AI newsletter vs writing it yourself: an honest comparison

What does writing it yourself actually cost?

For most people a researched, written and proofread issue takes two to four hours: reading the week’s sources, deciding what matters, drafting, checking the links, and formatting. Call it three hours. Fortnightly, that is roughly seventy-five hours a year.

The money cost is whatever your time is worth, but the real cost is variance. The three hours have to exist in a week where a job went wrong, a customer needed you, or you were on site. When they do not exist, the issue slips, and a slipped issue is how newsletters die.

Writing it yourself does buy something real: full control, genuine voice, and the thinking that happens while you write. Some people find the writing is where they work out what they actually think about their industry. If that is you, do not automate it away.

Break the three hours down and the shape of the problem becomes clearer. Roughly an hour is reading: opening the trade press, the regulator, the suppliers, and skimming for what changed. Half an hour is choosing: deciding which three of the fifteen things you read matter to your customers. An hour is writing and rewriting. Half an hour is the fiddly end: checking links, fixing formatting, sending a test, fixing the thing the test revealed. Only the middle half hour, the choosing, needs to be you. The rest is work that happens to be done by you because there was no one else.

What does an AI newsletter cost?

Money, a little: from nothing on a free tier to NZ$9.99 to NZ$99.99 a month depending on list size, excluding GST. Time, about an hour to set up and five minutes an issue after that. The five minutes is reading the draft and its scorecard and pressing approve, or sending it back with a sentence about what to change.

The less obvious cost is the setup hour. It is not hard, but it is the hour where you decide what the newsletter is: which sources, what voice, what shape. Skipping it and accepting defaults produces a generic newsletter, and a generic newsletter is worse than none because it trains readers to ignore you. Do the hour properly.

The cost that people worry about and should not is the AI itself. On a managed platform it is included in the plan; there is no per-issue charge that turns a busy news week into an expensive one. What you should worry about instead is whether the platform checks its own work, because an unchecked draft costs nothing to produce and a great deal to have sent.

Where does AI clearly win?

The tedious parts. Scanning a dozen sources, discarding the 90% that is irrelevant, summarising what is left, and producing a consistent first draft on schedule. That work is real, it is repetitive, and it does not benefit from your judgement until the draft exists.

It also wins on consistency, which is the thing that actually determines whether a newsletter is worth having. Software does not have a busy fortnight. The blank page never stalls an issue, because there is never a blank page.

And it wins on the mechanical parts of quality: checking that every link resolves, that a quoted figure matches its source, that nothing has been asserted without support. A person doing that carefully at eleven at night does it worse than software doing it every time.

It wins on breadth, too. A person keeping up with an industry reads the three or four sources they have always read. Software reads fifteen every week without getting bored, which is how the item from the regional council’s obscure consents report ends up in your issue and not your competitor’s.

Where does AI alone fall short?

Judgement about what matters to your specific customers. Software can tell you that a regulation changed; it cannot always tell you that this particular change is the one your three biggest clients will ring you about. That is domain knowledge and it lives with you.

Opinion, too. The most valuable line in a trade newsletter is often "here is what we think this means for you", and that sentence has to be yours. AI-written analysis that has no stake in the outcome reads exactly like what it is.

And accuracy without supervision. An unsupervised model will cite a link that does not say what it claims, confidently. Without a checking pass and a human approval, that goes out under your name.

It also cannot know what happened in your business this week. The job that went sideways, the customer question you have now been asked three times, the supplier who has just changed terms: these are the things readers most want to hear about and the things no source publishes. If they are going in, you are typing them.

What does the hybrid look like in practice?

AI researches and drafts, a checking pass verifies every claim against its source, and you approve before anything sends. You keep the editorial control and the voice; you lose the weekly grind and the blank page.

In time terms, that turns three hours an issue into about five minutes: read the draft, read the scorecard showing what was verified, softened or cut, add a line of your own opinion if you have one, and approve.

The five minutes are the important part. They are what keeps it yours, and they are what stops the failure mode where a business quietly stops reading its own newsletter before its customers do.

Here is how a fortnight actually runs. On the Wednesday evening before the send, the research runs and the draft arrives in your inbox with its report card. Thursday morning, with a coffee, you read it on your phone. Two of the four items are fine. One has been softened by the checker because the source said "proposed" where the draft had said "confirmed", which you agree with. One you would rather lead with, so you tap Edit and type "lead with the WorkSafe item, drop the funding one, shorter intro". Twenty minutes later a revised draft arrives. You add one sentence at the top about the job you saw last week that illustrates the WorkSafe point, and approve. It sends at 7am Friday to two hundred people. Total attention: eight minutes across two mornings.

A side-by-side

Time per issue: by hand, two to four hours; hybrid, five to ten minutes; fully automated with no review, zero, which is the problem.

Consistency: by hand, depends entirely on the fortnight you are having; hybrid, the draft always arrives, and the only thing that can slip is your approval; fully automated, perfect consistency including when it is wrong.

Voice: by hand, unmistakably yours; hybrid, yours by configuration plus whatever you type into the intro, which most readers cannot tell apart from the real thing after the first few issues; fully automated, generic within a month.

Accuracy: by hand, as good as your attention that evening; hybrid, every claim checked against its source and a person reading the result; fully automated, whatever the model produced.

Research breadth: by hand, the sources you remember to check; hybrid and automated, every source you configured, every time.

