In this blog post I'm going to walk you through what AI helps with when you're operating a newsletter that real people read, based on running mine for 21 years. The current state: 15,000 subscribers, 70 per cent open rate, sent weekly, written from scratch every time. The version a working operator would give. Not the version from an AI tool vendor or a "newsletter growth coach" who's never had to hit send.
Most AI newsletter advice in 2026 confuses output volume with newsletter value. The advice goes: write three newsletters a week, use AI to do them faster, segment aggressively, A/B test everything, growth-hack with referrals. Some of that helps. Most of it produces newsletters that nobody reads after the first month because the content has no point of view.
I've been writing newsletters for 21 years. Mine currently lands in 15,000 inboxes every week. Open rate consistently 65-72 per cent (Kit dashboard, last 90 days). Click-through 8-12 per cent, well above industry averages for unsegmented blasts. The newsletter generates inquiries, sales, partnerships, and brand reach. None of this happens because I AI-generate the content. Most of it happens despite the standard AI newsletter playbook.
The frame below is what works. By the end of this blog you'll know where AI saves time without flattening voice, what specifically to never automate, and the boundary line that separates a newsletter people read from one they unsubscribe from.
TL;DR
AI helps with: research before writing, transcription of voice notes, light copy editing, audience pattern synthesis from replies, repurposing one newsletter into related formats. AI hurts: when it generates the opening line, picks the topic, drafts the body, or replies to anyone.
The rule: AI handles the surrounding work, you handle the actual writing and the reader relationship.
The 5 specific AI workflows that work: - Voice-note transcription as raw material - Subject line variation generation (then human pick) - Reply-pattern synthesis from inbox - Old-content surfacing for repurposing - Light proofreading
The 4 things that never get automated: - The actual writing - The topic decision - The reply to any reader - The "in case you missed it" link choices
Why the standard AI newsletter playbook fails
Three structural reasons most AI newsletter strategies produce churn rather than growth.
Volume goals replace value goals. AI makes it easier to send three newsletters a week instead of one. The math seems obvious, more touches, more engagement. The actual math is the opposite. Open rates drop. Unsubscribes climb. The newsletter becomes background noise. The operators with the strongest newsletter relationships in 2026 send less often, not more.
Generation replaces thinking. When AI drafts the newsletter, the operator's role becomes editing rather than thinking. The output is plausible. Sometimes it's even well-written. But readers can detect when the thinking is missing. They might not consciously identify what's wrong. The unsubscribe rate climbs anyway.
Optimisation overrides relationship. Heavy A/B testing, aggressive segmentation, behavioural triggers, these tactics treat the newsletter as a conversion mechanism. Newsletters that build loyal audiences treat the newsletter as a relationship. The two are different operations. Most AI tools optimise for the conversion frame because that's what the metric dashboards reward.
The 5 AI workflows that help newsletter operators
These are the workflows I use every week.
1. Voice-note transcription as raw material
Walk for 30 minutes on Monday morning with your phone recording. Talk through whatever you're thinking about, business, the week, what you noticed, what you're working on, what's bothering you. Transcribe with Whisper or any decent tool.
Output: 3,000-5,000 words of your actual thinking in your actual voice with all the rough edges. This becomes the source material for the week's newsletter.
The transcription itself is the AI's work. The thinking it captures is yours. The newsletter you write afterwards is built from the marked-up sections of the transcript, not generated from a prompt.
I've used this approach for every newsletter I've sent in 2025-2026. Open rates went up after I switched to this method, not down.
2. Subject line variation generation
You have your subject line draft. You want to know if there are better ways to express the same idea. Ask AI for 8-12 variations of the subject line, holding the meaning constant but varying the phrasing.
You read all 12. Pick the one that feels right (or stick with your original, sometimes the original is best). The AI did combinatorial work that would have taken you 20 minutes. You did the editorial decision in 60 seconds.
The trap: letting AI pick the subject line based on "best practice" patterns (curiosity gaps, numbers, urgency). Those patterns are over-used and detected. Pick the subject line your specific audience will respond to, which AI doesn't know.
3. Reply-pattern synthesis from inbox
Newsletter replies are gold. Most operators are too busy to read them carefully. AI can synthesise.
Once a month, export the last month's replies (Kit, ConvertKit, Mailchimp, all have export). Paste into Claude. Ask for: top 5 recurring themes, top 10 common questions, top 5 phrases people use to describe their problems, surprising patterns.
Output: a one-page summary that shapes the next month's newsletter topics. I do this monthly. The themes that emerge inform 60-70 per cent of my next month's content. The patterns are real. The synthesis takes 5 minutes that would have taken an afternoon manually.
