How to use AI to write email newsletters that people actually open

Most email newsletters are written under pressure. They’re due Friday, the writer has been staring at a blank doc since Wednesday, and the final version reads exactly like that — rushed, generic, forgettable. We started using AI to draft newsletters about eighteen months ago, and the results were not immediately good. The drafts were too formal, too even-toned, and covered every angle so thoroughly that they said nothing in particular. The problem wasn’t the model. It was the brief. Once we fixed how we handed work to AI, open rates on client newsletters climbed an average of 22% over three months. This post explains the workflow — from subject line to sign-off — and the specific inputs that separate a newsletter people open from one they archive without reading. It’s written for marketing teams, agency operators, and founders who want to use Choco Media‘s approach to making AI email newsletter writing actually useful.

This is not a post about which AI tool to use. Claude, GPT-4o, and Gemini can all produce a serviceable first draft with the right brief. The workflow matters more than the model. What we’ve found is that most teams skip the brief entirely — they open a chat window, type “write a newsletter about X,” and then spend an hour editing out the corporate warmth and the unsolicited analogies. We’re going to show you a faster path.

By the end of this post you’ll have a brief template, a subject line testing approach, a section-by-section editing checklist, and a clear sense of where the human needs to stay in the loop. Let’s get into it.

Why AI struggles with email newsletters (and why it doesn’t have to)

Email newsletters are one of the hardest formats for AI to get right on the first pass. Unlike a blog post, where structure and completeness carry the day, a newsletter lives or dies on voice — the specific cadence, word choice, and implied personality that make a reader feel like a person wrote this for them. AI defaults to a kind of helpful neutrality that works fine for documentation and terrible for inbox.

The other issue is that newsletters reference context that AI doesn’t have: what happened this week, what the company shipped, what the audience is anxious about right now. Without that, the draft fills the shape of a newsletter without filling the meaning.

• Generic opening line problem — “We hope this finds you well” or “It’s been a big week in [industry]” are the first signs of an under-briefed draft.
• Tone collapse — AI tends to slide into either corporate press-release mode or over-friendly influencer mode. Neither is right for most brand voices.
• Missing the hook — newsletters need a reason to exist. AI often produces “here is a round-up of things” without an argument or a specific thing the reader should do with the information.
• No internal signal — without knowing what the company is working on, celebrating, or thinking about, the draft feels like it could have been sent by anyone.

All of these are brief problems, not model problems. Fix the brief, fix the output.

The 7-field brief that produces usable first drafts

We use a structured brief that takes about 15 minutes to fill in and produces a draft that needs 20–30 minutes of editing rather than a full rewrite. Here are the fields:

1. Audience and context

Who is reading this, what do they already know, and what is their emotional state right now? “Our B2B SaaS audience has heard a lot about AI cost-cutting this month and are nervous about headcount” is a brief. “Marketing professionals” is not.

2. The one thing

Every good newsletter has one argument, one story, or one thing the reader should think or do differently after reading it. Write it in one sentence. If you can’t, the newsletter isn’t ready to be written yet.

3. This week’s specific angle

What happened recently that makes this the right time to send this? A product update, an industry shift, a case study result, a conversation you had. This is the hardest field to fill in, and also the most important. It’s what makes the newsletter feel alive rather than evergreen.

4. Voice notes

Three to five adjectives describing the tone, plus one example sentence from a previous send that represents the voice at its best. This field alone lifts AI output quality significantly. For more on getting AI tools to respect your brand voice, our post on building a brand voice document AI can follow goes deeper on the documentation approach.

5. What to avoid

List three phrases or structural moves the newsletter should never include. Ours, for example, bans “in today’s fast-paced landscape,” any phrasing that implies the reader is behind, and any call-to-action that uses “just” as a minimiser (“just hit reply,” “just click here”).

6. Subject line direction

Not the final subject line — a direction. “Curiosity gap about email deliverability” or “Straightforward announcement of the new feature” gives the model a frame without over-constraining it. We’ll address subject lines in detail below.

