AI Private Coach
Home · Use cases · Content production with AI
For creatives and freelancers · Implementation

Create content with AI that sounds like you

Anyone with a recognisable voice loses it the moment they trade it for a smooth average.

In short: The problem with AI texts isn't that they're bad. They're mediocre-good — and that's where the damage lies. The way out isn't to give up AI, but to align it with your own voice before letting it write. That takes ten of your own texts and an analysis concrete enough for a model to hit the mark.

How it works today

  • Blank document, blinking cursor, three attempts, none of them lands — 60 min
  • Or: ask ChatGPT, get a text, can't use it — 20 min lost
  • Last week's good idea never became a post
  • Between two client projects, your own marketing is the first thing dropped

How it works afterwards

  1. Capture the thought as a voice note or three keywords — 2 min
  2. A draft in your structure, your rhythm, your words — 1 min
  3. Rewrite what doesn't sit, cut what's too much — 15 min

What we actually build

1. Take your voice apart

We take ten to fifteen texts you genuinely like and analyse what makes them yours. Sentence length, rhythm, how you open paragraphs, which words you never use. The result is a written description of your voice concrete enough for a model to hit — and one that teaches you something about your own writing along the way.

2. Build a voice profile that stays

From the analysis comes a fixed profile that runs with every job: allowed and forbidden phrasings, typical structures, examples. Not a prompt you retype every time — a saved foundation.

3. From raw material to draft

Most creatives don't have an idea problem, they have a writing-it-down problem. We build the path from your voice note, workshop scribble or three keywords to a first draft — in several formats from one piece of source material.

4. Draw the line

We define together where AI contributes and where it doesn't. For most creatives, the raw material and the final pass are off-limits, the in-between is open. Drawing that line deliberately is the difference between a tool and a slow dilution.

Tools — and where it doesn't help

For texts with a distinct voice, Claude generally delivers better results than the alternatives — it holds style constraints more stably across longer texts. For research and current references, other tools are added. Which one for what, we decide on your material, not on a ranking from the internet.

Where it doesn't help: the first thought. What you have to say comes from you — AI can shape it, not invent it. If you have nothing to say, AI just helps you produce nothing faster.

Your texts, your rights

Anyone feeding their own texts into a model should know whether they're used for training — consumer tiers partly yes, business tiers generally no. Client projects add another layer: does the job contain confidential client information? And what does your contract say about it? We settle this before the first job runs through.

→ How we bring law and AI together

What this means in time

Realistically, 50 to 60 percent of the time per piece — from 90 down to 35 minutes is typical. The bigger effect is frequency: someone who needs ninety minutes per post writes one a month. Someone who needs 35 minutes writes one a week. That's not a time saving — that's a different operation.

Frequently asked questions

Do readers notice AI was involved?

With smooth standard output, instantly — uniform sentence lengths, symmetrical lists, a certain stock of words you learn to spot quickly. With a text that follows your voice profile and that you've reworked, generally not. The effort sits exactly in that difference.

Will I lose my own writing ability over time?

A fair concern, and I take it seriously. My experience: those who use AI for the rough draft and rework it themselves write sharper after six months — because constantly editing someone else's phrasing trains your own judgement. Those who publish unchecked really do lose the ability.

How many of my own texts does the voice profile need?

Ten good ones are enough. Selection matters more than volume: texts you'd happily write the same way again. Feeding in fifty mixed texts gives you an average — and the average is exactly what we're trying to avoid.

Does this work for German and English in parallel?

Yes, but it takes two profiles. Your voice is a different one in each language — most people are looser in English and more precise in German. A shared profile for both delivers a mediocre result in both.

Step 3 of our approach · Implementation

Bring ten texts you genuinely like

In the free intro call I'll show you, on your own material, what a voice profile makes visible. No sales pitch.

1:1 coaching from CHF 320Flat hourly rate CHF 180/h · online or on-site in Basel
Book a free call — 20 min All use cases

Related use cases