Using Claude for customer discovery: what works and what makes it worse

Customer discovery is the most important thing an early-stage founder does, and it is the area where AI use can go most wrong. The failure mode is not using Claude for the wrong task — it is using it in a way that gives you false confidence in insights you have not actually earned.

This is the version that works: Claude for preparation and synthesis, humans for the actual conversations.

What Claude is good for in discovery

Preparing for interviews. The quality of a customer interview depends almost entirely on the questions you ask. Generic questions get generic answers. Specific, hypothesis-driven questions get the information you actually need.

Before each interview, spend ten minutes in Claude:

"I'm interviewing a [role] at a [company type] tomorrow. My current hypothesis is [hypothesis]. I think they have [problem]. Help me generate 10 questions that would either confirm or kill this hypothesis. Make them open-ended. Avoid leading questions. Start with their current behavior, not my product."

The resulting questions are better than what most founders prepare on their own, because Claude will push you toward behavioral questions ("Walk me through the last time you did X") over opinion questions ("Would you use a product that did Y?"). Behavioral questions get real data. Opinion questions get polite speculation.

Synthesizing across interviews. After five interviews, patterns start emerging but are hard to articulate. You have pages of notes, a handful of quotes, and a general sense of what you heard — but translating that into a clear hypothesis requires real synthesis work.

Paste your raw notes into Claude (redact anything sensitive):

"Here are my notes from five customer discovery interviews. What patterns do you see? What do multiple people mention that I might have dismissed as one-off? What seems like a strong signal vs. noise? What question should I have been asking that I wasn't?"

Claude will surface connections across interviews that are hard to see when you're too close to the individual conversations. The synthesis is not a substitute for your own analysis — it is a starting point that saves you two hours of staring at sticky notes.

Writing the synthesis doc. Most founders keep their discovery insights in their head or in scattered notes. Writing a clear synthesis document — what you learned, what you believe, what you are still uncertain about — forces clarity and creates something you can share with co-founders, advisors, or early investors.

Claude is genuinely useful here: you dump your insights in rough form, it organizes them into a readable document. The thinking is yours. The formatting is Claude's.

The interview simulation (use carefully)

You can ask Claude to roleplay as a customer and practice your interview script. This has real value — it surfaces questions that are confusing, leading, or too narrow before you burn an actual interview slot on them.

"Roleplay as a senior CS manager at a 100-person SaaS company. I'm going to practice my customer discovery interview with you. Push back when my questions are leading or too vague. Stay in character."

The critical caveat: Claude's simulated responses are drawn from its training data about how people generally behave, not from the actual person you are trying to understand. It will validate plausible things and miss unusual things. Use it to stress-test your questions, not to draw conclusions about your customer.

The trap: founders who run five simulated interviews and feel like they have done discovery. They have not. Simulated interviews are warmup, not research.

What Claude cannot do

Tell you whether your hypothesis is right. Only real customers can do that. If you describe your hypothesis to Claude and ask if it sounds valid, you will get a thoughtful analysis of whether it seems plausible — not evidence that it is true.

Replace the unscripted parts of an interview. The most valuable moments in customer discovery are usually the tangents — the thing the person mentions offhand that was not in your question list, the moment of visible frustration when you ask about a workaround, the pause before they answer a question in a way that tells you they're being careful. These only happen in real conversations.

Catch what you are missing in your mental model. Claude synthesizes what you give it. If you went into five interviews with a blind spot — a whole category of problem you were not asking about — Claude will not invent it. It will organize the blind spot you already have. The cure for this is diversity of interview subjects and active effort to disconfirm your assumptions, not more Claude.

If your discovery is informal (20-minute calls, not structured interviews)

Most discovery articles — including this one — describe formal structured interviews. Many early-stage founders do not run formal discovery. They have 20-minute conversations with potential users, catch-up calls that turn into discovery, or Slack DMs where someone describes their problem. This is still discovery.

Claude is useful here too, just differently:

After an informal conversation: Write three sentences about what you heard immediately after the call. Not polished notes — just what you remember. Then ask Claude: "Here is what I heard in a 20-minute call with [person's role]: [three sentences]. What hypothesis is this consistent with? What would I need to hear from the next 2-3 conversations to confirm or kill it?"

This forces you to extract a signal from a conversation that might otherwise stay in your head as a vague feeling.

When you have talked to 5+ people informally: Paste your rough notes from all of them into Claude and ask: "I've had informal conversations with five potential customers. Here are my rough notes from each: [notes]. What patterns do you see across these? What is the same complaint in different words? What do people keep mentioning that I might be treating as background noise?"

The synthesis from informal notes is often more valuable than from formal structured data, because informal conversations surface the things people actually say rather than the things they say when they know they are being interviewed.

The discipline that makes it work

Use Claude between interviews, not during them. During the interview: phone down, notes brief, full attention on the person. After the interview: immediately write rough notes while the memory is fresh (ten minutes, no polish), then use Claude to help you think through what you heard.

The founders who extract the most from customer discovery do the hard part — the actual conversations, the uncomfortable silences, the follow-up probes — and use Claude to make the surrounding work faster. The founders who use Claude to shortcut the conversations end up with organized notes about things they never actually learned.

Further reading

• Claude takes research to new places — using Deep Research for discovery interviews