It’s Thursday morning. Your newsletter goes out tomorrow. You don’t have a topic, much less a draft. This tutorial replaces the “stare at the blank doc until 11pm” cycle with a 60-minute Thursday-morning routine: 10 minutes of AI-assisted ideation, 25 minutes of drafting, 10 minutes of subject-line A/B prep, 15 minutes of editing and send. It’s aimed at creator newsletters, brand newsletters, and indie product newsletters where the bottleneck is consistency, not voice.
TL;DR
- Run the loop weekly: ideate (10 min) → outline (5 min) → draft (20 min) → subject A/B (10 min) → send (5 min) → log learnings (10 min).
- Let AI generate topic candidates and the first draft. Write the personal note and the CTA yourself — that’s where AI flattens your voice.
- Keep subject lines to 30–50 characters; only the first ~30 reliably survive Gmail’s mobile truncation, so front-load the hook.
- Pick the platform by the feature you actually need: Kit for automation (free to 10,000 subscribers as of June 2026), Beehiiv for analytics and A/B testing (free to 2,500), Substack for fastest setup (free, but 10% of paid revenue and no native subject-line A/B test).
- Judge results on clicks and replies, not raw “opens” — Apple Mail Privacy Protection inflates reported opens by 15–20+ points.
Who this is for
Creator newsletters (a writer’s personal list, an indie expert), brand newsletters (a SaaS company weekly, ecommerce news), and indie product newsletters (changelog plus tips). It works best when you have a defined audience and three themes you cover. It works worst when you don’t know your audience yet — AI cannot manufacture a voice for an audience you have not met.
It’s also useful for relaunching a dormant newsletter: the workflow’s discipline restarts the habit. Newsletter is usually one stage of a bigger funnel — see the AI content marketing stack guide for the full ideation-to-repurposing tool set that the workflow below plugs into.
When this is NOT the right tool
Skip AI on these:
- The first issue of a newsletter with no clear voice yet. Write it by hand, twice, before bringing in AI.
- Issues that are highly personal (a writer’s life updates). AI smooths the voice into a stranger’s.
- One-off announcements you occasionally drop into a newsletter. The setup overhead exceeds the value.
Pick your platform first (June 2026 pricing)
The article repeats prompts that work on any platform, but two early decisions — does your platform support subject-line A/B testing, and what does it cost as you grow — shape the rest of the workflow. Here are the four most common choices for solo and small-team newsletters, verified June 2026.
| Platform | Free tier | First paid step | Native subject A/B test? | Best at |
|---|---|---|---|---|
| Substack | Unlimited subscribers; 10% of paid-sub revenue + Stripe ~2.9% + $0.30 | No flat fee (revenue share only) | No | Fastest setup, built-in discovery |
| Beehiiv | Launch: up to 2,500 subscribers | Scale: ~$43/mo at 1,000 subscribers | Yes | Analytics, A/B testing, monetization |
| Kit (formerly ConvertKit) | Up to 10,000 subscribers | Creator: ~$39/mo at 1,000 subscribers | Yes | Automation and sequences |
| Mailchimp | No A/B testing on Free | Essentials: $13/mo at 500 contacts (2-variant A/B) | Paid only | Broad marketing-automation suite |
Two corrections worth flagging because old guides get them wrong: Substack has no native subject-line A/B test as of June 2026, and Mailchimp’s Free plan can’t A/B test subject lines — you need at least Essentials. If subject-line testing is core to your loop (it should be), Beehiiv and Kit are the safer picks. If you’d rather not pay until you monetize, Substack’s revenue-share model wins until your list is large.
For deeper platform-by-platform feature notes, the AI content marketing stack guide covers where each one fits.
Pick your model
Any current frontier model handles this workload; the difference is voice control on the draft step. As of June 2026:
- Claude Sonnet 4.6 (Claude Pro, $20/mo) — strongest at matching a pasted voice sample without over-formalizing. The default for the draft step here. Pro now bundles Claude Code and Cowork.
- GPT-5.5 (ChatGPT Plus, $20/mo; default model since ~April 2026) — fast ideation and subject-line variants; the Instant/Thinking/Pro picker lets you keep ideation cheap.
- Gemini 3.1 Pro (Google AI Pro, $19.99/mo) — useful if your newsletter draws on Google Docs/Sheets research; 1M-token context for pasting in long source material.
You don’t need a paid plan to start — free tiers of all three draft a 350-word issue fine. The paid tiers matter mainly for higher daily limits when you batch a month of issues at once.
Before you start
- Document your audience in one sentence (“indie SaaS founders running 10–50 person teams, mostly technical, US/EU”). AI mirrors what you tell it — vague audience, vague newsletter.
- List your three themes (“scaling without funding, AI tooling, founder mental health”). Every issue should visibly be one of these.
- Confirm your platform supports subject A/B testing (see the table above) and that you can set the test window.
- Build a “swipe file” of three past issues you were proud of. Paste these into the AI as voice references — it learns the voice faster from examples than from descriptions.
Step by step
1. Topic ideation (10 min)
Prompt:
Audience: [paste one-sentence audience].
Themes: [paste your three themes].
Last 3 issues covered: [list].
Generate 8 topic candidates for this week's issue — each as a one-line
title plus a one-line "why now." Avoid topics close to what I just covered.Pick one. If none are good, the prompt or the theme list is too thin, not the model. Save the winners to a running doc (see “How to reuse” below).
2. Outline the issue (5 min)
Standard structure: a one-line hook, three numbered takeaways, one personal note (your voice, your story), one CTA. Draft this yourself — the personal note carries the voice, and that’s the one part AI cannot do well.
