Automated Newsletters with AI

Why a newsletter (and why AI)

An email list is one of the few audiences you actually own — a direct line that does not depend on a feed algorithm deciding to show you. The cost is that a newsletter is a recurring commitment, and that is what kills most of them. AI removes a large part of the production load: curating, drafting, proposing subject lines.

Decide the promise before you write issue one

Most newsletters die of vagueness rather than effort. Somebody starts one, publishes four good issues, then cannot decide what the fifth should be about and never sends it.

The fix is a sentence you write before the first issue, covering three things: who it is for, what they get, and how often. "Every Tuesday, one automation you can build in an hour, for small business owners with no developer." That sentence is the subscribe box, the editorial brief for every issue, and the test for whether an idea belongs in it.

It is also what the reader is agreeing to. A newsletter is an appointment they accepted, and the two ways to lose them are breaking the schedule and sending something other than what was promised. Both are decisions you make, not accidents.

Write it down. Put it at the bottom of every issue. It costs one line and it prevents the drift that ends most lists.

What to automate, and what never to

There is a contradiction sitting in the middle of this topic, and it is worth naming rather than stepping around.

The appeal of automation is a newsletter that writes itself. The reason anyone subscribes is that a specific person sifted the noise and told them what mattered — which is judgement, and judgement is the one thing that cannot be automated out without removing the product. A fully automatic newsletter is a feed, and nobody needs another feed in their inbox.

So the split is:

• Automate the gathering. Monitoring sources, pulling articles, deduplicating, assembling raw material into one place. This is genuinely tedious and completely mechanical.
• Automate the drafting scaffold. Summaries, a filled-in structure, subject-line options, formatting, scheduling.
• Never automate the selection or the opinion. What goes in, what gets cut, and what you think about it. That is the newsletter.

A practical version: the machine gives you twelve candidate items with summaries; you pick four and write two sentences of your own view on each. Ten minutes of your judgement on top of an automated pile is a newsletter people keep. An hour of your time reformatting links is not.

Curation is mostly deletion

• Summarise sources: condense articles, releases and trends in your niche into short paragraphs.
• Repurpose: turn existing content — a blog post, a video — into an issue. See repurposing.
• Find angles: ask for topics based on what your audience keeps asking about.

The value a curator adds is not the summaries; those are now free. It is what they left out, and the sentence explaining why the remaining thing matters to this specific audience. A model can hand you fifteen things that happened. Only you can say "ignore all of this except the third one, and here is what it means if you run a small agency".

A useful constraint: fewer items than you want to include. Three items with a view on each beats ten with a link each, because ten links is a search result and the reader already had one of those.

Writing & personalisation

• Fast drafts: give the model a fixed structure — intro, three sections, one call to action — and let it fill it.
• Consistent voice: a good newsletter sounds like a person. Define the tone once and reuse the description verbatim.
• Segmentation: different emphasis per segment, generated quickly.
• Always edit: the opinion, the story and the aside are what earn the next open.

Personalisation is worth a caution: inserting a first name is the least valuable form of it and everyone can see the mechanism. Sending different content to people with different interests is real personalisation, and it requires the tagging discipline described in the platform comparison — one contact, many tags, segments as queries over them.

Subject lines and the field nobody fills in

The subject line decides whether anything else gets read. Generate ten to fifteen and select rather than accept:

• Curiosity — a gap worth opening for, without being a trick. Clickbait that under-delivers trains people to ignore you.
• Benefit — what this issue gives them.
• Short. Mobile clients truncate; assume around forty characters are safe and put the important word first.
• No spam bait. Exclamation marks, all caps, "FREE!!!" — these hurt both deliverability and your credibility.

Then fill in the preview text — the snippet shown after the subject line in most clients. Left empty, it pulls the first words of the email, which is usually "View in browser" or your logo's alt text. It is a second headline you are given for free, and most senders waste it. Write it to continue the subject line rather than repeat it.

The same goes for the from name. A person's name, or a person at a company, consistently outperforms a faceless brand address, and it is the first thing the eye lands on in a crowded inbox.

Open rates are no longer a reliable metric

This matters because a lot of newsletter advice still treats the open rate as the headline number, and it has quietly stopped meaning what it used to.

Opens are measured with a tracking pixel. Mail privacy features now preload images on the recipient's behalf, which registers an open whether or not anyone read anything. The result is inflated and inconsistent open rates that vary by how many of your subscribers use which mail client — so comparing your rate to a published benchmark, or to your own rate from three years ago, tells you very little.

What to watch instead:

• Clicks — a real action, still measurable.
• Replies. The strongest signal there is, both for whether the thing landed and for deliverability. Asking a genuine question at the end of an issue is the cheapest improvement available.
• Unsubscribes and spam complaints per send. A rising complaint rate is the one number to act on immediately.
• The outcome you actually want — replies to an offer, bookings, purchases. Everything else is a proxy.

Staying out of spam

A newsletter nobody receives is worse than one nobody opens. The essentials, covered more fully in the platform comparison:

• Authenticate your sending domain — SPF, DKIM and DMARC. Bulk senders are now rejected outright by major providers for missing these, not merely penalised.
• Send from your own domain, not a free mailbox address.
• Remove the dead weight. Suppress people who have not opened or clicked in six to twelve months. It feels like deleting your audience; it is deleting the part dragging your reputation down.
• Make unsubscribing obvious. Someone who cannot find the link presses "report spam", which costs you far more.
• Never buy a list. It is against every platform's terms and it is the fastest route to a suspended account.

