How to Use ChatGPT for Market Research: A Source-Checked Workflow

ChatGPT is most useful in market research as a structuring and drafting partner, not as a source. Use it to frame questions, organise messy inputs, cluster themes, and draft interview or survey instruments — then verify every factual claim against primary sources such as company websites, filings, official statistics, standards bodies, or your own collected data. Treat every output as a provisional hypothesis that must survive a source check before it enters a deck, a strategy document, or a decision.

The workflow below is written for early-stage research: the phase where you are still deciding which market, segment, competitor set, or message deserves real investment of time and budget.

What ChatGPT can and cannot do in market research

The distinction is simple but easy to lose under deadline pressure: the model generates plausible language, not verified evidence.

What it does well:

• Framing. Turning a vague question (“should we enter this segment?”) into a set of answerable sub-questions.
• Structuring. Building research briefs, discussion guides, screener logic, and coding frameworks.
• Summarising inputs you supply. Condensing pasted reviews, call notes, transcripts, or public product pages into themes.
• Language work. Rewriting jargon into customer language, drafting neutral survey wording, generating alternative message variants to test.
• Adversarial review. Arguing against your own conclusion so you can see where it is thin.

What it cannot do reliably:

• Produce market sizes, growth rates, shares, pricing, or headcounts that you can trust without checking.
• Tell you what a named company currently offers, charges, or claims.
• Represent your customers. Synthetic “personas” are writing aids, not respondents.
• Guarantee that a citation, report title, or statistic it names actually exists.
• Access anything you have not given it, unless you are using a version with live browsing and you check the links it returns.

A practical rule: if a sentence contains a number, a named source, a date, or a claim about a specific organisation, it is unverified until you have opened the source yourself.

Start with a one-page research brief

Most weak AI-assisted research fails at the beginning, not the end. Without a brief, the model fills the gap with generic content and you end up editing fluent text that answers no real question.

Write the brief first, in plain text, and paste it at the top of every session.

RESEARCH BRIEF

Decision this research supports:
  (What will change depending on the answer?)
Decision owner and deadline:
Business question (one sentence):
Sub-questions (3-6, each answerable):
Market / segment definition:
  Geography:
  Buyer role(s):
  Company size or customer type:
  In scope / explicitly out of scope:
What we already believe (assumptions to test):
Evidence we already hold:
  (internal data, past studies, sales notes, analytics)
Evidence we will need to collect:
  (interviews, survey, public sources, pricing pages)
Acceptable sources for factual claims:
  (official statistics, filings, standards bodies, vendor sites, our own data)
What "good enough to decide" looks like:
Known constraints (budget, time, access to respondents):

A brief this short takes fifteen minutes and removes most of the ambiguity the model would otherwise invent.

A five-step source-checked workflow

Step 1 — Define and decompose. Paste the brief and ask for a decomposition of your business question into sub-questions, each labelled with the evidence type that could answer it (desk research, internal data, qualitative interviews, quantitative survey). Reject sub-questions no realistic source can answer.

Step 2 — Map the landscape as hypotheses. Ask the model to list candidate competitors, substitutes, buyer roles, and objections — explicitly labelled as hypotheses to check. Then verify each item yourself: open the company site, the pricing page, the job listings, the documentation. Delete anything you cannot confirm.

Step 3 — Collect real inputs. This is the step that produces evidence: public product and pricing pages, official statistics and regulator or standards publications, industry associations, customer reviews, support tickets, sales call notes, search and analytics data, and interviews you conduct yourself. Keep a source list with URL, publisher, and date.

Step 4 — Analyse the inputs you collected. Now use the model on your own material: cluster review complaints into themes, compare competitor messaging, identify contradictions between segments, draft a coding frame for open-ended survey answers. Require every theme to cite the specific input it came from so you can trace it back.

Step 5 — Stress-test and write up. Ask for the strongest counter-argument to your emerging conclusion and for the evidence that would falsify it. Then write the deliverable yourself, marking each claim as verified, internal data, or hypothesis. Anything unverified either gets a source or gets cut.

Three reusable prompt patterns

Each pattern ends with the same discipline: separate facts, assumptions, and follow-up questions. That separation is what makes the output auditable.

Pattern 1 — Competitor and message scan

You are helping structure early-stage market research. Do not invent data.

Brief: [paste research brief]
Inputs: [paste text you copied from competitor sites, pricing pages,
docs, or ad copy - include source URL and date for each]

Tasks:
1. For each input, summarise the positioning, target buyer, and core
   promise in one line, referencing the source it came from.
2. Compare across inputs: shared claims, points of differentiation,
   claims that are vague or unsupported.
3. Identify gaps in the market conversation that our brief cares about.

Output in three labelled sections:
FACTS - only statements traceable to the inputs I pasted, each with
  the source label.
ASSUMPTIONS - your inferences and anything you filled in from general
  knowledge, each flagged as unverified.
FOLLOW-UP QUESTIONS - what I must check or collect next, and where.

Pattern 2 — Review and theme analysis

Analyse only the customer feedback below. Do not add outside examples.

Brief: [paste]
Feedback: [paste reviews, tickets, or call notes]

Tasks:
1. Cluster the feedback into themes. For each theme give a neutral name,
   how many items belong to it, and two short representative quotes.
2. Note where sentiment conflicts within a theme or differs by segment.
3. Flag themes supported by very few items as low-confidence.

