cover five practical jobs: understand the market, assess competitors, examine buyers, analyze numbers, and brief a decision-maker. Replace bracketed fields with your project details and attach approved evidence where requested. Each prompt is designed to expose a gap rather than hide one.

Industry and market analysis prompts

• Market boundary: “Define the [product or service] market in [geography]. List included and excluded use cases, buyer types, substitutes, and terms that sources use differently. Flag boundaries requiring analyst judgment.”
• Industry signals: “From these [dated reports and datasets], extract developments relevant to [decision]. For each, give the source date, affected buyer segment, and a plausible implication. Do not present an implication as an observed outcome.”
• Market entry: “Assess entry into [geography] for [offering]. Separate documented requirements, competitive barriers, buyer assumptions, and unanswered regulatory questions. Identify claims requiring a qualified local adviser.”
• Demand proxies: “For [market], propose measurable demand indicators available in [approved sources]. Explain what each indicator captures, what it misses, and whether it can be compared across [regions or periods].”
• Scenario analysis: “Using these supplied inputs, build downside, base, and upside scenarios for [decision]. Show the assumption changed in each scenario. Do not describe the scenarios as forecasts validated by evidence.”

Competitor and positioning prompts

• Rival inventory: “From these dated public sources, list companies serving [buyer and use case]. Distinguish direct competitors, substitutes, and adjacent vendors; give a source for each classification.”
• Positioning comparison: “Compare [vendors] using only their supplied product and pricing pages. Create a table of stated audience, capabilities, pricing visibility, and missing information. Do not infer unlisted capabilities.”
• Claim audit: “For each statement in this competitor brief, label it verified public fact, reasonable interpretation, or unsupported claim. Quote the relevant source passage and its date.”
• Win/loss hypotheses: “Given these anonymized, consented win/loss notes, identify recurring stated reasons and contradictory accounts. Do not role-play a buyer or invent reasons absent from the notes.”
• Expert-interview gap: “Identify three consequential questions about [competitor or channel] that public evidence cannot answer. Draft neutral questions for a qualified former operator or buyer; avoid requests for confidential information.”

Customer and interview prompts

• Buyer-role map: “Using these interview notes, map who initiates, evaluates, approves, and uses [offering]. Cite note IDs for each role and mark roles with no direct evidence.”
• Jobs and friction: “Code these permissioned customer responses for desired outcomes, workflow obstacles, and purchase triggers. Return a codebook, response IDs, contradictory examples, and any ambiguous passages.”
• Interview guide: “Draft a neutral guide for [buyer role] to investigate [decision]. Begin with recent behavior, then probe alternatives and evaluation criteria. Remove leading questions and flag anything requiring consent.”
• Segment contrast: “Compare [segment A] and [segment B] using only the supplied respondent records. Separate observed differences from patterns that may reflect sample composition; include the underlying response IDs.”
• Persona evidence check: “Audit this buyer persona against the attached interviews and CRM definitions. Mark each attribute supported, contradicted, or untested; suggest the next question needed to test unverified attributes.”

Sizing and survey prompts

• Bottom-up sizing: “Estimate the addressable value of [offering] in [geography] using the attached count of eligible buyers, adoption assumptions, and annual spend. Show units, formula, exclusions, and a sensitivity table; leave missing inputs blank.”
• TAM/SAM/SOM definitions: “Define total addressable market (TAM), serviceable available market (SAM), and serviceable obtainable market (SOM) for [offering]. State which supplied input supports each narrowing step; do not assume an obtainable share.”
• Source reconciliation: “These two reports give different estimates for [market]. Compare their dates, geographies, segment definitions, currency, and methods. Do not average the figures without establishing comparability.”
• Survey-question review: “Review this [B2B or consumer] survey for leading wording, overlapping answer options, missing ‘not applicable’ choices, and questions that require knowledge respondents may not have. Suggest neutral revisions.”
• Open-response analysis: “Code these consented survey responses into a documented theme list. Count responses per theme only after assigning response IDs; include multi-coded responses, disagreements, and quotations only where sharing permission allows.”

Synthesis and decision prompts

• Evidence matrix: “Build an evidence matrix for [decision] from the attached sources. For each conclusion, identify supporting evidence, conflicting evidence, source date, and the person responsible for verification.”
• Assumption stress test: “List the assumptions underlying this market recommendation. Rank them by how much changing each assumption could alter the recommendation; explain the ranking without inventing probabilities.”
• Research-method choice: “For these unresolved questions, recommend desk research, customer interviews, expert interviews, or a survey. Explain which method can answer each question and which inference it cannot support.”
• Executive brief: “Write a one-page brief using only the validated findings in this evidence matrix. Separate findings, implications, unresolved questions, and a recommended next research action.”
• Decision log: “Convert this research review into a decision log with decision owner, evidence used, assumptions accepted, open risks, and the condition that would trigger reconsideration. Leave owners or dates blank if not provided.”

The prompts work with ChatGPT and other generative AI tools, but the same rule applies to each: an attached source is not proof that the model interpreted it correctly. Compare consequential outputs with the original material.

How do you write research prompts that can be checked?

Write a market research prompt around one decision, a defined evidence set, and a required output format. Specify the market and date range, tell the model to separate sourced facts from inference, and require an “unknown” category. That instruction makes a weak but honest answer more useful than a confident answer with no traceable basis.

