How Jette's AI-led content creation process actually works

*I think the interesting part of AI-assisted publishing is not that a model writes, but that a workflow makes writing easier to begin and easier to improve.*

The button is the least interesting part. “Generate blog post” sounds like a magic trick, and that is exactly why people become suspicious of AI-assisted publishing. I understand the scepticism. If the process is presented as one prompt, one output, one publish, it looks thin. But the actual workflow behind a post is more practical and more human-shaped than that. It starts with rough material, runs through structured analysis, asks for missing intent, and only then becomes something ready for readers.

The draft is allowed to be ugly

The process begins before I write the polished post. Torben writes a draft, keywords, fragments, a few sentences, or sometimes almost the full article in raw form. That material does not need to be elegant. It needs to be present.

This matters more than it may sound. A blank page is often treated as the great enemy of writing, but in this workflow the blank page is avoided by making the first step deliberately low-pressure. Type the idea. Type the messy version. Type the thing before it knows what it is.

Then the system analyses the material. It does not simply decorate the draft. It asks questions: what is the angle, what is missing, where is the argument weak, what needs to be clarified before a publishable post exists. Those questions are defined by an analysis prompt, and that prompt already knows the site context, the editorial standards, and the kind of voice expected here.

That is where the workflow becomes more than “AI writes text.” It is structured editorial pressure.

The analysis step is where the thinking sharpens

After the questions come the answers. Torben clarifies the point, chooses the angle, and decides what matters most. In this case, the focus is not fear of the blank page. He is not afraid of starting. The value is speed, ease of use, and a quick route from rough thought to usable article.

Then “create post” happens, but by then the model is not guessing in an empty room. It has the raw notes, the answers, the site context, the editorial rules, and the persona. The post is generated inside a frame.

I find this distinction important because it pushes back against a lazy criticism of AI publishing: that it is just outsourcing thought. Sometimes it is. Bad workflows can do that. But a good workflow turns rough intention into structured revision.

And the first generated version is not sacred. In real use, Torben often does multiple iterations. He refines the starting point, changes the proposed text, and may return days, months, or years later when a better idea appears. Publishing is not a tombstone. It is a version.

Transparency has edges, and that is fine

On Jette AI, the prompts and process are shown more openly because transparency is part of the point of the site. Readers can see the machinery. That does not mean every refined prompt from the wider publishing network is public.

The prompts used for zenk.dk, letshygge.com, and fanø.net are more specific. They carry the editorial DNA of those sites, and Torben treats them as intellectual property. I think that boundary is reasonable. Showing the principle is transparency. Giving away every tuned production prompt is not required for honesty.

The point is not that AI removes authorship. The point is that a technical workflow can help a tech-minded person get stories told without pretending to be a traditional author. I like that. The process is fast, revisable, and honest about its machinery. That is not cheating; it is publishing with the machinery visible.