PromptDeck
Product ManagerCustomer Feedback Synthesis#PM#JTBD#synthesis

Feedback Synthesis · PM JTBD Lens

Synthesize 50 feedback items into a PM-actionable brief: JTBD validation, hypothesis challenges, roadmap shifts.

Pain addressed · PMs jam feedback into the backlog without synthesising — the backlog grows, real signal drowns in noise.

Feedback Synthesis · PM JTBD Lens

Prompt·Advanced
Also available in 中文 →
You're my PM feedback synthesis partner. Use JTBD lens to compress 50 items into a PM brief.

# Raw feedback
{paste_feedback}

# Product + this quarter's bets
{product_and_bets}

# Output five sections

## 1. JTBD validation (3-5 sections, 50 words each)
For each prior hypothesis about user job [X], which did the feedback validate / challenge / overturn?
Each: original hypothesis / what feedback shows / verdict.

## 2. 3 high-confidence problems (80 words each)
Based on frequency × intensity:
- User job (in user voice)
- How they currently hack it (often using our product weirdly)
- Build / buy / kill response

## 3. 2 hypothesis challenges (60 words each)
Are there signals in the feedback that challenge this quarter's bets? Be willing to say so:
- We bet X · feedback says Y · should we pivot?

## 4. 1 "didn't see this coming" insight (60 words)
- The most surprising item / cluster
- What it means · where our prior mental model was wrong

## 5. 3 roadmap actions (30 words each)
- One ADD (from a high-confidence problem)
- One CUT (something feedback says nobody uses)
- One INVESTIGATE (something challenging a bet, needs more data)

# Boundaries
- Don't treat every item as must-fix. Synthesis's job is to find signal and cut noise.
- The "didn't see this coming" must be real. Missing = the model didn't synthesize.
- Every action must be launchable within the quarter.
Saves ~120 min per use

Open directly on your favourite AI

One click copies the prompt and opens the platform in a new tab — just paste (⌘/Ctrl + V).

Workflow · 5 steps

  1. 1

    Pull the last 4-8 weeks of tickets, interview notes, NPS comments, and sales relays — 50-80 items into one block.

  2. 2

    Run in Claude (strongest at JTBD reasoning) or DeepSeek R1. Fill the 2 placeholders.

  3. 3

    Audit the JTBD validation section. If the model says 'validated' for everything, it's hedging — push back for at least one 'overturned'.

  4. 4

    The hypothesis challenge section takes nerve. After reading, ask: 'am I willing to share this with the team?' If yes, send.

  5. 5

    Bring the 3 roadmap actions to the next PM standup. Start with INVESTIGATE — cheapest action, highest information value.

Verify before you ship

AI's "almost right" is the #1 pain. Tick these off so reviewers can't find the obvious holes. (Checkbox state stays local — we don't track it.)

Done 0 / 5 checks

Why this prompt works

This prompt isn't a generic “write me a Customer Feedback Synthesis”. It folds in the concrete context a Product Manager actually faces (target reader, info density, deliverable format). LLMs deliver an order-of-magnitude better answer when they get role × scenario × constraints together — they can make the right tradeoffs about what to spotlight and what to cut.

When to use

Best moment: you know roughly that you need to do a Customer Feedback Synthesis, but you're stuck on “how do I even ask the AI?” Muscle memory has no template. Copy this card, tweak 2-3 fields per the workflow, and get a first revisable draft in 30 seconds.

Common pitfalls

  • Send without tweaking placeholders · the draft reads as default-GPT and your reviewer spots it in 3 seconds.
  • Skip the verify checklist · “almost right” failures usually hide in a thing you assumed the model handled but it didn't.
  • Stop after the first answer · a single follow-up like “make section X tighter / more formal” usually pushes a 70 → 90.

Variants & extensions

Want to use this card more aggressively? Three extensions:

  • Swap languages: both EN/ZH prompts ship in the card — use the EN version when shipping to a Western audience.
  • Swap platforms: the same prompt has slightly different taste on ChatGPT / Claude / Kimi. The recommended-platform pill marks the best fit.
  • Stack the checklist: paste the verify checklist back into the chat and ask the AI to self-audit — error rate drops sharply.
PromptDeck Pro Bundle

Want 25 more prompts like this one?

Get the full 50-card Pro deck — multi-step chains, vertical variants, lifetime updates.

  • 50 prompts · 25 free + 25 advanced multi-step chains
  • PDF printable + Notion template
  • Lifetime updates · ~5 new prompts a month, auto-synced
  • Dual-language · EN/CN + 5 China-exclusive scenarios
Get Pro · $9 lifetime14-day refund · no questions · powered by Lemon Squeezy

Related cards