Skip to content
Skills from A Smart Bear

Interview Questions: Test your hypotheses without leading the witness

Turn each hypothesis into an open-ended, unbiased interview question — a miniature experiment that uncovers what customers actually think, not what you hope to hear. Adapted from Chapter 10 of Hidden Multipliers: A multiplier of one's own.

Input
A HYPOTHESES.md file — or a single hypothesis you want to test in an interview.
Output
A QUESTIONS.md of numbered questions in interview order, each mapped to the hypotheses it tests (single-hypothesis mode: the final question, in chat).

What this is about

Ask a customer "Blogs get hacked all the time and it's devastating, right? Would you like extra security?" and everyone says yes. They would look foolish otherwise. You led the witness to the answer you wanted. Now you will never know the answer you needed. Most customer interviews fail here. The problem is not who you talk to. The problem is questions that hint at the wanted answer and collect polite agreement instead of real validation.

This is the third step of the interview method. The method runs goals → hypotheses → questions. Every question links to a hypothesis. This makes writing good questions easier, not guesswork. Each question is a small experiment. The book sets four criteria for each question. The question must be able to confirm or deny the hypothesis. The question must not hint at any particular answer. The question must draw out specific facts, not feelings. The question must invite information you did not know to ask for. The skill drafts candidate questions. It tests every version, including yours, with a polite-stranger test. It phrases each question in your customers' own words. It will not finalize a question that breaks the criteria. A leading question does not just weaken an interview. It creates false evidence. The source allows one exception: near the end of an interview, you may state a specific price. The customer's gut reaction to a real number is itself useful data.

Give the skill your whole HYPOTHESES.md file and it works through the list. It groups related hypotheses so the set fits a real conversation. It orders the result like an interview plan. It keeps QUESTIONS.md updated from the first finished question, so a long session, or one you resume later, never loses work. Or give it a single hypothesis. The skill then shapes that one question with you and gives it back in chat. No files, no extra steps.

Example invocation

You can invoke the skill like this:

/asb-interview-questions Here's my HYPOTHESES.md from the hypotheses
exercise — turn these into the questions I'll actually ask.

The session works through your hypotheses one at a time. It drafts and tests each question until the question is truly open-ended. Then it writes QUESTIONS.md as a ready-to-use interview plan.

Or bring one hypothesis with no file at all:

/asb-interview-questions My hypothesis: gym owners spend at least 2
hours a week chasing failed membership payments. How do I ask about
this without planting the answer?

In single-question mode, the skill shapes one phrasing with you and hands back the final question in chat.

One practical note: say where you are working. Name a directory, or point at your existing files, when you invoke the skill. It then keeps the method's files together there. If you do not say where, it asks before it creates anything.

From the source

This skill rests on two sources:

  • As in Chapter 10 of Hidden Multipliers: A multiplier of one's own (section "Questions: What to Ask During the Interview") — hiddenmultipliers.com — the questions step itself: the four criteria, the leading-the-witness failure and its security-question example, and the bad-versus-good question table this skill's testing is built on.
  • The Iterative-Hypothesis customer development method — the published article version of the method. It covers the price-quoting exception, the value-testing question pattern, and the follow-up probes for surprising answers.

Supporting articles each supply one part of the method: