decision-doc is a user-prompt template you paste into Claude/ChatGPT along with raw notes about a decision your team needs to make. It produces a one-page memo in the format every senior team eventually converges on: context, options with trade-offs, a clear recommendation, the risks, and the sign-off list.
Designed to replace the 40-message Slack thread that ends without a decision.
At a glance
| Field | Value |
|---|---|
| Type | User prompt (with optional system prompt) |
| Use case | Turn raw notes about a decision into a structured memo |
| Output | One-page markdown, ~400-600 words |
| Replace before use | Two placeholders: {{TOPIC}} and {{NOTES}} |
| Length | ~180 words of prompt + your notes |
The prompt
I need to write a decision memo for: {{TOPIC}}
Below are my raw notes (Slack threads, scratch ideas, partial
arguments). Turn them into a structured one-page memo using
this exact section structure:
# Decision: {{TOPIC}}
**Status:** Draft / In review / Decided
**Owner:** [name — fill from notes if mentioned, else "TBD"]
**Date:** [today's date]
## Context (3-5 sentences)
What's true today, why we're deciding now, what changes if we
don't decide. No history lessons — only what's needed to follow
the rest of the memo.
## Options
For each viable option, one paragraph:
### Option A: [name]
**What it is.** One sentence.
**Pros.** 2-3 bullets, concrete.
**Cons.** 2-3 bullets, concrete.
**Cost.** Money / time / opportunity. One line.
(Repeat for B, C. Cap at three options. If the notes mention
more, fold the weaker ones into a "Other options considered
and rejected" line at the bottom of the section.)
## Recommendation
One option, named. One paragraph explaining why it wins on the
criteria above. If the recommendation is conditional ("Option A
if budget is approved, else B"), state the condition explicitly.
## Risks and what we'd watch
2-4 bullets. The realistic things that could go wrong with the
recommended path, and the early signal we'd watch for each.
## Who needs to sign off
List by role. Format: "Role — for what specifically".
---
Rules:
- Don't invent facts not in the notes. Use "TBD: [specific
question]" for anything missing.
- Don't soften pros and cons to seem balanced. If Option A is
obviously better, say so.
- Don't recommend "we need more information" as the answer.
If the recommendation is "decide later, here's the smaller
decision we can make now," say THAT.
My notes:
{{NOTES}}How to use
- Replace
{{TOPIC}}with the actual decision (e.g., “Database for the new analytics service”). - Paste your raw notes after
{{NOTES}}— Slack messages, half-thoughts, “Sarah said X but Dan disagreed”, whatever. - Send the whole thing as a user message.
- Review the memo. The most common edit is the Recommendation paragraph — make sure it actually picks something.
- Ship to the relevant Slack/Notion/email.
Worked example
Topic: “Self-host or use a managed service for the new search index”
Notes (excerpt):
Sarah: managed (Algolia/MeiliCloud) is fast to ship but expensive at scale, ~$2k/mo at our volume
Dan: self-hosted Meilisearch on existing k8s, marginal infra cost, but someone owns ops
Tania: search relevance tuning matters more than infra — both options support it
Open question: do we need multi-region search? Sarah says no for v1
Decision needs to land by Friday so the search redesign isn't blockedOutput: A one-page memo with three options (Algolia, MeiliCloud, self-host Meilisearch), pros/cons/cost for each, a recommendation of “self-host Meilisearch, revisit at 10x volume”, risks (“ops headcount required”, “no multi-region story”), and a sign-off list (Sarah for cost, Dan for ops capacity, Tania for relevance).
Customising
- Different memo style. Replace the section structure with whatever your team uses (RFC, ADR, “DACI”, etc.). The rules at the bottom still apply.
- Pair with brutal-editor. Run the output through brutal-editor for a tighter pass before sending.
- For one-way-door decisions. Add a section after Risks: “## What’s reversible” — bullet what’s still movable after we commit, what isn’t.
- Hebrew memos. Add: “Output in Hebrew” at the top of the rules block.
The most useful side effect. Forcing yourself to dump notes into the {{NOTES}} block usually clarifies your own thinking before the model even runs. Half the value is the structure, not the LLM call.
Related
More drop-in prompts in the Prompts library. For the system slot patterns this prompt assumes, see System vs User Prompts.