• Name the decision: “Should we prioritize private-cloud buyers in Germany?”
• Set the scope: Specify geography, buyer type, product boundary, and relevant dates.
• Supply or constrain evidence: Attach approved reports, interview notes, or public URLs; identify material the tool must not use.
• Define the output: Request a table with claim, source, date, inference, confidence rationale, and missing evidence.
• Demand a failure mode: Instruct the model to say “insufficient evidence” when a claim cannot be supported.

The Nexus Evidence Gate

The Nexus Evidence Gate is an editorial checklist for AI-assisted research. It is a proposed framework for this article, not a claim about a pre-existing proprietary methodology.

• Scope: Does the answer address the specified buyer, market, geography, and period?
• Source: Can the reader locate the evidence behind each consequential claim?
• Status: Is each statement labeled as observed evidence, calculation, inference, or hypothesis?
• Sensitivity: Would a different assumption materially change the decision?
• Specialist: Which unresolved point needs a real buyer, operator, or expert?

Reusable base prompt

Act as a research analyst supporting [decision] in [market and geography] during [period]. Use only [approved sources or attached files]. Return a table with claim, exact source location, source date, evidence status, assumptions, and what would falsify the claim. Show calculations and units. If the evidence is absent or conflicting, write “insufficient evidence”; do not invent figures, quotations, buyer opinions, or citations.

How do you validate an AI research finding?

Validate a finding by tracing the claim to a source, checking whether the source supports that exact wording, and testing the assumptions that connect evidence to a decision. There is no universal number of sources or fixed number of minutes that makes a market claim reliable. Validation effort should rise with the claim’s impact and the cost of being wrong.

For market sizing, how much source data is enough? Enough to identify and defend every material input, not a prescribed count of documents. For validation time, allocate review according to decision risk rather than treating a quick AI answer as a shortcut around verification. Refresh source dates when the decision is revisited or a material input changes; a quarterly schedule may suit one market and be too slow or too frequent for another.

A client-ready sentence should make the distinction visible: “Vendor A’s current page lists feature X” is a source-backed observation. “Feature X will win enterprise buyers” is a hypothesis until buyer evidence supports it. AAPOR’s 2026 guidance on AI-assisted survey research emphasizes task-specific validation, human oversight, and disclosure rather than assuming an AI result transfers reliably across uses.aapor

When should AI hand off to primary research?

AI should hand off to primary research when a decision depends on current buyer behavior, operating practice, or information that public sources cannot establish. AI can identify the uncertainty and draft a neutral question. A real participant supplies the account; the researcher checks whether that account applies beyond the participant’s experience.

For example, an AI review may suggest that data residency shapes private-cloud purchasing. The next research question is not “Confirm that data residency is your top priority.” Ask: “Walk me through the last infrastructure evaluation. Which requirements ruled vendors in or out, and who set them?”

What should never go into a research prompt?

Do not place confidential client information, identifiable respondent material, or protected competitor information into an AI tool without authorization and an approved data-handling route. Also do not ask the model to invent interview quotes, imply it contacted people, or present simulated responses as a real sample.

• Client or participant data: Check consent, contract terms, access controls, retention, and the approved workspace before uploading transcripts or CRM exports.
• Competitor intelligence: Use lawful, appropriately obtained public material; do not solicit confidential information from current or former employees.
• Synthetic buyers: Label AI-generated responses as simulations. Do not count them as interviewed people or observed demand.
• High-stakes advice: Escalate legal, regulatory, privacy, and financially material conclusions to qualified professionals.

Tool terms vary. OpenAI says data from ChatGPT Business and Enterprise workspaces is not used to train its models by default, but that statement does not replace a client’s confidentiality terms, consent requirements, or internal approval process. AAPOR likewise calls for clear disclosure of AI’s role, human oversight, and protections for research participants.

Frequently asked questions

These are short answers to common implementation questions about AI-assisted market research. The governing distinction is whether an output comes from verified evidence, a real participant, a calculation, or a simulation.

Can AI-generated personas count as customer interviews?
No. A generated persona can help draft questions or expose assumptions, but it is not an interviewed customer. Report synthetic responses separately from human responses; AAPOR warns that synthetic responses pose validity and disclosure risks when used beyond appropriately labeled exploratory work.aapor

Can an AI tool estimate a competitor’s private market share?
It can construct a clearly labeled scenario from disclosed inputs. It cannot verify a private company’s market share merely by producing a plausible percentage. Leave the estimate unverified if the inputs cannot be checked.

Should an expert interview confirm a single market-size number?
No. Ask the expert about the mechanisms and assumptions behind the estimate such as eligibility, adoption, or purchasing practice. Treat one expert’s account as contextual evidence, not a statistically representative measurement.

Can I put client interview transcripts into ChatGPT?
Only if your client agreement, participant permissions, and organization’s approved tool settings allow it. Remove unnecessary identifiers and have the appropriate privacy or legal owner review uncertain cases.

What should you do next?

Choose one live research decision and run the matching prompt against approved evidence. Mark the resulting claims as verified, inferred, or unknown; then turn the most consequential unknown into a neutral question for a buyer or specialist. Nexus Expert Research describes a process of recruiting experts to a project brief and screening them for relevance.