7. Call-to-action

What is the one thing you want the reader to do? Be specific. “Read the blog post” is a CTA. “Forward this to a colleague who is planning a campaign launch in Q3” is a better one.

The brief is not a nice-to-have. It is the product. When a newsletter draft is bad, the brief was bad. We’ve stopped accepting “just write something about X” as a brief internally — it produces output that requires more editing time than writing from scratch would have taken.

Subject line testing: the AI-assisted approach

Subject lines are where AI earns its keep fastest. Given a brief, most models can generate 10–15 subject line variants in under a minute. The challenge is evaluating them intelligently rather than just picking the cleverest one.

We use a four-filter pass on subject line candidates:

• Specificity filter — eliminate any line that could apply to a newsletter from a different company or a different week. “What we’ve been thinking about” fails this test. “The case against a four-day content sprint (from someone who just finished one)” passes.
• Preview text alignment — the subject line and preview text work together. Evaluate pairs, not lines in isolation.
• List fatigue scan — if your last three subject lines used the same structure (e.g. all questions, all numbers, all “we did X”), break the pattern regardless of quality.
• Click-intent clarity — open rates measure curiosity; click-through rates measure relevance. Subject lines that create curiosity without signalling relevance inflate opens and tank clicks. Make sure the line sets accurate expectations about what’s inside.

For clients sending to lists over 5,000, we run a 20% A/B split between the top two subject lines before committing to the full send. For smaller lists, pick the one that scores best on the four filters and move on — statistical significance takes longer to achieve than the send cycle allows.

Structure: what a good AI-drafted newsletter looks like section by section

We use a consistent structure that the brief instructs the model to follow. Consistent structure means the human editor knows exactly where to look for problems, which makes the edit faster.

Opening paragraph (written by human, every time)

The first 40 words of a newsletter are the hardest and the most important. They are also the place where AI is most likely to produce something generic. We write the opening paragraph ourselves, always. Even two good opening sentences dropped into the AI draft significantly lift the quality of everything that follows because the model can calibrate to the established voice.

Body sections (AI-drafted, lightly edited)

Once the opening is set and the brief is solid, the body sections — whether that’s a how-to, a case study summary, or a list of observations — are where AI saves the most time. The edit pass here is for: factual accuracy, brand voice consistency, any claim that needs a source, and any sentence that falls into the banned phrase list.

CTA and sign-off (template with variables)

We keep a rotating bank of CTA formats and sign-off lines. The AI fills in the variables; the human checks that the tone matches the newsletter’s emotional register. A newsletter about a difficult industry topic shouldn’t close with an upbeat “have a great week.” It sounds like the writer didn’t read what they wrote.

The human-edit pass: what to check in 20 minutes

A structured edit pass is faster than an open-ended read. We work through these checks in order:

• Voice check (2 minutes) — read the first sentence of each paragraph aloud. Does it sound like the brand or like a helpful assistant? Flag anything that sounds like a summary rather than a statement.
• Fact check (5 minutes) — anything statistical, anything attributing a quote, anything claiming a specific outcome. AI hallucinates plausible-sounding data. Every number needs a source or needs to be removed.
• Redundancy pass (3 minutes) — AI drafts often restate the same point in the opening paragraph, the body, and the conclusion. Cut anything that’s already been said. Most newsletters are 20% longer than they need to be.
• Link check (2 minutes) — verify every link resolves correctly and the anchor text accurately describes the destination. On newsletter links, tracking parameters matter; make sure your UTMs are appended before the draft goes to the sending platform.
• CTA sharpness (3 minutes) — is the ask clear and singular? Does it match the newsletter’s topic? Is it appropriate to the relationship stage of this list segment?
• Mobile preview (5 minutes) — send a test to your own inbox and read it on a phone. Line breaks that look fine in a desktop editor often create awkward white space on mobile. Paragraph length matters differently on a 375px screen.

