How to start an AI newsletter for your small business

What is an AI newsletter, exactly?

An AI newsletter is a regular email to your own subscribers where the research and the first draft are done by software, and a person approves it before it sends. The AI reads the sources you nominate, picks what matters to your audience, and writes it up in your voice. You read the draft and decide whether it goes.

That last part is what separates a useful AI newsletter from the ones that give the category a bad name. Fully automated sending with no human check is how businesses end up emailing their customers a confident, well-written claim that is simply wrong. The value is in removing the blank page and the weekly research, not in removing you.

It is worth being clear about what it is not. It is not a mass-mail blast to a purchased list, it is not spun content rewritten from someone else’s blog, and it is not a substitute for having something to say. If your business has nothing your customers want to know about, no tool fixes that.

It is also different from asking a chatbot to "write me a newsletter". A chat window gives you one draft from whatever it happens to know, with no sources, no memory of your voice from last time, and no check on whether the claims are true. A newsletter tool holds the configuration (sources, voice, sections, schedule), does the reading every week, and keeps a record of what it verified. The difference shows up by the third issue, when the chat version has drifted and the configured one has not.

Is a newsletter still worth it for a small business?

Email remains the only audience you actually own. Your followers on any social platform are rented: the platform decides who sees your posts, changes that decision without telling you, and can remove your account. A subscriber list is a direct line that nobody can throttle.

For a small New Zealand business the practical case is narrower than the marketing case. A newsletter earns its place if you have something genuinely useful to say on a repeatable schedule, and if the people on your list would notice its absence. A builder telling subcontractors what changed in the retentions rules is useful. A cafe sending a monthly "hello from the team" is not, and the open rate will tell you so within three issues.

The reason most small-business newsletters die is not that the writing was bad. It is that a busy fortnight broke the cadence and it never came back. Consistency, not brilliance, is what makes the list worth having, and consistency is exactly the part software is good at.

There is a second, quieter benefit. A newsletter that arrives reliably every fortnight keeps you in front of customers between jobs. Most small businesses lose work not to competitors but to being forgotten: the customer needed a plumber, could not remember yours, and searched. A short, useful email every couple of weeks is the cheapest way to be the name they remember, and it does not require you to be good at social media.

Who is it for, and who should skip it?

It suits businesses whose customers have an ongoing reason to hear from them: trades and construction (rule changes, product recalls, council consents), professional services (tax dates, legislation, case law), agriculture and horticulture (prices, weather windows, biosecurity notices), retail with a knowledgeable audience (new stock, how-to content), and any business that sells to other businesses in a regulated sector.

It suits people who already spend time keeping up with their industry and would rather that time produced something. If you read the trade press over breakfast anyway, an AI newsletter turns that reading into an asset your customers see.

It does not suit a business whose whole appeal is personality on the page. If your readers subscribe for your writing, they will notice when it is not yours. It also does not suit a business with nothing published to draw on: if your sector has no trade press, no regulator and no suppliers who announce anything, there is nothing to research and the issues will be thin. And it does not suit anyone unwilling to read a draft before it sends. That five minutes is not optional.

How do you choose what goes in it?

Start with sources rather than topics. Pick between five and fifteen places your industry actually publishes: a regulator, a trade body, two or three trade publications, the suppliers whose prices move, and the councils or agencies whose decisions affect your customers. In New Zealand that often means MBIE, WorkSafe, your regional council, and one or two sector bodies.

Then say what to avoid. Excluding your competitors’ marketing blogs, press-release wire copy and anything paywalled will do more for quality than any amount of prompt tuning. A source list of ten good publications beats fifty indiscriminate ones.

Finally, decide the shape of an issue and keep it. Three to five items, each one paragraph, each with a link to where it came from, is a format readers learn quickly. Brieva builds every issue from the sections you configure, so the shape stays the same even when the content does not.

A useful test for any source: would you forward an article from it to a customer? If the answer is no, it does not belong in the list, no matter how often it publishes. Volume is not the problem you are solving. Relevance is.

What should the first issue look like?