The shape of an issue people finish

Format matters more in email than anywhere else, because the reader is standing up, on a phone, between two other things.

1. Open with the payoff. One or two lines saying what is in this issue and why it is worth two minutes. Not a greeting, not a preamble about your week unless the personal note is the product.
2. Three items, maximum. Each with a heading someone can skim, a summary, and your view. The view is the part they came for.
3. One link per item. Two links means a choice, and a choice at speed usually resolves to neither.
4. One ask, at the end. Reply, click, book — one, and only when you have something worth asking for.
5. The promise line in the footer, so anyone wondering why they are receiving this can see it, alongside the unsubscribe.

Two things that consistently improve engagement and cost nothing: write as though to one person, using "you" and not "everyone", and send from a person. A newsletter that reads like an email from someone gets replies; a newsletter that reads like a publication gets archived.

On design: plain text, or something close to it, usually outperforms a heavily designed template for this kind of newsletter. It renders reliably everywhere, it loads with images off, it looks like a message rather than a mailshot, and it takes less of your time. Save the designed template for a product announcement.

The first ten issues

Almost every newsletter that fails does so in the first two months, and usually for reasons that were decided before issue one.

1. Write three issues before you publish the first. Having a buffer is what carries you through the week when everything goes wrong, and it will tell you quickly whether the promise is one you can actually keep.
2. Send to a small list without embarrassment. Thirty engaged people is a real newsletter; waiting until you have a thousand means never starting.
3. Ask a real question in every early issue and reply to everyone who answers. Those replies are your research, your testimonials and your topics for the next ten issues.
4. Keep the format fixed for the first ten so that changes in engagement mean something.
5. Then review honestly at ten: are people replying, and would you subscribe to this? If the answer to the second is no, change the promise rather than the send schedule.

Notice how little of that is about tooling. The automation described above is worth building once you know what you are producing every week — building it first is a common and expensive way to avoid the harder question of what the newsletter is for.

Where subscribers actually come from

Growth is usually the real constraint, not writing. Three things do most of the work:

• A specific reason to join, stated where people already are. "Subscribe for updates" converts poorly; the promise sentence from earlier converts better because it says what arrives.
• Something worth trading an address for — a template, a checklist, a tool. Make it genuinely useful and directly related to what the newsletter covers, or you will attract people who want the freebie and not the emails.
• Asking at the moment of value. At the end of the article they just read, not in a pop-up before they have read anything.

And send the first email immediately. A welcome email while they still remember subscribing is the highest-engagement message you will ever send, and it sets the expectation for everything after it.

The automation flow, with the gate in it

Build it in n8n or Make: gather sources → summarise and assemble a draft → send it to you for approval → publish through your email platform at a fixed time. See automation templates.

The approval step is not a formality, and it should be a real editorial pass rather than a glance — picking items, cutting two, adding your view. Build the flow so that nothing sends without a positive action from you: no timeout that publishes by default, because the one week you are busy is the week something wrong goes out under your name to your whole list.

Two safeguards worth adding while you are in there. Send yourself a test to a real inbox first and look at it on a phone. And keep a suppression check so an unsubscribe is honoured even if your automation pulled the list earlier — a person who unsubscribes and then receives another issue is a complaint, and usually a fair one.

The rules that apply

Marketing email is regulated in most markets, and the obligations follow your subscribers rather than your location. The constants: people must have opted in, your real identity and a postal address must appear in the email, unsubscribing must be easy and honoured promptly, and you should be able to show how each person joined. Double opt-in costs you some signups and produces exactly the evidence a complaint asks for.

Making money from it, without spending the trust

A list is an asset because people let you into a private space on a recurring basis. Everything you can earn from it is drawn against that, so the order you do things in matters.

• Your own products and services — the highest return and the least damaging, because a relevant offer to people who already trust you reads as useful rather than as an advert. This works at list sizes far too small for anything else.
• Affiliate recommendations — viable early, and only for things you would recommend unpaid. Disclose every time, in the email itself and not only in a footer. One recommendation that disappoints costs more than the commission on ten that did not.
• Sponsorship — real money once the audience is specific enough that a sponsor wants exactly those people. Being narrow is worth more than being large here.
• Paid subscriptions — the hardest, and the honest test is whether people would miss it. If nobody replies to your free issues, nobody will pay for more of them.

One rule that keeps the asset intact: never send an issue whose only purpose is to sell. If the value and the offer arrive together, the offer is part of something they wanted. If the offer arrives alone, it is an advert they did not subscribe to, and the unsubscribe link is right there.

When not to start one

• You cannot sustain the cadence. Monthly that arrives beats weekly that stops. Promise the interval you will keep on a bad week.
• You have nothing to say that a search would not answer. Summaries are free now; a view is not.
• Nobody has asked. If your existing content gets no replies or questions, there is no demand to convert yet.
• The plan is to sell from issue one. The list is an asset because of trust, and trust is built before it is spent.

Common mistakes

• A fully automated issue. No opinion, no selection — a feed with a subject line. It is the failure this whole format is prone to.
• Too much, too often. A good weekly beats a thin daily. The inbox is not yours.
• Selling only. Give first, and keep the ratio honest.
• Neglecting deliverability. Authentication and list hygiene decide whether any of this happens at all.
• Optimising the open rate. It is no longer a trustworthy number. Watch clicks, replies and complaints.
• An empty preview text. A free second headline, wasted on "View in browser".