TL;DR
Feed an AI your theme, three links, and your one-sentence take on each link. Ask for a 600-800 word issue built around your reactions, not the links. Edit for voice, cut to one CTA, and write the subject line last. AI structures and tightens; the opinion stays yours. Skip that and you ship a summary nobody re-opens.
The task
Your newsletter is due in 3 hours. You have a theme, three links, and a few raw thoughts. The risk is shipping a curated link list that reads like a feed: no take, no voice, no reason to open next week. The job is to turn the links into a piece of writing with your reaction at the centre, not the links.
Why it matters: the bar is rising. On Beehiiv’s 65,000-newsletter dataset, reported open rates climbed to roughly 41% in 2025, but most of that is noise. Apple’s Mail Privacy Protection preloads images and fires the tracking pixel even when nobody reads the email, inflating reported opens by 15-20+ points (some platform medians sit near 49%). Clicks are what survive that distortion, and click rates have slipped year over year as inboxes get more crowded. A healthy newsletter click rate in 2026 is about 2-5% (3-6% for media and creator-led sends, 1.5-2% for B2B). That is the number a take-driven issue moves; a link dump does not.
When AI helps, and when it does not
AI is good at structuring an issue, varying section lengths, and tightening a rambling first draft. It is bad at having a take. That part is yours. Feed it your raw reactions, not just the links; otherwise you get a competent summary that nobody re-opens.
A practical split as of June 2026: any current frontier model (GPT-5.5, Claude Sonnet 4.6, Gemini 3.1 Pro) handles the structuring well. Claude Sonnet 4.6 and Opus 4.7 tend to hold a pasted voice more faithfully across a full issue, which matters most here. Whichever you use, the model copies your opinion only if you give it one to copy.
What to feed the AI
- The theme of this issue (one sentence)
- 3-5 links with title + URL + your one-sentence take on each
- Audience (segment, why they subscribed, what they expect)
- Brand voice: paste a previous issue you are proud of (200-300 words)
- The CTA (drive to a paid tier, share, reply, click out, or none)
- Banned phrases, your AI tells (“dive in,” “let’s get into it”)
Copy-ready prompt
Write a newsletter issue.
Theme (one sentence): [line]
Audience: [segment, why they subscribed, what they expect]
Brand voice (previous issue I am proud of): "[paste 200-300 words]"
CTA: [paid / share / reply / click / none]
Banned phrases: [list]
Links + my take:
1. Title — URL — my take: [sentence]
2. ...
3. ...
Return a 600-800 word issue with:
- Hook (75-100 words) that sets up the theme
- Three sections, each: one-line link reference + 100-130 word
essay using my take as the spine. The link is the prompt, not
the body.
- A "what to do with this" callout: one practical action for the reader
- A 50-word closer with the CTA
Constraints:
- No "let's dive in," no engagement bait, no emoji headers
- Vary sentence length, alternating short and long
- Do not summarise the linked article in detail; assume the
reader trusts me to pick
- Keep my take recognisable as mine; flag any paragraph that
drifted into "AI summary" voiceFor a paid newsletter, add: “Add a 100-word teaser of next issue’s topic, positioned to make a free reader curious.”
Subject line and preview text
Write these last, after the issue exists, so they reflect the strongest idea you actually wrote. As of June 2026 the practical targets are:
- Subject line: 30-50 characters. Gmail on Android truncates around 33, so put the hook in the first ~33 characters.
- Preview (preheader) text: roughly 40-90 characters on mobile; use it to add context, not repeat the subject.
Have the AI generate 5-8 subject options from the finished draft, then pick the one that reads like a person, not a campaign. Subject lines that look automated get ignored even when the writing is good.
Recommended output structure
Hook → 3 sections → action callout → closer with CTA. Subject + preview line at the top. Total 600-800 words. Separate sections with a single line break, not heading hierarchy, because newsletter clients render heading tags inconsistently.
How to check the output is usable
- Each section reads as a take, not a summary
- Your voice from the reference is recognisable
- Sentence length varies (no flat “AI cadence”)
- The CTA is one ask, not three
- A reader could skip every link and still get value from your take
- The subject line says something specific, not “This week’s roundup”
Picking a platform
The right home depends on whether you charge and how big your list is. Figures below are current as of June 2026, verify the live page before you commit.
| Platform | Free tier | Paid entry | Revenue cut on paid subs |
|---|---|---|---|
| Substack | Unlimited subscribers, free forever | None (free to publish) | 10% + Stripe fees |
| Beehiiv | Up to 2,500 subscribers | Scale ~$49/mo (to 10,000 subs, ad network, referrals) | 0% |
| Kit (formerly ConvertKit) | Up to 10,000 subscribers, one automation | Creator from ~$39/mo (to 1,000 subs) | 0% |
Substack is the lowest-friction start and takes a 10% cut once you charge. Beehiiv and Kit take 0% of subscription revenue but charge a monthly fee that scales with list size, which is cheaper at scale if you have paid subscribers. None of these write the take for you; that is the part this prompt is for.
Common mistakes
- Long link list, thin take. Readers can get a link list from Reddit.
- Stacking CTAs (subscribe + share + reply + click). Pick one.
- Letting the AI summarise the linked article. That defeats the link.
- Same-rhythm sentences, the giveaway of AI prose.
- Forgetting the subject line. It decides whether the work gets seen.
- Chasing the open-rate number. Track clicks and replies instead; opens are inflated by Apple MPP.
FAQ
- How long is too long? 600-800 words is the sweet spot for a curated issue; 1,200 is the upper bound for most newsletters. Longer content belongs behind a paid tier with its own format.
- Which metric should I actually watch? Click rate and reply rate. Apple Mail Privacy Protection inflates open rates by preloading the tracking pixel, so a 40%+ open rate can hide a dead list. A 2-5% click rate is a healthier signal in 2026.
- Should I include images? One hero image, optionally. Inline images frequently break across clients and add load time, so do not rely on them to carry meaning.
- How consistent does my schedule need to be? Same day, same time, weekly is the safe default. Open and click rates degrade fast on irregular schedules because subscribers stop expecting you.
- Can I let AI write the whole thing unattended? No. The structuring is safe to delegate; the opinion and the final edit are not. The reason anyone subscribes is your judgment, which the model cannot fake.
Related
- Newsletter prompts — alternative phrasing
- AI newsletter tutorial — end-to-end workflow
- Cross-platform repurpose — repurpose the newsletter
- Email marketing copy — promo emails, not newsletter
- Email subject line prompts — subject lines for the issue
- Storytelling outline — newsletter intro as a story
Tags: #AI writing #Content creation