4. Old-content surfacing for repurposing
You have 2,000 newsletter issues in your archive. You've forgotten what was in most of them. AI is excellent at scanning historical content and surfacing pieces relevant to current topics.
When I'm writing about a current topic, I'll ask Claude: "I'm writing about X this week. From this list of newsletter titles I wrote between 2010-2025, which past pieces touch the same theme?" Get back 5-10 historical references. Pick one or two for "in case you missed it" mentions. This builds long-term coherence in the newsletter (readers see a worldview developing, not isolated posts).
5. Light proofreading
After you've written the draft, before you send. Paste into Claude with instructions: "Proofread for typos, grammar errors, and clear repetition. Do not rewrite. Do not suggest different phrasing. Do not change voice."
Output: a small list of errors to fix. Accept 60-70 per cent, reject the rest (sometimes "errors" are intentional voice choices).
Notable absence: I never use AI to "improve" the writing beyond basic proofreading. The friction in the writing is what makes it sound like me. AI optimised toward "improvement" often removes the friction.
The 4 things that never get automated
Hard rules. No exceptions in my newsletter operation.
1. The actual writing
AI doesn't draft my body copy. Ever. The reasoning: my newsletter exists because 15,000 people want to know what I'm thinking. If AI drafts it, they're getting what AI is thinking with my name on it. That's a betrayal of the relationship.
This rule eliminates a lot of "time-saving" AI advice. Saving time on the writing means the readers get less of me. Less of me is the opposite of what they signed up for.
2. The topic decision
What I write about each week comes from my own observation, what I'm noticing in my work, what's in the air with clients, what unresolved question I'm sitting with. AI can suggest topics, but they'll be averaged toward what's trending. Trending topics are everyone's topics. My newsletter's job is to bring something only I can bring.
3. The reply to any reader
When someone replies to my newsletter, I read it. If I respond, I write the response myself. If I can't write it that day, it waits. If it requires research, I do the research.
AI-drafted replies, even ones I "review," eventually drift. Drift gets noticed. Noticed drift ends the trust. Hard line on this one.
4. The "in case you missed it" link choices
What I link to from past content matters strategically. AI can surface relevant pieces (workflow 4 above) but the actual decision about which to feature, in what order, with what framing, is editorial work that shapes how the audience perceives the body of work over time. Keep that human.
My actual weekly newsletter workflow
Concrete, in order, with AI involvement called out.
Monday morning, walk: Voice notes, 30-40 minutes, no AI.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Monday afternoon: Whisper transcription. 4,000-5,000 words raw. Skim, mark sections with real ideas, ignore the rest. AI did the transcription only.
Tuesday morning: Write the newsletter draft directly from marked sections. ~1,500-2,000 words. No AI.
Tuesday afternoon: Light AI proofreading pass. Fix obvious errors. Reject suggestions that touch voice. 5 minutes.
Tuesday late afternoon: AI generates 8-10 subject line variations. I pick one. 60 seconds.
Wednesday morning: Send.
AI's total contribution to the weekly newsletter operation: about 12 per cent of the time, all of it on surrounding work. The 88 per cent that matters, what to write about, the actual writing, the voice, the takes, the response to replies, is human.
Newsletter operators who flip that ratio (AI doing 80 per cent, human doing 20 per cent) report short-term time savings and longer-term audience erosion. The math doesn't work over more than 6-12 months.
Specific tools I use
Not affiliate recommendations. Just what's in my actual stack.
- Kit (formerly ConvertKit): newsletter platform. The reason: the audience-tag system is friendly to manual segmentation, the deliverability is decent, the editor doesn't get in the way of writing
- Whisper (via various wrappers): transcription
- Claude: the proofreading, audience pattern work, archive scanning described above
- My own brain: the parts that matter
Notable absences from my stack: AI newsletter generators that promise to "write your newsletter in your voice," scheduling tools that auto-draft sequences, AI-based audience scoring tools that decide who gets what when, voice-cloning tools that produce audio newsletters from text without my reading.
I tried several of those in 2024-2025. None survived more than 8 weeks. Either the output quality dropped or the audience signal dropped or both.
Subscriber growth: where AI helps and where it doesn't
Subscriber acquisition strategy in 2026 is different from 2020. Worth being specific about where AI fits.
Where AI helps for growth: - Researching potential collaboration targets (other newsletter operators) by analysing their content quickly - Drafting outreach for collaborations (then heavily personalising) - Identifying gap topics by analysing your own content history for missing pieces - Generating lead magnet ideas based on reader reply patterns
Where AI hurts for growth: - Auto-DMing on LinkedIn or Twitter (kills brand trust) - Buying sponsorship placements based on AI-recommended audiences (the recommendations are usually wrong) - Running paid acquisition with AI-optimised creative (the creative starts off generic and erodes from there) - Cross-promotion swap programmes run by AI matching engines (matches are usually poor)
The non-AI things that still grow newsletters most reliably in 2026: lead magnets that solve a real problem, guest appearances on podcasts your audience already listens to, direct mentions on other newsletters with audience overlap, sponsorship of high-quality smaller newsletters in adjacent niches.
