synthesizing-user-research

name	synthesizing-user-research
description	Turning a stack of interview transcripts or research notes into themes, Jobs-To-Be-Done statements, and a recommended next step. Use when the user says "synthesize this research", "I have N interview transcripts", "what did we learn from the user interviews", "pull out the themes", or "what's the JTBD here".
allowed-tools	["Read","Write","Edit"]

Rule: Synthesizing user research

Goal

A research synthesis answers one question: what should we do differently because of what we heard? Everything else — themes, quotes, frameworks — only earns its place by serving that answer. The mistake most PMs make is producing a beautiful Notion page that nobody acts on. This skill ends with a recommendation, not a wall of color-coded sticky notes.

Process

1. Receive input — transcripts (raw or summarized), call notes, survey results, support tickets, sales-call snippets.
2. Ask the clarifying questions below. The research goal changes which framework fits.
3. Read everything once before synthesizing. Don't synthesize as you read.
4. Cluster observations into 3–7 themes. Fewer than 3 means you didn't push; more than 7 means you didn't cluster.
5. Write JTBD statements for each theme that has a clear user motivation.
6. Surface the surprises — what contradicted your prior beliefs, not what confirmed them.
7. Recommend a next step. One paragraph. Specific.

Clarifying Questions

Research goal — what were you trying to learn?

• 1. Problem discovery (do users have the pain we think they do?)
• 2. Solution validation (does this approach work?)
• 3. Usability (where do people get stuck in the flow?)
• 4. Pricing / willingness to pay
• 5. Churn / win-loss

Synthesis depth

• A. Themes only (a one-pager)
• B. Themes + JTBD statements
• C. Themes + JTBD + opportunity sizing
• D. Full opportunity-solution tree (Teresa Torres style)

Sample size + segmentation

• How many interviews? Roughly how were they segmented (new vs. churned, paid vs. free, role)?
• If under 5 interviews: say so explicitly in the output. Patterns from <5 are hypotheses, not findings.

Where the output goes

• A. Internal Notion / Linear doc
• B. Exec memo
• C. Input to a PRD
• D. Customer-facing report (rare; needs a tone shift)

Output Structure

# <Research topic> — synthesis

**Goal of the research.** <One sentence — what you were trying to learn.>
**What we did.** <N interviews / surveys / sessions, period, segment.>
**What we'd do differently because of this.** <One paragraph. The recommendation.>

## Top themes

### 1. <Theme name — a short noun phrase, not a sentence>
- What we heard: <Plain-language summary of the pattern.>
- Strength: <How many of the N participants showed this.  "/">
 Representative quote: > "" — 
 So what: 


…



  ,  ,  .
  ,  ,  .

(One per theme that has a clear underlying user motivation.)



 



| Opportunity | T-shirt | Confidence | Notes |
|---|---|---|---|
|  | S / M / L | low / med / high |  |



. Decision point: Apr 12 — go / no-go on shipping the redesign.">



 



 Participants: 
 Interview protocol: 
 Recruiter / script source: 

Voice rules

• Quotes are sacred — verbatim only. No paraphrasing inside quotation marks.
• Counts beat adjectives. "7 of 9" beats "most".
• Name the surprise. Synthesis without a surprise is a synthesis you didn't push hard enough on.
• Themes are nouns. "Pricing confusion" — not "Users were confused about pricing".
• Don't recommend the obvious. "Talk to more users" is not a recommendation.

Target Audience

The doc usually has two readers: an exec who reads only the top three paragraphs, and a teammate who reads it carefully and acts on it. Write so both get value: the top is dense and declarative; the body has enough detail that the doer can do.

Tiny worked example

Theme excerpt (from 9 onboarding interviews):

2. People skip the sample data step because they don't trust it'll be deleted later

• What we heard: Six of nine new accounts skipped the "load sample data" toggle. Three said they were worried about cleaning it up before going live.
• Strength: 6/9 skipped; 3/9 said it out loud.
• Representative quote: > "I don't want fake invoices floating around in my real account. I'll figure it out without the sample stuff." — P4, ops lead
• So what: We're hiding the product behind a trust problem we can fix with one sentence and a "delete sample data" button.

JTBD:

When I'm setting up the tool for my real team, I want to evaluate the workflow without polluting my live data, so I can adopt confidently without rework later.

Recommended next step:

Add a one-click "remove sample data" button on the dashboard and a "you can delete this anytime" line under the sample-data toggle. Ship within the sprint. If sample-data usage doesn't move 15 points in 30 days, the trust issue is deeper than copy and we run a longer test.

Anti-patterns

• Synthesis = quote highlight reel. Quotes support themes; they don't replace them.
• 20 themes that are really 4 themes with synonyms. Cluster harder.
• No recommendation. If you don't end with one, the work is incomplete.
• Confirming what you already believed. If nothing surprised you, you led the witness.
• Hiding tiny-N caveats. Five interviews is a hypothesis, not a finding. Say so.

Output

• Format: Markdown (.md)
• Location: research/ or tasks/research/
• Filename: <YYYY-MM-DD>-<topic>.md

Final instructions

1. Always read all transcripts before clustering. Synthesis-as-you-go biases toward the first interview.
2. End with the recommended next step. If you can't write one, the doc isn't done.
3. After delivering the draft, ask: "What was surprising — and what would change your mind?" Then revise.

Attribution

Synthesized from: anthropics/knowledge-work-plugins · user-research-synthesis (Apache 2.0); phuryn/pm-skills · summarize-interview + opportunity-solution-tree (MIT); Teresa Torres' Opportunity-Solution Tree framing from Continuous Discovery Habits; Tony Ulwick's Jobs-To-Be-Done formulation; voice principles from the Runtime brand skill.