These 14 customer research prompts help you plan a study around a real decision, recruit the right customers, write interview guides and surveys that don't lead, and turn transcripts into findings backed by quotes. The model does the drafting and sorting; the evidence still comes from talking to customers.
How to use these prompts
Each prompt works in ChatGPT, Claude or Gemini. Replace the blanks in double braces, or keep the default after the colon: {{segment:small agencies}} means "small agencies" unless you change it.
Start every study with the decision it will inform. "Learn about our customers" produces a pile of notes; "decide whether to build team billing this quarter" produces findings someone acts on.
These prompts lean on one principle from Nielsen Norman Group's article First Rule of Usability? Don't Listen to Users: "To design the best UX, pay attention to what users do, not what they say." That is why the interview prompts ask about past events, and the synthesis prompts separate what people did from what they said they would do.
For usability testing, screeners and heuristic reviews, see the separate UX research prompts.
Which prompts plan customer research?
1. Research plan from a decision
Keeps the study small and pointed at something you will decide.
I need to decide: {{decision:whether to build team billing this quarter}}.
What we already know (data, past research, support tickets):
{{known}}
Write a one-page research plan: the 3 to 5 questions we must answer to make
this decision, which customers to talk to and why, the method for each
question (interviews, survey, usage data, support history), what result
would make us say yes or no, and a timeline of {{weeks:2}} weeks. Flag any
question we could answer from existing data without new research.
2. List the assumptions to test
Most plans rest on beliefs nobody wrote down.
Here is the plan or idea we're researching:
{{idea}}
List every assumption it depends on about customers: who has the problem,
how often, what they do today, what it costs them, and whether they'd pay
or switch. For each, note the evidence we already have (data, interviews,
tickets) or mark it "none". Rank them by how much the plan fails if each is
wrong. For the top 3, write the interview or survey question that would
test it, and the answer that would make us change the plan.
Which prompts help you recruit customers to talk to?
3. Pick who to talk to from a customer list
Avoids talking only to your friendliest customers.
Here is a customer export with columns {{columns:plan, signup date, last active, seats, MRR, churned}}:
{{customer_list}}
For the research question "{{question}}", propose a recruiting mix of
{{count:12}} customers across segments that matter, including people who
churned or barely use the product, not just fans. Explain the mix, then list
the criteria to filter by. Don't pick named individuals; give me the rules.
4. Interview request email
Short, honest about time, and clear that you're not selling.
Write an email asking a customer for a {{minutes:30}}-minute research call
about {{topic:how they handle invoicing}}.
Who it's from: {{sender:the founder}}
Thank-you we offer: {{incentive:a $50 gift card}}
Rules: under 110 words, say clearly this is not a sales call, say what we
will and won't do with what they tell us, give a scheduling link
placeholder, and make it easy to decline. Plain subject line. Give 2
versions: one for active customers, one for customers who left.
Which prompts write interview guides?
5. Discovery interview guide about past behavior
The core guide. Questions about the last real time beat questions about "usually".
Write a {{minutes:30}}-minute customer interview guide about
{{topic:how they handle invoicing}} for {{segment:small agencies}}.
Rules:
- Ask about specific past events ("Tell me about the last time you..."),
not opinions or predictions ("Would you use...").
- Never mention our product idea or a solution.
- Include follow-up probes for each question: what happened next, what it
cost them, what they tried, who else was involved.
- Order: warm-up (2 min), the story of the last time (15 min), workarounds
and spending (8 min), wrap-up (5 min).
End with a short list of things to listen for that would prove our
assumptions wrong: {{assumptions}}
6. Churn interview guide
For customers who left. The goal is the story of the decision, not a defense of the product.
Write a {{minutes:20}}-minute interview guide for customers who cancelled
{{product:our product}} in the last {{months:3}} months.
Cover: what they hoped it would do when they signed up, the moment they
first thought about leaving, what else they considered, what finally
triggered the cancellation, and what they use now. Neutral wording, no
attempts to win them back during the call, no defending the product. Add
probes for vague answers like "it was too expensive" or "we didn't use it".
7. Win-loss interview guide
Asks buyers why they chose you or someone else.
Write a win-loss interview guide for {{deal_type:B2B deals over $10k a year}}.
Two versions: one for deals we won, one for deals we lost.
Cover: the problem that started the search, who was involved in the
decision, which options they compared, how they compared them, what almost
changed their mind, and the deciding factor. Questions must be neutral
about competitors. Include a short intro the interviewer reads that
explains the call is not a sales follow-up.
8. Find and fix leading questions
Run it on any guide before the first call.
Review this interview guide or survey for questions that lead the
participant, assume an answer, ask about two things at once, ask for
predictions about future behavior, or mention our solution.
