ChatGPT Prompts for Market Research
Around a dozen prompts you can paste today — planning, instrument design, synthesizing your own data, competitor analysis, and messaging. All built on one principle: the prompt does not fix hallucination. Grounding does.
The one thing to understand before you copy a single prompt
Most prompt lists for market research are quietly dangerous. They tell you to ask ChatGPT for market size, top competitors, customer demographics, or the leading pain points in a category — and the model answers fluently, with numbers and names, as if it knew. It does not. A plain ChatGPT answer is generated from training data with a cutoff date and no live sources. When pushed for facts and citations it will invent them. In a 2023 study published in Scientific Reports, Walters and Wilder found that 55 percent of GPT-3.5 citations and 18 percent of GPT-4 citations in literature reviews were entirely fabricated — references to papers that do not exist.
So here is the principle this whole page is built on: a prompt does not fix hallucination. Wording a question more cleverly does not make the model more truthful. What changes the outcome is grounding — giving the model your real material (notes, transcripts, survey exports, competitor pages, support tickets) and asking it to structure, summarize, or synthesize that material rather than supply facts from memory.
Every prompt below is written to be grounded. Either you paste in source material and the model reasons over it, or — for anything that needs current external data — you run it in a browsing or Deep Research mode and you verify the links yourself before you trust them. Used this way, ChatGPT is a fast analyst working on your evidence. Used the other way, it is a confident fabricator.
How to read the library
The prompts are grouped by research stage: planning, instrument design, synthesizing data you already have, competitor analysis from material you paste, and positioning and messaging. The table gives you the prompt, what it does, and a one-line note on how to ground it. The grounding note is not optional decoration — it is the part that makes the prompt safe to use.
Where a prompt needs real customer language or real pain signals as input, you will need genuine source material to feed in. That material has to come from somewhere honest: your own interviews and tickets, or a research tool that returns sourced, real-world text rather than a summary the model wrote from memory.
The prompt library
The synthesis and positioning prompts only work if the material you paste is real. Inventing example posts and feeding them in just launders the model's guesses back to you. You need genuine sourced text — your own transcripts and tickets, or a research tool that returns real threads with links.
Where the customer-language material comes from
Several of the strongest prompts above — theme extraction, voice-of-customer phrasing, pain-to-value-prop — are only as good as the raw text you feed them. If you have interviews, support tickets, or survey exports, start there. If you do not, you need a way to gather real customer language at scale without the model inventing it.
This is where a grounded Reddit research tool fits. rawneed pulls real discussion threads for a claim you define and returns a ranked report of those threads, classified by pain, willingness to pay, sentiment, and the tools people mention — each with a link back to the original post. That output is exactly the kind of sourced input the synthesis and positioning prompts are built for: real human wording, traceable to its source, ready to paste in. The model then structures what real people said instead of guessing what they might have said.
Honest caveats
None of these prompts make ChatGPT reliable on their own. Keep these limits in mind.
- A clever prompt does not fix hallucination. Grounding does. If the model has nothing real to work from, better wording just produces more confident fiction.
- Plain ChatGPT has a training cutoff and no live sources. Anything time-sensitive — market size, current pricing, who exists today — must come from a browsing or Deep Research mode, and you must verify the links.
- Citations are the highest-risk output. Walters and Wilder found a large share of model-generated references were entirely fabricated, so treat every citation as unverified until you open it.
- The model will quietly paraphrase even when you tell it not to. Spot-check verbatim quotes against your source before reusing them.
- Counts and percentages from a synthesis pass can be wrong. If a theme says nine people raised it, confirm nine actually did.
- Garbage in, confident garbage out. Fabricated or unrepresentative input produces clean-looking output that is still wrong.
- A prompt is a starting draft, not a finding. Treat every output as something to verify, not something to publish.
Want prompts grounded in real customer voices, not the model's memory?
The synthesis and positioning prompts here need real, sourced customer language to work. rawneed produces exactly that — a ranked report of real threads classified by pain, willingness to pay, sentiment, and tools, each linked to its source — so the model structures what people actually said. See how the research is gathered and graded before you trust it.
See how rawneed grounds its researchA workflow that ties it together
Run the planning prompts first to turn your hypothesis into questions you can falsify. Use the instrument-design prompts to build interviews or surveys that do not bias the answers. Gather real material — transcripts, tickets, or sourced threads. Feed that material into the synthesis prompts to find themes, frequencies, and verbatim language. Paste competitor copy into the competitor prompts to map the positioning landscape. Finally, use the positioning prompts to turn verified pains into messaging, rejecting any line that has no source behind it.
At no point does the model supply the facts. It plans, structures, summarizes, and drafts — and you supply and verify the evidence. That division of labor is the whole game.
Frequently asked questions
What are good ChatGPT prompts for market research?
The good ones ask the model to reason over material you supply — your transcripts, survey exports, or competitor copy — rather than to recall facts from memory. The library on this page is organized by stage: planning, instrument design, synthesis of your own data, competitor analysis, and positioning. Each prompt includes a note on how to ground it so the output stays tied to real evidence.
Can ChatGPT do market research on its own?
Not reliably on its own. Plain ChatGPT answers from training data with a cutoff and no live sources, and it will invent facts and citations when pushed. It is a strong analyst for material you give it — structuring interviews, coding survey text, comparing competitor copy — but the evidence has to come from you, and anything current must be gathered in a browsing mode and verified.
Will a better prompt stop ChatGPT from making things up?
No. A prompt does not fix hallucination. Better wording can reduce some sloppiness, but if the model has no real source to work from, a cleverer prompt just yields more confident fiction. The fix is grounding — giving it your real material and asking it to summarize or structure that, not to supply facts.
How do I get ChatGPT to use real customer language instead of guessing?
Feed it real customer text and tell it to keep the exact wording and not paraphrase. The text has to be genuine — your own interviews and tickets, or a tool that returns real sourced threads with links. If you paste invented examples, the model just hands you its own guesses dressed up as customer voice.
Are ChatGPT citations for market research trustworthy?
Treat them as unverified until you open them. In a 2023 Scientific Reports study, Walters and Wilder found 55 percent of GPT-3.5 citations and 18 percent of GPT-4 citations in literature reviews were entirely fabricated. Use a browsing or Deep Research mode for anything that needs sources, and click through every link before you rely on it.
Keep reading
See what people really say about your competitors
Track how buyers really compare tools and why they switch.
Read →ChatGPT for Market Research
What ChatGPT genuinely does well for market research, what it gets dangerously wrong, and the grounded workflow that keeps it useful.
Read →ChatGPT prompts for customer research
A copy-pasteable prompt library for understanding customers — grounded in real interview, support, and review data, not invented personas.
Read →AI vs traditional market research
AI is fast, cheap, and good at synthesis but it fabricates and cites nothing; traditional research is slow, costly, and verifiable. Here is how to use both.
Read →How to fact-check AI research
An analyst built a slide on a stat the model gave her. The citation looked real. She opened it — the source said the opposite. Here is the workflow that catches that before it ships.
Read →Reddit market research: reading a market in its own words
A founder spends six weeks and a chunk of budget on a survey, then learns more from one afternoon reading a niche subreddit. This is how to do Reddit market research on purpose instead of by accident.
Read →Reddit research tool: the honest guide to every type
Reddit is the most candid place on the internet, and the hardest to read at scale. This guide maps every type of Reddit research tool — from free keyword alerts to structured-report engines — so you can pick the one that fits the question you are actually asking.
Read →