Learn from visitor conversations.

Founders usually hear from friendly customers, sales calls, and support tickets. That leaves a gap: evaluators, skeptics, confused visitors, and people comparing alternatives before they are ready to commit.

Their questions can reveal gaps your existing customers no longer notice. Record what visitors actually say, while remembering that people who choose to chat are only one part of your audience.

The Product Discovery skill should not ask leading roadmap questions. It should ask about the last time the problem happened, what the visitor tried, who was involved, what broke, what it cost, and what happens if nothing changes.

The Stand-in should stop when the visitor becomes rushed, frustrated, terse, or asks for direct help. The goal is learning with consent, not extracting a survey.

A discovery question is only useful if the answer could change something: the headline, the pricing explanation, the docs path, the onboarding promise, the roadmap priority, or who the founder interviews next.

Use these questions only after the visitor has agreed to a research interview. A factual pricing or documentation question still deserves a direct answer, even when the visitor matches your research audience.

Do not ask the Stand-in to learn everything. Give it the current questions that would actually change your next product or positioning decision.

Open Stand-ins → the chosen Stand-in → Edit → Skills → Product Discovery and enable it. Define whom to interview, the learning goal, and up to three research questions. These questions guide the interview; they are not a script the visitor must complete.

Start with Adaptive depth, which follows the visitor’s engagement. Use Light for brief exchanges or Deep for willing participants with a complex story. Set guardrails for subjects to avoid and when to stop; enable notifications only if you can act on an interview in progress. Save, then test an engaged visitor, a refusal, and a request to return to support.

To receive synthesized Product Discovery findings, open Digest on Pro or Business, enable Add AI insights to my digest, and save. Choose a cadence or send yourself the last seven days. The report draws on relevant conversations available in that digest period and scope; an empty period cannot produce research evidence.

Read the numbered evidence and caveats, then open the cited conversations. Check that the visitor actually described the problem, distinguish their words from the generated interpretation, and look for counterexamples. Several similar confusion reports may justify a copy test; one enthusiastic feature request is a reason for further research, not a roadmap commitment.