Short. The first issue sets the expectation for every issue after it, and a long first issue promises something you will not sustain. Aim for one screen on a phone: a two-sentence introduction in your own words, three items with a source link each, and one line at the end that says what you want the reader to do, if anything.

A concrete example. An electrical contractor in the Waikato sends a fortnightly issue to about two hundred customers, mostly landlords and small commercial tenants. A typical issue has: a note on a change to the healthy homes standards deadline with a link to the Tenancy Services page; a supplier bulletin about a recalled brand of RCD, linked to the recall notice; a short item on the council’s consent processing times from the council’s own report; and one line offering a pre-winter switchboard check. That takes a reader ninety seconds and every claim in it has a source a reader can open.

Notice what is not in it: no stock photography, no "welcome to our newsletter", no essay about the state of the industry. The reader learns three things they did not know and sees one offer. That is the whole job.

Once the shape is set, resist adding to it. The temptation after a good first month is to add sections. Every added section is more to read and more to approve, and the open rate will drift down as issues get longer. Add a section only when you have removed one.

How do you keep it sounding like you and not like AI?

Voice is a configuration problem, not a prompting problem. The things that make writing sound like a specific business are concrete: the words you use for your own services, the ones you refuse to use, how formal you are, whether you use first person, how long your sentences run, and what you always say at the end.

Give the tool those rules once. Ban the vocabulary that gives AI away in your industry: "delve", "landscape", "leverage", "in today’s fast-paced world", and any sentence that begins "In conclusion". Specify whether you write "we" or "I". Say whether you swear. These constraints do more work than any amount of asking for a "friendly, professional tone".

Then read the first three issues properly, end to end, and correct what grates. After that the corrections become rare. If you find yourself rewriting every issue heavily, the voice configuration is wrong, not the model.

The single most effective voice rule is the list of words you never use. Every trade has them: the words your suppliers’ marketing departments love and your customers roll their eyes at. Write down ten of them. A draft that avoids those ten words will sound more like you than one that was asked, in the abstract, to sound like you.

Keep one paragraph of every issue genuinely yours. The intro is the natural place. Two sentences typed by you before you approve, saying what you think about the week’s items, does more for the newsletter’s character than any amount of configuration. It is also the part readers reply to.

How do you stop it publishing something wrong?

Assume the draft contains at least one claim that is not quite right, because it usually does. The defence is a checking pass that opens every link and compares each claim to what the source actually says, and a human who approves before anything sends.

Brieva runs that pass automatically and hands you a scorecard with the draft: which claims were verified against their source, which were softened because the source was weaker than the draft implied, and which were cut because no source supported them at all. You approve or you do not.

This matters more than it sounds. An AI-written newsletter that is 95% accurate is worse than no newsletter, because the 5% goes out under your name to the people most likely to know it is wrong. Your customers do not grade on a curve.

The failure modes are predictable, which is why they can be checked for. Numbers get rounded into something the source did not say. A proposal gets reported as a decision. A quote gets attributed to the wrong person. A link goes to a page that has since changed. A checking pass catches most of these; a reader who knows the topic catches the rest. That is why the approval step is a person, and why it should be the person in the business who knows the subject best.

When the scorecard flags something, take the flag seriously even when the sentence reads well. The most dangerous errors are the fluent ones.

Should you send from your own domain?

Yes, once you are past the trial stage. Sending from your own verified domain means the from line says your business rather than a platform’s, and it means the reputation you build belongs to you rather than being pooled with every other customer of that platform.

Technically it requires three things in your DNS: DKIM records so receiving servers can verify the mail really came from you, an SPF record listing who is allowed to send on your behalf, and ideally a DMARC record telling receivers what to do when something fails. Brieva generates the records; you or your web person adds them, and verification usually completes within an hour of the DNS propagating.

On the free tier Brieva sends from brieva.news, which is fine for testing but means the reputation is shared. Every paid plan sends from your own domain.

If you have never touched DNS, it is less frightening than it sounds: it is a handful of copy-and-paste records in the same control panel where your website’s address is set, and the guide on sending from your own domain walks through each one. Most people do it in twenty minutes, most of which is finding the login.

