When to use this prompt
When you have a folder of notes, half-formed observations, or a swipe file of links from the past two weeks and you need to ship a newsletter on Friday. The most common newsletter failure is trying to cover everything in the notes — the result is a list, not a piece. Lists get unsubscribed; pieces get forwarded.
This prompt forces a single idea per issue, with the rest of the notes either supporting that idea or getting dropped for next week. It’s also useful when you write a newsletter on a recurring beat (GEO, AI search, marketing operations) and need a way to convert raw inputs into shippable copy.
The prompt
<role>Newsletter editor specializing in single-idea issues that get forwarded.</role>
<task>Draft a newsletter issue from the rough notes below. Pick one main idea. Build the issue around it. Drop or sideline notes that don't support that idea.</task>
<inputs>
<newsletter_voice>[2-3 sentences describing the newsletter's voice and audience, e.g., "Direct, opinionated, written for B2B marketers who want to stay ahead, not catch up. No marketing fluff."]</newsletter_voice>
<typical_length>[Word target, e.g., "600-900 words" — newsletters that hit a consistent length build trust]</typical_length>
<sender_name>[NAME]</sender_name>
<rough_notes>
[5 to 15 bullets of raw notes. Mix of observations, links, half-takes, and data points. Don't pre-curate; the prompt does that.]
</rough_notes>
<recent_issues>
[OPTIONAL: 1-2 sentence summaries of the last 2-3 issues so the model doesn't repeat themes.]
</recent_issues>
</inputs>
<instructions>
1. Read all the notes. Pick one as the central idea — the one that, if developed, would create the strongest single-issue piece. Briefly justify the pick.
2. Identify which other notes (if any) support the central idea. Use only those. Drop or sideline the rest. List the dropped ones at the end so the user can save them for next week.
3. Draft the issue with this structure:
- **Subject line** (≤55 characters): names the specific tension or claim, not the topic. Not "AI search update" — "Why your dashboard hasn't noticed pipeline is breaking."
- **Opening** (1-2 sentences): a hook the reader can't stop reading. Specific, not a setup.
- **Body** (3 to 5 short sections): the central idea developed with evidence. Each section has its own one-line lead.
- **Closing** (2-3 sentences): the takeaway the reader can act on this week. Not "rethink your approach" — a specific action.
- **PS** (optional, 1 sentence): a quick mention of something else interesting from the notes, framed as "also seen this week."
4. Constraints:
- Total length within 10% of <typical_length>.
- Match <newsletter_voice>. Do not flatten an opinionated voice into generic prose.
- Do not invent statistics, quotes, or named companies that are not in <rough_notes>. If a fact is unverified, mark [VERIFY].
- Do not repeat themes from <recent_issues>.
5. Sign off with the sender_name. No "Best regards" or generic closers.
</instructions>
<output_format>
**Central idea picked:** [one sentence]
**Why this idea (not the others):** [one sentence justification]
---
**Subject line:** [≤55 characters]
**Body:**
[The full newsletter issue, ready to paste]
— [sender_name]
---
**Notes saved for next week:**
- [Note 1]
- [Note 2]
- ...
**Word count:** XXX
**[VERIFY] markers:** [List any unverified claims so the user can fact-check]
</output_format>How it works
The single-idea constraint is the entire game. Newsletters that try to cover everything dilute every individual point. Newsletters built around one idea, supported by 2-3 specific examples, get read and forwarded. Forcing the model to pick one idea and explain the pick prevents the “list of links” failure mode.
The “notes saved for next week” output is what makes this prompt sustainable. Without it, dropped notes either get rewritten into the issue (defeating the single-idea constraint) or vanish (wasting future-issue material). The save list keeps the discipline humane.
The subject-line constraint deserves special attention. Subject lines that name the topic (“AI search update”) underperform subject lines that name the tension (“Your dashboard hasn’t noticed yet, but pipeline is breaking”). The instruction to name the tension, not the topic, applies frontier-model literal-instruction following to the highest-leverage line in the issue.
The [VERIFY] markers are the same hallucination guard used across the library. Newsletter writers cite numbers; models invent numbers; the marker forces honesty about which numbers are sourced and which need a check.
Example output
Central idea picked: AI engines are rewarding brands for content already on their site, not new content. Why this idea (not the others): Five of the rough notes touch this directly; the others are tangential. Stronger single-idea issue than any of the alternatives.
Subject line: The content you already have is the audit you haven’t done
Body:
Most teams reading this are about to spend Q3 producing more content for AI search. They will be surprised when their citation share doesn’t move.
The reason: AI engines reward content you already have. They don’t reward more.
[Issue continues for 700 words covering 3 specific examples, the audit framework, and a closing action: “Run a 50-prompt audit on your existing top 20 pages this week. Don’t write anything new until you do.”]
— Arthur
Notes saved for next week:
- The Reddit thread I found about agentic AI in B2B
- The data point about Google AI Overviews coverage hitting 70%
- The take on whether Claude Opus 4.7 changes prompt engineering practice
Word count: 731 [VERIFY] markers: None this issue.
Variations
- Curation-first mode: For link-heavy newsletters where the format is a curated list. Constrain the body to 5 links with a 2-3 sentence editorial note per link.
- Series mode: When the issue is part of a multi-issue series, add an instruction to reference the prior issue at the top and tee up the next at the bottom.
- Listener question mode: For newsletters that take reader questions, supply the question as part of the inputs and ask the model to build the issue around answering it specifically.