This is an uncomfortable opening for a piece that's supposed to set us apart: almost everyone says this by now.

We thought so too, until we measured it. And then it turned out the principle wasn't our position but the consensus of an entire category. That's an unwelcome outcome to write down and, at the same time, the only honest place to start.

Because the question that remains isn't whether a human decides. That question is answered. The question is what happens on the day nobody's watching.

01 / the definitionWhat exactly does draft-only mean?

Draft-only means a system does all the preparatory work but never makes the final call. The post is written, the email is ready, the proposal is built. Exactly one action is missing: sending. That action stays with a human, every time, with no override button.

The difference looks small. One button more or less. In practice it's the difference between a system that works ahead of you and a system that speaks on your behalf without you knowing. In the first case you save on preparation and keep control. In the second case your name is attached to texts you never saw.

02 / the measurementDoesn't every agency say this by now?

Yes. We measured it on 26 July 2026 by literally asking an answer engine why AI agencies don't publish automatically. The composite answer came out to: everything comes in as a draft, a human makes the publication decision. Eight agencies in the source list. We weren't among them.

That was more instructive than a citation would have been. The principle isn't the distinguishing factor anymore — it has become the category standard. Everyone says a human decides, the same way every contractor says he works neatly.

A principle everyone states out loud says nothing anymore. What it's worth shows in where it's actually built in.

Because there's a difference between an attitude and a boundary. An attitude is a sentence on a website. A boundary is something that stops the machine, even when there's a rush, even when it would have been more convenient, even when nobody would notice. The first costs an afternoon of copywriting. The second costs a design decision you can't undo afterward without rebuilding it.

And that's exactly where it chafes in this category. Automatic publishing sits here on wish lists — as a feature still to come, as something clients ask for and that gets delivered "later." Whoever has it as a wish on their roadmap hasn't built the boundary. They've postponed it.

03 / the buildSo how is that gate actually built?

With keys, not agreements. The roles that do the work run on access that can read and prepare drafts. Publishing requires a different key, and that one doesn't sit in the machine. It sits with the human responsible for what goes out.

That's how it's built at our own place. We're customer number one of our own system, and that's not a marketing line but the reason we can be concrete about this:

The nice thing about this setup is you can verify it without taking our word for it. Ask which permissions the automation has. Ask whether publishing runs on the same key as producing. A provider who doesn't have a sharp answer to that hasn't built it this way.

This post itself came in that same way. Written by a system, placed as a draft, read and approved by a human before you could read it. If it had gone differently, you wouldn't have seen this.

04 / the boundarySo what is the system allowed to do on its own?

Everything without judgment and without tone. An appointment confirmation, a reminder, a receipt confirmation, a status change in your own admin. Messages that contain nothing anyone could read as your opinion. Those go out the door on their own, and that's how it should be.

The boundary lies where the content says something about you. Content, sales, advice, a reply to a human. Anything where a reader thinks: this business owner thought or wrote this themselves. A human belongs in that loop, no matter how good the system gets.

That boundary is deliberately drawn wider than strictly necessary. It would save time to let even the standard replies go out on their own too. We don't do it, because the exception wrecks the pattern: the moment one category exists that goes out without a human, the question is no longer whether something slips past the gate but when.

05 / the catchDoesn't approving eat the time you meant to save?

No, and that's the math that's usually skipped. Writing a good post yourself takes an hour. Reviewing a well-prepared post takes a few minutes. The saving is in the preparation, not in the button. Whoever tries to save on the button is saving on the wrong part.

But there is a real catch, and it's not this one. The catch is that approving is work that comes back every day. Not much work, but daily. Whoever puts it off builds up a queue and then experiences the system as pressure instead of help. A fixed moment each day solves that. No fixed moment means a pile.

Against that stands something you only get by looking. Every correction you make is the input that sharpens the system's grasp of your way of writing. Adjust the same phrasing three times, and that adjustment ought to become the default. Approving isn't the leftover task at the end, then. It's the point where your judgment enters the system.

06 / usableWhat should you ask any provider about this?

Four questions that expose the difference between attitude and build. They work just as well on us as on anyone else, and that's the point.

• Does publishing run on the same key as producing? One key for both means the boundary is an agreement, not a boundary.
• Is automatic publishing on your roadmap? A yes means the principle is negotiable the moment a client pushes hard enough.
• Which messages already go out the door without me right now? If the answer is longer than confirmations and reminders, the boundary is drawn somewhere other than you thought.
• What happens if I don't approve for a week? A system that sends anyway doesn't have the gate.

We mark what came about with AI support, and we arrange the disclosure for you and your team. Not because we know the legal outcome — that's work for a lawyer, not for us — but because for every piece of communication you need to be able to show who approved it. At the bottom of this article is exactly such a line.

This text was produced with AI support and reviewed and approved by a human before publication.