What does the law require in New Zealand?

Two pieces of legislation apply. The Unsolicited Electronic Messages Act 2007 requires that you have consent to send commercial email, that the message clearly identifies who sent it and how to contact them, and that it includes a functional unsubscribe that works for at least 30 days after sending. Consent can be express or, in limited circumstances, inferred from an existing business relationship, but express is what you want on record.

The Privacy Act 2020 governs the subscriber data itself: you must collect it for a stated purpose, keep it secure, let people see and correct what you hold about them, and not keep it longer than you need. A subscriber list is personal information.

In practice, double opt-in satisfies both comfortably. Someone enters their address, receives a confirmation email, and is only added when they click it. You have a timestamped record of consent, and you have proof the address belongs to a real person who wanted it. Brieva does this on every subscription and does not offer a bulk import, precisely so a purchased list cannot get in.

The question people actually ask is whether they can add their existing customers. The honest answer is: only if those customers would reasonably expect it, and even then, ask. An email to existing customers inviting them to subscribe, with a link to a form, is compliant, polite, and converts well because these are people who already trust you. Silently adding them is the thing the Act was written to stop, and it is also how you get your first spam complaint.

And in Australia, the UK and the US?

The shape is the same everywhere: consent, identification, unsubscribe, and a duty of care over the data. The statutes differ.

Australia: the Spam Act 2003 requires consent (express or inferred from an existing relationship), that the message identifies the sender, and a working unsubscribe actioned within five business days; the ACMA enforces it and fines businesses that keep sending after an unsubscribe or send to bought lists. The Privacy Act 1988 and the Australian Privacy Principles govern the list for businesses over the turnover threshold. There is a separate Australian guide to newsletter sources and ideas by industry.

United Kingdom: PECR requires consent for marketing email to individuals, with a "soft opt-in" for existing customers, and UK GDPR governs the data; the ICO enforces both. Double opt-in is not strictly required but is the accepted way to evidence consent.

United States: CAN-SPAM is opt-out rather than opt-in, but requires an honest subject line, a physical postal address in the message and an unsubscribe honoured within ten days; state privacy laws are adding consent requirements on top. Mailbox providers, not the law, are the real constraint there: Gmail and Yahoo require one-click unsubscribe and low complaint rates for bulk senders regardless of jurisdiction.

Brieva applies the strictest common denominator to every newsletter: double opt-in, one-click unsubscribe actioned immediately, sender identification in every footer, and no bulk import. That satisfies all four regimes without configuration.

How do you get subscribers?

From the places people already interact with you. The sign-up form goes on your website, ideally on the pages people visit when they are deciding whether to use you. A one-line invitation goes in your email signature, on invoices and quotes, and on the thank-you page after a booking. If you have a counter or a workshop, a small card next to the EFTPOS terminal works better than anyone expects.

Tell people what they will get and how often. "Fortnightly: what changed in the rules, what it means for your property, no sales pitch" will out-perform "Subscribe to our newsletter" every time, because it answers the only question the reader has, which is whether this will be worth opening.

Do not expect the list to grow quickly, and do not need it to. A trade newsletter with two hundred subscribers who all own or manage property in your area is worth more than one with five thousand who found you through a giveaway. Growth of a few subscribers a week from the right places is a healthy list.

Never buy a list, never scrape one, and never import a list from a previous business without fresh consent. The subscribers on those lists did not ask for your email, they will mark it as spam, and a spam complaint rate above a fraction of a percent will get your sending paused before the second issue. Brieva refuses bulk imports for exactly this reason.

How often should you send?

As often as you can reliably say something useful, and no more. For most small businesses that is fortnightly. Weekly works if your industry genuinely moves weekly, which is true of construction, agriculture and anything regulated. Monthly is the floor: less often than that and readers forget they subscribed, which is where bounces and complaints come from.

Daily is a different product. It suits market and trade news, where readers want a short morning brief and the sources publish every day. It does not suit a business whose customers hear from it a few times a year. If you are wondering whether to send daily, the answer is almost certainly no; the businesses that should send daily already know.