Opinion and news from inside the business: by hand, naturally; hybrid, one typed sentence per issue, which is enough; fully automated, none.

Cost: by hand, your hourly rate times three, every fortnight; hybrid, a plan fee and a few minutes; fully automated, a plan fee and a reputation risk.

What about hiring someone to write it?

The third option, and for some businesses the right one. A freelance writer or a marketing agency will produce a fortnightly newsletter for somewhere between a few hundred and a couple of thousand dollars a month in New Zealand, depending on length and how much research is involved.

What you get: a human who can develop opinion, interview you for the inside-the-business items, and write with real personality. What you give up: money, obviously, and a surprising amount of time, because the writer needs to be briefed every fortnight on what matters, and a good writer’s questions take longer to answer than reading a draft does. Turnaround is days, not hours, so an issue cannot react to something that happened yesterday.

The failure mode is that the writer does not know your industry, and it shows. A generalist writing about retentions rules for a building company produces something accurate and slightly off, the way a translation is slightly off. Specialist writers exist and cost more.

Where hiring wins outright: when the newsletter is a genuine editorial product, long-form, opinion-led, and part of how the business positions itself. Where the hybrid wins: when the newsletter is an information service to customers and the value is that it arrives, is accurate, and is short. Most small-business newsletters are the second kind.

Three businesses, three right answers

A sole-trader accountant with three hundred clients, who sends tax dates and rule changes fortnightly. She tried writing it herself for a year and sent eleven of twenty-six issues. The hybrid suits her exactly: the sources are Inland Revenue and MBIE, the items are factual, the checking pass matters because a wrong date is a real problem, and her one typed sentence per issue is the reason clients read it. Cost: NZ$9.99 a month and ten minutes a fortnight.

A boutique architecture practice whose monthly letter is a short essay about a building, a material or an idea, written by one of the partners. It is the practice’s marketing and its voice is the point. AI would make it competent and forgettable. The right answer is to keep writing it, and if the partner’s time is the constraint, to send it less often rather than automate it. A chat assistant for a first pass at the least interesting paragraph is fine.

A regional produce wholesaler that wants a daily morning brief for two hundred retail customers: prices, weather windows, what is short this week. Daily is impossible by hand and a freelancer cannot turn it round overnight. The hybrid is the only option that works, with the internal price list added as a source and the sales manager approving from the phone at 6:30. Cost: NZ$19.99 a month for the daily cadence, and five minutes a morning.

Questions people ask before deciding

Will my customers know it was written with AI? Not from the text, if the voice rules are set and you type the intro. Whether you tell them is a choice; some businesses put "researched with AI, checked and approved by us" in the footer and readers seem to like the honesty. What readers notice is not the tool but the absence of a person: generic phrasing, no opinion, nothing from inside the business. The five minutes are what prevent that.

What if it says something wrong? The checking pass catches the factual errors before you see the draft and shows you what it changed. The one it cannot catch, the claim that is accurate but unwise, is why you read the draft. In practice an approved issue from the hybrid has fewer errors than a hand-written one, because the checker opens every link and a tired person does not.

Can I write some issues myself and let it draft others? Yes. The configuration does not care who wrote the last issue. Some people draft the year-end issue by hand and let the tool handle the fortnightly ones.

How long until it sounds like me? Two or three issues, if you send the first ones back with specific notes. The notes are the training: "shorter sentences", "never say leverage", "always end with the phone number". After the third issue the corrections become rare.

What happens when I go on holiday? The draft still arrives and waits for approval. Nothing sends until someone approves it, so either you approve from the beach, delegate approval to someone in the business, or the issue skips. All three are fine. What does not happen is an unread issue going out.

When should you not use an AI newsletter?

If your newsletter is a personal essay, do not automate it. If the value is your writing itself rather than the information in it, software will produce something competent and characterless, and your readers will notice within two issues.

If you send fewer than about six issues a year, the setup is not worth it. If your industry has no reliable published sources, there is nothing to research, and the drafts will be thin.

And if you are not willing to read the draft before it sends, do not do this at all. An unread AI newsletter is a liability with a schedule.

A less obvious case: if the newsletter is mostly about you (new staff, jobs completed, the office dog), there is nothing external to research and the AI has little to add. That kind of newsletter is fine, and short, and you should keep writing it yourself in twenty minutes a month.

Can you switch back?

Yes, and it is worth knowing that before you start. A newsletter platform holds three things of yours: the subscriber list, the configuration, and the archive of sent issues. The list should be exportable at any time as a plain CSV; if a platform makes that hard, that tells you what it thinks of you. The configuration is a page of notes you can rewrite anywhere. The archive is HTML you can keep.

What does not transfer is the sending reputation, unless you were sending from your own domain. That is the practical reason to set up your own domain early even if you are not sure you will stay: it makes the decision reversible.

How do you decide?

Answer one question honestly: over the last twelve months, did you send every issue you intended to send? If yes, and you enjoy the writing, keep writing. You have the rarest thing in this category, which is a sustainable habit.

If no, the choice is not between a human newsletter and an AI one. It is between an AI-drafted newsletter that actually goes out and a human one that mostly does not.

And if you have never sent one, start with the hybrid and a small list. Three issues will tell you whether your customers want it, which is the only question that matters, and it will have cost you an hour and a few dollars to find out.