Every week another tool promises to write your entire AI newsletter for you. Brief it in the morning, send it by lunch, spend the afternoon doing something more interesting. The pitch is seductive enough that a lot of newsletter operators have tried it, and a lot of them have quietly gone back to writing most of it themselves. That's not because the tech is useless. It's because most people are applying it to exactly the wrong things.
After watching this space closely and talking to creators running newsletters from a few hundred subscribers to well over a hundred thousand, the picture is pretty clear. AI genuinely earns its keep in specific, unglamorous parts of the workflow. In others, it produces something that looks like a newsletter the way a waxwork looks like a person.
The Stuff AI Actually Gets Right
Start with the tasks nobody enjoys. Research synthesis is the obvious one. If you cover a beat, you're spending real time each week reading through sources, pulling out the relevant bits, and deciding what's worth your readers' attention. A well-prompted AI can cut that triage time dramatically. Not by replacing your editorial judgement, but by doing the first pass so you're editing rather than starting from scratch.
Subject lines and preview text are another genuine win. Most writers are terrible at these, not because they lack skill, but because after you've spent two hours on an issue you're too close to it. Ask an AI for fifteen subject line variations and at least three of them will be better than what you'd have written in a hurry at the end of a long session. The click-through rate improvements some operators report from this alone, often in the 15 to 25 percent range, are hard to argue with.
Structural scaffolding also works well. Telling an AI 'I want to cover these four topics in roughly this order, give me a section outline' and then writing into that skeleton is a legitimate productivity approach. You're not outsourcing the thinking. You're outsourcing the blank page problem, which is a different thing entirely.
Repurposing is underrated here too. Taking a long interview transcript and asking AI to surface the five most quotable moments, or turning a detailed how-to piece into a shorter digest version for a different audience segment, these are tasks where the technology is genuinely useful and where the output quality is high enough that your editing time is minimal.
Where the AI Newsletter Collapses
Here's where it gets ugly. Ask AI to write the actual body of a newsletter, the analysis, the opinion, the 'here's what I think this means' section, and what you get is technically correct, competently assembled, and completely forgettable. It hedges where you should take a position. It summarises where you should provoke. It writes like someone trying very hard not to be wrong rather than someone with a genuine point of view.
Readers can't always tell you what's off about it, but the numbers don't lie. Open rates hold because your subject lines are good. But replies drop. Forwards drop. The feeling of being in someone's trusted inner circle, which is what great newsletters sell, just evaporates. That's a slow bleed, not a sudden crisis, which makes it particularly dangerous because you might not catch it until you've already lost the audience's trust.
Anything time-sensitive is also a problem. AI training data has cutoff dates, and even tools with web access get things wrong often enough that you'd need to verify every claim anyway. If your newsletter's value is in being fast and accurate on a fast-moving topic, AI is at best a neutral tool and at worst a liability.
Voice is the hardest thing to replicate. Not your vocabulary or your sentence structure, though those matter. The deeper thing: the specific cultural references you reach for, the instinct about when to be irreverent and when to be straight, the things you choose not to say. These come from a particular life and a particular editorial sensibility built over years. No amount of prompting recreates that. You can get AI to approximate your style. You will know it's an approximation, and eventually so will your readers.
'The newsletters I love feel like getting an email from a smart friend. I've never once thought that about AI-generated content. It feels like getting an email from a smart friend's very competent assistant who has never actually met me.'
The Hybrid Model That Actually Works
The operators doing this well aren't using AI to write their newsletters. They're using AI to write around their newsletters. Research, structure, subject lines, repurposing, reformatting for different channels. The actual analysis and opinion is theirs, written by them, unassisted.
One way to think about it: AI handles the connective tissue. You provide the beating heart. That division of labour is sustainable. It saves real time without hollowing out the product.
The practical workflow that keeps coming up in conversations with experienced operators looks something like this:
- Use AI to scan and summarise your source material before you start writing
- Write your own takes, opinions and analysis entirely without assistance
- Use AI to generate subject line and preview text options once the issue is done
- Use AI to repurpose the finished issue for social posts or a shorter digest
- Use AI for any templated or recurring sections, such as events listings, job boards or sponsored content bridges
Notice what's not on that list. The main body copy. The editorial voice. The reason people subscribed in the first place.
Prompting Makes a Bigger Difference Than You Think
If you're going to use AI for the tasks where it earns its keep, the quality gap between a lazy prompt and a careful one is enormous. Most people write prompts like search queries and then complain that the output is generic. It is generic. Because generic input produces generic output.
For research synthesis, give the AI context about your audience, what they already know, what angle you're likely to take, and what you want to ignore. For subject lines, tell it the tone of the issue, the single most interesting thing in it, and your readers' biggest pain point right now. For structural outlines, share your previous issues as examples so it understands your format rather than defaulting to a format it's seen ten thousand times before.
The investment in building these prompts properly, writing them out, saving them, refining them over time, pays back quickly. It's the difference between AI that saves you forty minutes a week and AI that produces output you spend forty minutes fixing.
What This Means If You're Building for the Long Term
The newsletters with genuinely loyal, engaged audiences have something in common. The reader has a relationship with a person, not a publication. They know the writer's quirks, trust their judgement, and feel like the newsletter is written specifically for someone like them. That relationship is built slowly and lost quickly.
AI newsletter tools, used wrong, chip away at exactly that. You save three hours a week and spend twelve months slowly losing the thing that made your newsletter worth reading. That's a bad trade even if it doesn't feel like one in the moment.
Used right, the same tools give you back time to do the things that actually build the relationship. More reporting. More talking to readers. More thinking before you write rather than thinking while you write. That's a good trade.
Platforms like Aldus are designed around this reality, giving writers tools that handle the operational side of newsletter production without nudging them toward outsourcing the editorial core. That's the right instinct. The writers who will still have loyal audiences in five years are the ones who treat AI as a capable production assistant, not a ghostwriter.
The technology will keep improving. The fundamental problem won't change. Readers subscribe to people. As long as that's true, the writer who shows up authentically every week will outlast the one who automated themselves out of their own newsletter.