Pick a day and a time and keep them. Readers build a habit around a newsletter that arrives at the same time; they do not build one around a newsletter that arrives when someone got round to it. Brieva holds the schedule for you: research runs, the draft arrives for approval, and the issue goes at the time you set once you have approved it.

What should you measure?

Three numbers, and only three to begin with. The open rate tells you whether the subject line and sender are trusted. The click rate tells you whether the content was worth acting on. The unsubscribe and complaint counts tell you whether you are annoying people.

Small, well-built lists open at rates that would look absurd to a mass marketer, often well over half, because everyone on the list asked to be there and knows the sender. If your open rate is under a third, the list has the wrong people on it or the subject lines are not saying what the issue contains. If clicks are low but opens are high, the items are interesting but not actionable, which may be fine.

Watch the trend rather than the number. A list that opens at 55% one month and 48% the next has told you something: the issues got longer, or drifted off topic, or the send time moved. Fix that before it becomes 40%.

Ignore vanity comparisons with industry averages. Those averages are dominated by large consumer lists and say nothing about a trade newsletter to two hundred landlords. Compare this month to last month.

What are the common mistakes?

Too many sources. Fifty feeds produce a draft full of marginal items and a research step that takes longer and finds less. Ten good sources is the target.

Too much in an issue. Readers give a newsletter about ninety seconds. Three items read beat eight items skimmed.

No opinion. A draft that summarises the week without saying what you think about it reads like a press digest. One sentence of your view per issue is enough to fix this.

Approving without reading. The checking pass catches most errors; it does not catch the item that is technically accurate and commercially embarrassing, like enthusiastically reporting a supplier’s price rise to your customers. Read it.

Changing the format every issue. Readers learn a shape. Keep it.

Letting the list go cold. A list that has not heard from you for six months will bounce and complain when you come back. If you are going to pause, tell them, and when you restart, send a short re-introduction first.

What does an AI newsletter cost?

Brieva is free for lists under 50 subscribers, with one newsletter sent weekly or less often from brieva.news. Paid plans start at NZ$9.99 a month (Solo) for up to 1,000 subscribers and sending from your own domain, NZ$19.99 (Pro) for up to 5,000 subscribers, three newsletters, daily issues and a custom logo, and NZ$99.99 (Scale) for up to 25,000 subscribers and ten newsletters. Prices exclude GST.

Compare that to the real alternative, which is not another tool but your own time. A researched, written and checked issue takes most people two to four hours. At any sensible hourly rate, a fortnightly newsletter costs more in time than any of these plans cost in money, and the time cost is the one that makes people stop.

The honest caveat: if you would only send four issues a year, none of this is worth setting up. The economics work when the cadence is regular.

There is no per-send charge and no charge for the AI research. The plan price is the whole cost, which matters because it means a busy news week does not become an expensive one.

How long does it take to set up?

The first issue takes about an hour of your attention: choosing sources, setting the voice rules, picking the sections and the cadence, and reading the first draft carefully. Domain verification runs in parallel and depends on how quickly your DNS updates.

After that, each issue is one decision. The draft arrives on your schedule with its scorecard, you read it, and you approve or send it back. Most people spend five minutes an issue once the voice settles.

The setup hour breaks down roughly as: fifteen minutes finding and pasting sources, ten minutes on voice rules and the words you avoid, ten minutes choosing sections and a template, five minutes on the schedule, and twenty minutes reading the first draft and sending it back with notes. The test run is the important part; run it before you invite anyone to subscribe.

What is the first thing to do?

Write down the five sources you would check if you had an hour a week to keep up with your industry, and the three questions your customers ask you most often. That list is the whole configuration, and it is the part no software can do for you.

Then start on the free tier with one newsletter and a small list. Read three issues before you decide whether it is worth paying for. If the drafts are not useful by the third issue, the sources are wrong, and no plan upgrade will fix that.

Invite ten customers you know well before you invite anyone else. They will tell you what is missing, and their replies will be the most useful feedback the newsletter ever gets.