Output in three labelled sections: FACTS (grounded in the pasted items,
with quotes), ASSUMPTIONS (interpretation, motivation, or causation you
inferred), FOLLOW-UP QUESTIONS (what to probe in interviews or measure
in a survey).

Pattern 3 — Interview and survey question design

Help me design research instruments. Do not write findings.

Brief: [paste]
Hypotheses to test: [list]

Tasks:
1. Draft a 30-minute interview guide: warm-up, current behaviour,
   problem and workaround, decision process, close. Open, non-leading
   questions only, with probes.
2. Draft up to 12 survey questions mapped to my hypotheses, with answer
   options and screener logic.
3. Review your own draft for leading wording, double-barrelled items,
   loaded terms, and missing answer options.

Output in three labelled sections: FACTS (which hypothesis each question
tests, based only on my brief), ASSUMPTIONS (anything you assumed about
my market, buyers, or terminology), FOLLOW-UP QUESTIONS (what I must
confirm before fielding).

What to verify before you use an output

Output	Useful role	Required human or source check
Market size, growth, or share figures	Rough hypothesis to test	Replace with figures from official statistics, filings, or a named published dataset you have opened; otherwise cut
Competitor list	Starting shortlist	Confirm each company exists, is active, and competes in your defined scope
Competitor features, pricing, or claims	Prompt for where to look	Read the live product, pricing, or documentation page and record URL and date
Customer personas or segments	Hypothesis and discussion aid	Validate against real interviews, CRM data, or survey responses
Themes from feedback you pasted	Analysis shortcut	Spot-check a sample of quotes against the source items; check counts
Cited reports, articles, or authors	Lead to follow	Locate the original publication; treat as non-existent until found
Survey and interview questions	Draft instrument	Review for leading and double-barrelled wording; pilot with 3-5 people
Regulatory, clinical, or financial statements	None	Out of scope for this workflow; consult a qualified professional
Strategic recommendation	Argument to pressure-test	Confirm every supporting claim is verified, and state the decision risk

Common failure modes

• Treating fluency as evidence. Confident phrasing is a property of the model, not of the claim.
• Skipping the brief. Generic prompts produce generic output that survives editing because it is never wrong enough to notice.
• Fabricated sources. Report titles, author names, and URLs can be invented. Always open them.
• Synthetic respondents replacing real ones. Personas cannot tell you what people will pay or switch away from.
• Anchoring. The model’s first landscape map quietly becomes your scope, including its omissions.
• Leading questions. Instruments drafted from your hypotheses tend to confirm them unless reviewed.
• Losing provenance. Once a claim is copied into a slide without its source, it becomes permanent.
• Pasting confidential material. Check your organisation’s policy and the tool’s data settings before entering customer or commercial data.
• One long chat. Long sessions blur what was verified and what was generated; run separate sessions per sub-question and keep a source log outside the chat.

FAQ

What are some good ChatGPT prompts for marketing?

The useful ones are structured and grounded in material you supply. Strong patterns include: summarise the positioning and core promise of these pages I pasted; cluster these customer reviews into themes with quotes; rewrite this feature description in the language customers used in these call notes; draft five message variants for this segment and list what evidence would decide between them; critique this landing page copy for vague or unsupported claims. Add the same closing instruction each time — separate facts, assumptions, and follow-up questions — and refuse any output that states numbers you have not verified.

What are the 7 basic questions in market research?

Different textbooks phrase them differently, but a widely used working set is:

1. What decision does this research need to support?
2. Who is the customer, and how do we define the segment?
3. What problem are they trying to solve, and what do they do today?
4. How large and reachable is the opportunity?
5. Who else serves this need, and how are they positioned?
6. What would make someone switch, buy, or pay more?
7. What evidence would change our mind, and how will we collect it?

Answer these in your brief before prompting anything.

How to use AI in market research?

Use it at the structuring and synthesis stages, and keep evidence collection human. Concretely: decompose the business question, draft the brief and instruments, organise and cluster inputs you have gathered, translate findings into clear language, and stress-test conclusions. Keep sourcing, fieldwork, sampling decisions, and final judgement with the researcher, and maintain a source log so every claim in the deliverable can be traced.

Closing checklist

• One-page brief written, with the decision it supports named.
• Business question decomposed into answerable sub-questions.
• Every prompt includes the facts / assumptions / follow-up questions split.
• Real inputs collected, with publisher and date recorded.
• Analysis run on your own material, not on model recall.
• Every number, name, and citation opened at its source or cut.
• Instruments reviewed for leading wording and piloted.
• Counter-argument documented alongside the conclusion.
• Claims labelled verified, internal data, or hypothesis in the write-up.
• Confidentiality and data-handling policy checked before pasting anything.

📌 This article is editorial guidance on process, not a source of market data, and it offers no medical, legal, or financial advice. Responsibility for verifying any claim produced during your own research remains with you. For further reading on the same topic, see Zapier’s “How to use ChatGPT for market research” and Voxpopme’s “ChatGPT in Market Research”; both are third-party sources listed for context only and are not the basis of the workflow above.