If you’re also running AI automation across other parts of your marketing workflow, the same principle applies everywhere: the edit pass is where quality is maintained, not where it’s created. Build the edit into the workflow rather than treating it as an optional final step.

Segmentation and personalisation: where AI adds more than time savings

The workflow above produces a strong single newsletter. The bigger opportunity is using AI to personalise at segment level without proportionally more work.

If you’re sending to three audience segments — say, enterprise buyers, agency operators, and solo founders — you don’t need three separate briefs or three separate edit passes. You need one core brief plus three variation briefs that specify the angle shift for each segment. The body stays largely the same; the opening, the specific examples, and the CTA change.

• Core body: AI drafts once from the main brief.
• Segment openings: AI drafts three variants from variation briefs; human edits all three openings (takes 15 minutes total rather than three separate production cycles).
• Segment CTAs: templated with segment-specific variables.
• Subject lines: five variants per segment, filtered down to two for testing.

This approach lets a two-person team run a segmented newsletter programme that would previously have required a dedicated email specialist. It also pairs well with a full AI content production workflow, where the newsletter becomes one output in a larger content system rather than a standalone effort.

What to keep human: the non-negotiable list

We’re direct about where AI doesn’t belong in the newsletter workflow, because the failure modes are visible to every subscriber:

• Crisis or sensitive topics — if your newsletter is addressing a difficult industry moment, a company mistake, or anything emotionally charged, do not draft this with AI. The neutral helpfulness of model output is exactly wrong for these situations.
• Personal anecdotes — AI can simulate a personal story convincingly enough that it passes a quick read. Subscribers who know the sender notice. Don’t use AI to fabricate first-person experience.
• Replies and responses — if your newsletter encourages replies and someone sends a genuine, thoughtful response, a human writes back. Using AI to draft reply emails to individual subscribers is the fastest way to erode the relationship the newsletter was built to create.
• Founding voice newsletters — some newsletters work specifically because a named founder writes them. If the audience relationship is built on that identity, AI drafts need a heavier human-pass than usual, or should not be used at all for the main copy.

Measuring whether the workflow is working

Three metrics tell you whether the AI-assisted approach is producing better newsletters or just faster ones:

Open rate trend over 90 days: if open rates are flat or declining while your list is growing, the subject line system needs work. If they’re growing alongside the list, the content is earning its place in the inbox.

Click-to-open rate: this measures content relevance to people who were already interested enough to open. A rising CTOR alongside a stable open rate means the content inside is sharper than before. A falling CTOR means the content isn’t matching what the subject line promised.

Reply rate: harder to measure systematically, but a meaningful signal. Newsletters that provoke real replies are landing. Track this qualitatively — read the replies, note the pattern, feed it back into the brief for the next send.

Getting started: the first three newsletters

The first AI-assisted newsletter will take about as long as writing it from scratch, because building the brief template and voice notes from nothing takes time. The second will take half as long. By the third, the process is embedded.

Here’s the order we recommend:

1. Write the voice notes section of the brief by pulling three past newsletters the sender is proud of and extracting the specific phrases and patterns that make them work.
2. Write the “what to avoid” section by pulling one newsletter the sender is embarrassed by and identifying what went wrong.
3. Run the first brief on a newsletter that isn’t time-sensitive — a welcome email rewrite, or an evergreen “here’s what we believe” piece. This gives you space to iterate the brief without deadline pressure.
4. Build the subject line bank before you need it. Generate 30 subject line candidates across different structures and store them. You’ll draw from this when time is short.

If you’re running newsletters alongside a broader content programme, our post on AI email lifecycle automation covers how to connect the newsletter into a full email strategy that includes onboarding, re-engagement, and campaign sequences.

If you want a second set of eyes on your current newsletter setup — brief structure, send cadence, or subject line approach — we’re happy to take a look. Get in touch and we can start with a quick audit of your last five sends.