{{guide}}
Return a table: original question, problem, neutral rewrite. Keep the
intent of each question. Then list any research question from this plan
that the guide doesn't cover: {{research_questions}}
Which prompts write surveys?
Pew Research Center's guide to writing survey questions says "It is also important to ask only one question at a time," and that "Pretesting a survey is an essential step in the questionnaire design process." The two prompts below apply both points.
9. Survey draft from research questions
A short survey with a reason behind every question.
Write a customer survey of no more than {{max_questions:10}} questions to
answer these research questions:
{{research_questions}}
Audience: {{segment:small agencies}}, reached by {{channel:email}}.
For each survey question give: the wording, the answer type (single choice,
multiple choice, scale, open text), the answer options, and which research
question it serves. One idea per question, simple words, a "none of these"
or "other" option where needed, and balanced scales. Put sensitive or
demographic questions last. End with 3 questions to ask 5 people in a
pretest.
10. Analyze open-text survey answers
For the column of free text nobody wants to read.
Here are {{count:300}} open-text answers to the survey question
"{{question}}":
{{answers}}
Group them into themes. For each theme: a one-line name, how many answers
fit it, and 3 representative quotes copied exactly. Put answers that fit no
theme in "Other" with a count. Don't merge themes to make the list shorter,
and don't infer anything the answers don't say.
Which prompts make sense of what customers said?
11. Transcript to quote-backed notes
Quotes first, conclusions second. Anthropic's guide to reducing hallucinations recommends that, for long documents, you ask the model to extract word-for-word quotes before doing the task, which grounds the answer in the actual text.
Here is a customer interview transcript (names removed):
{{transcript}}
Step 1: extract the exact quotes that relate to {{topic:how they handle invoicing}}.
Number them.
Step 2: using only those quotes, write notes under: what they did (past
events), problems and their cost, workarounds and tools, what they said
they want, and surprises. Cite quote numbers for every point. Mark
statements about the future ("I would...") separately from things they
actually did.
12. Synthesize across interviews
Turns a stack of notes into findings with counts, so one loud customer doesn't set the roadmap.
Here are notes from {{count:10}} customer interviews, one block per
participant, each labeled with a participant ID and segment:
{{all_notes}}
Find the patterns. For each finding: a one-sentence statement, how many
participants support it (with their IDs), the strongest quote, and any
participant who contradicts it. Separate findings about behavior from
findings about stated wishes. Finish with what we still don't know and
which assumptions from this list were supported or challenged:
{{assumptions}}
13. Mine reviews and support tickets
Unprompted feedback is often more honest than an interview.
Here are {{source:app store reviews and support tickets}} about
{{product:our product}} or competitors:
{{feedback}}
Pull out: the jobs people are trying to get done, the problems they hit,
the words they use for them, and what they compare us to. For each, give a
count and 2 exact quotes. Keep praise and complaints in separate lists.
Don't summarize away specific details like prices, features or workflows
customers mention.
14. Findings to decisions
Closes the loop with the decision from prompt 1.
Our research question was: {{decision:whether to build team billing this quarter}}.
Here are the findings with their evidence:
{{findings}}
Write a one-page readout: the recommendation in one sentence, the 3
findings that matter most for it with evidence counts and one quote each,
what would change the recommendation, risks, and next steps. State the
confidence level and why. Don't overstate findings that rest on fewer than
{{min_support:3}} participants.
How do you keep customer research honest?
Interview real people. Nielsen Norman Group's guide to user interviews says to "Treat user interviews like a research study, not an informal chat," and warns that "If the interviewer asks many leading questions, the validity of the data will be compromised." A model can check your questions for that, but it cannot replace the conversation.
Keep behavior and opinion apart. What a customer did last month is evidence. What they say they would pay for is a hint. Prompts 11 and 12 label the two separately so a wish doesn't get reported as a fact.
Ask for counts and quotes. A finding that says "most customers" with no number and no quote is a summary of the model's impression. OpenAI's prompt guide suggests asking the model to "list any assumptions you made" before finishing, which surfaces leaps you would otherwise miss.
Protect participants. Remove names and identifying details before pasting, and only use transcripts your consent form covers.
Keep your research prompts in one library
If you run the same interview or synthesis prompt every study, turn it into a skill with your rules built in, so Claude, Claude Code or Codex applies them every time. Paste it into the free SKILL.md generator. When research turns into a release, the product launch prompts pick up from there.
To have these ready for the next study, save them in Kipt with their blanks, tag them by stage (plan, recruit, interview, synthesize) and press Cmd+K to find one. Your library lives as Markdown in a private GitHub repo you own.