3. Draft the body (20 min)
Prompt:
Write a 350-word newsletter issue from this outline: [paste outline].
Match the tone of this past issue: [paste a swipe-file issue].
No corporate-speak. No filler adjectives. Keep my numbered takeaways intact.Expect to rewrite the personal-note paragraph entirely. AI flattens voice on that part every time — paste your draft note in and ask the model only to tighten grammar, not reword it.
4. Subject lines (10 min)
Prompt:
Generate 10 subject-line variants for this issue: [paste issue].
Mix: curiosity gap, specific outcome, question, contrarian, personal.
Keep each to 30–50 characters. No emoji unless I ask.Why 30–50 characters: Gmail’s mobile app shows only about 30 characters before truncating, and over 60% of newsletter opens happen on mobile (as of 2026). Front-load the hook in the first 30 characters, and let the preheader text do the rest. Pick your top two variants for the A/B test. Personalized subject lines (a first name, a specific detail) measurably out-open generic ones — roughly 46% vs 35% in 2026 benchmark tests — so favor specificity over cleverness.
5. Send (5 min)
On Beehiiv, Kit, or Mailchimp Essentials+, A/B test the subject on a 10–20% sample, then send the winner to the rest. Set a 2–4 hour test window; longer tests waste the open window. On Substack, which has no native A/B test, alternate subject styles issue-to-issue and compare opens manually — slower, but it still surfaces patterns.
6. Log learnings (10 min, one week later)
Record: open rate, click rate, replies, unsubscribes, and which sections drove clicks (tag links with UTM parameters). Then feed it back into next week’s ideation prompt — for example: “Last week ‘X topic’ got 38% reported open rate and replies clustered on subtopic Y; explore Y this week.”
A note on the open-rate number: Apple Mail Privacy Protection pre-loads images and inflates reported opens by 15–20+ percentage points (as of 2026), so a “38% open rate” might be ~22% real human opens. Treat opens as a rough trend line and weight clicks and replies — which can’t be faked by image pre-loading — far more heavily.
First-run exercise
- Run the full workflow once for next week’s actual issue. Real stakes; don’t rehearse on a fake issue.
- Time each step. If ideation took 25 minutes instead of 10, your audience and themes aren’t specific enough.
- Save the prompts that worked. The ideation prompt is the most reused; tune it for a month before declaring it done.
- For the second issue, change only one variable: a tighter theme list, a longer voice reference, or a different model.
Quality check before you send
- The issue solves something the reader cares about today. Not “interesting” — useful. Re-read the takeaways: would the reader thank you for this on Monday?
- The personal note sounds like you, not AI. Read it aloud; if it sounds smooth and corporate, rewrite by hand.
- The CTA is one specific ask, not three vague invitations. “Reply with your hardest hiring decision this quarter” beats “let me know what you think.”
- The subject line is 30–50 characters, with the hook in the first 30; no emoji unless your audience expects them; no clickbait the body can’t deliver.
- All links work and carry UTM tags. Broken links destroy the trust you spent the issue building.
How to reuse this workflow
- Save the ideation, draft, and subject-line prompts as a Notion or text snippet. Each week, swap the audience and theme details and keep the structure. For ready-made versions, see newsletter prompts and email marketing prompts.
- Maintain a “winning topics” doc — issues with open rates above your baseline. Patterns emerge, and future ideation gets faster.
- Quarterly, audit unsubscribes by topic. Topics that consistently drive unsubscribes have leaked into the wrong audience.
Sending the issue in a second language? Run it through our AI translation self-critique workflow — the region-aware critique catches tone slips that flat translation would silently ship.
Common mistakes
- Spending 90 minutes on the body and 10 seconds on the subject line. The subject drives opens, the body drives replies; both matter, neither should be an afterthought.
- No CTA. Readers don’t know what’s next, and a clear ask is what earns the relationship.
- Not logging what worked. The same writer rediscovers the same insight every six months without a log.
- Letting AI write the personal note. It smooths the voice into a stranger’s, and unsubscribes follow.
- Stuffing every issue with five links. Subscribers leave when an issue feels like a digest. One link is enough.
- Chasing the reported open rate. Apple MPP inflation makes it noisy. Optimize clicks and replies instead.
FAQ
- Which platform is best for an AI newsletter workflow?: It depends on the feature you need. Kit (formerly ConvertKit) is free to 10,000 subscribers and best for automation; Beehiiv has the strongest analytics and native A/B testing (free to 2,500 subscribers); Substack is the fastest to set up but takes 10% of paid-subscription revenue and has no native subject-line A/B test (all as of June 2026).
- Which AI model should I draft with?: Claude Sonnet 4.6 matches a pasted voice sample most faithfully, which matters most on the draft step. GPT-5.5 and Gemini 3.1 Pro are both fine for ideation and subject lines. Free tiers of all three handle a 350-word issue.
- How long should an issue be?: 300–500 words for most niches. Tech-industry newsletters tolerate up to 800. The friction is at the open, not the read.
- What about images?: One header image, maybe one inline. AI-generated header images are fine; AI-generated stock photos of people are not.
- Should I trust my open rate?: Not at face value. Apple Mail Privacy Protection inflates reported opens by 15–20+ percentage points (as of 2026), so a 35–40% reported open rate often maps to ~20–28% real human opens. Weight clicks and replies more heavily.
- What if I miss a week?: Skip it. A wobbly schedule beats a forced bad issue. Resume next week — no apology paragraph needed.