Playbook · 2026

How to design customer research studies with Claude: step-by-step.

Good research starts with the right questions, asked without bias. This playbook gives you a prompt template that turns a decision into a sound study design.

Short version

Tell Claude the decision you need to inform, who you can talk to, and your constraints, then ask for objectives, the right method, and an unbiased question guide. You get a defensible research plan, fast.

The prompt template

This template starts from the decision (not the questions), which is what keeps research useful, then drafts a non-leading interview or survey guide.

Copy, paste, and fill in the brackets

You are my UX and customer research partner. Design a research study. Context: - Decision this research will inform: [the choice you face] - What we think we know / assumptions: [hypotheses] - Who we can reach: [segment, sample size, access] - Constraints: [time, budget, qual vs quant] Produce: - 3-5 clear research objectives tied to the decision. - The recommended method and why (interviews, survey, usability test). - A question guide with NON-LEADING, open questions. - What a good vs misleading result would look like. Rules: flag any leading questions and rewrite them. Do not use em dashes.

Designing from the decision and screening for leading questions is what separates research that informs from research that confirms your bias. For applying findings, see how to use AI in your business.

The step-by-step workflow

Set up a Claude Project. Add your the decision at hand, your assumptions, and prior research as project knowledge so you never re-paste context. Claude Projects keep brand voice, examples, and rules in one place.
Paste the prompt template. Fill in the bracketed fields with your specifics. The more precise the inputs, the less editing the output needs.
Generate two or three variations. Ask for the plan, then a pass that hunts for leading questions. Pick the strongest and tell Claude what you liked so the next pass sharpens it.
Iterate, do not accept the first draft. One follow-up instruction (tighter, warmer, shorter, more specific) usually does more than re-prompting from scratch.
Edit for voice and accuracy, then save the prompt. Claude gets you most of the way; you own the final 20 percent. Save the working prompt so next time is a two-minute job.

A worked example

Say you are deciding whether to build a feature. Tell Claude the decision and your assumptions, and it returns objectives, recommends interviews over a survey, drafts open and non-leading questions, and flags any that lead the witness, giving you a study that tests your assumption rather than confirming it.

What to avoid

Vague prompts. 'Help me write this' produces generic output. Specifics in, specifics out.
Publishing or sharing without editing. Always do the final human pass for voice, facts, and judgment.
Skipping the Project setup, which forces you to re-paste context every single time.
Treating Claude as a vending machine instead of a thinking partner you iterate with.

Frequently asked questions

Can Claude design a customer research study?

Yes. Give it the decision, your assumptions, and constraints, and it produces objectives, a recommended method, and a non-leading question guide. Use the template here.

How does AI help avoid biased research questions?

Ask Claude to flag and rewrite leading questions. It is good at spotting questions that presume an answer, which is one of the most common research mistakes.

Should research start with questions or the decision?

The decision. Prompt Claude to tie every objective to the choice you face, so you gather what you will actually act on, not interesting trivia.

Can Claude analyze the results too?

Yes, once you have transcripts or survey data. Paste them and ask for themes and counter-evidence. See analyzing call recordings with Claude for the interview side.