Deliverability: the AI angle
If your newsletter isn't being delivered, none of the other strategy matters. AI's role here is narrow but useful.
Where AI helps: - Pattern-matching deliverability issues against documented causes (spam triggers, domain reputation, list quality) - Analysing your subject line history for words/patterns that correlate with lower deliverability - Drafting clean unsubscribe sequences that reduce spam complaints - Cleaning email list against engagement patterns (with manual review of the cuts)
Where AI hurts: - Auto-cleaning lists without review (you'll remove people who'd convert later) - Subject line "optimisation" tools that push toward generic patterns - Engagement-prediction tools that auto-suppress content from "low engagement" subscribers
My current deliverability is strong because I keep the list clean manually (suppress non-openers after 90 days of zero engagement, with manual review of the cut list), I write subject lines that match the body (no clickbait), and I never use AI to "improve" subject lines past my own draft.
Monetisation: where AI is changing things
Monetisation is where AI is shifting the economics of newsletter operation in 2026. Worth being specific.
Sponsorship still works but the rates need re-thinking. Newsletter sponsorship rates in 2026 are softer than 2022-2023 because AI tools have made it cheaper to produce content (which raises supply) without making readers more attentive (which limits demand). Operators getting strong sponsor relationships are focusing on niche audience quality, not list size.
Affiliate works for some niches but increasingly transparent disclosure is mandatory. Audiences in 2026 can tell when a recommendation is paid. Smart operators are explicit and recommend only what they'd recommend unpaid.
Direct services from your newsletter (consulting, products, courses) is the strongest 2026 model. Your newsletter becomes your trust-building mechanism, your services close the revenue. AI's role here: helping you respond to inquiry emails faster (drafts only), and scaling the content that supports the services (with all the constraints described above).
I monetise primarily through direct services. The newsletter generates inquiries. The inquiries become discovery calls. The discovery calls become engagements.
Frequently asked questions
Can I use AI to write the whole newsletter and still build an audience? Short answer: no. Slightly longer answer: short-term you'll grow with volume, medium-term the audience erodes as readers detect the lack of person.
Is it okay to use AI to draft "in case you missed it" sections? I use AI to surface candidates, then write the framing myself. Pure AI-drafted recap sections feel mechanical.
How long should a newsletter be in 2026? Mine averages 1,200-2,000 words. Goes down to 600-800 when the topic is tight. Goes up to 3,000+ for deep dives. The right length is whatever the content needs, not a target.
Should I use AI to translate my newsletter for international audiences? Only if you'll have a native speaker review. Auto-translation produces content that sounds non-native to readers who'd notice. If you can't have native review, leave it untranslated.
What about generating images for the newsletter with AI? I rarely use images in mine, the writing is the product. If you do, AI-generated images work for illustrative purposes (concepts, abstract ideas). Don't use them for things readers might think are photographs.
How do I write newsletters when I'm short on time? Send less frequently. A weekly newsletter you wrote beats a thrice-weekly one AI helped draft.
Should I disclose AI involvement in the newsletter? I don't disclose proofreading, transcription, or research synthesis. I would disclose if AI were drafting body copy. Your call varies by audience and trust expectations.
What's the right frequency for a newsletter in 2026? For most operators: weekly. Daily only works for news/curation formats. Monthly works for very deep content. Twice-weekly tends to be the worst-of-both for most operators.
Can AI help with audience research without compromising privacy? Yes, synthesise aggregated patterns from reply content, never identify individuals, never share reader data with external AI services without consent.
Want to talk through your newsletter operation?
If you're running (or planning) a newsletter and want a thinking partner on where AI fits and where it doesn't, that's a conversation I have regularly with operators across coaching, consulting, agency, and SaaS businesses.
I'm Lilach Bullock. I've been a marketing consultant for twenty-one years. My newsletter has been running in various forms throughout that period and currently sits at 15,000 subscribers with consistently strong engagement. I went all in on AI in 2024, and the newsletter has grown stronger since because I kept the AI in its lane and kept the writing mine.
Related reading on AI and content operations
- /ai-marketing-consultant-2026/, main hub
- /ai-for-content-marketing/, broader content marketing with AI
- /ai-for-personal-branding/, AI for personal brand operators
- /what-is-ai-implementation/, what AI implementation means
