Deep Dive (one topic, taken all the way down)

A focused counterpart to broad customer research: the user names one thing that's on their mind, and you take it all the way down into the evidence — returning not a plausible guess but a quote-rich, sourced brief drawn from real conversations. This skill is often reached right after the initial report, when a pattern caught the user's eye and they want to go deeper. The move is: ask what's top of mind, then ground the answer in what customers actually said.

Step 1 — Ask what's top of mind

Open by asking the user which topic is top-of-mind for them these days — the one thing they most want to understand right now. Keep it warm and singular; you're inviting one topic to go deep on, not a checklist. For example:

What's top of mind for you these days? Name the one topic you'd most like me to dig into — onboarding, discovery, churn, a specific feature, a segment — and I'll take it all the way down into what your customers have actually said.

Then stop and await their answer. Don't guess a topic and start searching — the whole point is to go deep on their topic. If the reply is too broad or genuinely ambiguous, ask one quick narrowing question (a decision it feeds, a segment, a time window) before grounding. Try to get at least a few relevant keywords so that search works correctly.

Step 2 — Ground the topic in the corpus (required)

Once you have the topic, follow using-evermuse skill's grounding mechanics to find the relevant evidence.

Step 3 — Deliver a cited brief

Cluster the evidence into themes, and for each: demand strength (mentions across distinct accounts), the sentiment split, who said it and when, and the sharpest verbatim quote. Represent dissenting voices — don't flatten disagreement into a false consensus. Keep evidence (what customers said) separate from any context (market) you pulled.

Lead with the answer, then themes with quote blocks:

**Short answer:** [1–2 sentences that directly answer the topic they raised.]

### Theme 1 — [name] ([N mentions / M accounts], mostly [sentiment])
> "[verbatim quote]" — [Name], [Meeting], [Date] [`1`](URL)
[one line of interpretation]

### Theme 2 — …

**Dissent / nuance:** [minority view, cited]

Honesty when evidence is thin

If the grounding greps come back sparse, say so — "only 2 mentions, both from one account" — rather than dressing a weak signal as a trend. Thin evidence is itself a finding, and on a deep dive it's often the cue to recommend an interview (find_skills:interview-script).

Hand off

Close by offering the natural next move: "Want a spec from this?" (find_skills:write-feature-spec), "Brainstorm solutions?" (find_skills:brainstorm-ideas), "Prep an interview to fill the gaps?" (find_skills:interview-script), or "See everything I can do?" (find_skills:list-capabilities).

Further reading

• Grounding mechanics: using-evermuse and its references/search-patterns.md + references/citations.md. Companion to customer-research (broad synthesis) — this skill is the single-topic deep cut.