ChatGPT produced a usable client email on the first try. Lukas keeps testing anyway: what does a style brief built from your own writing samples change, does the reply hold up when a client asks for free extra work, and can a German sample produce an English email?
The starting point is unusual for a prompt video: the first answer was already good. That is not down to style, it is down to the facts being in the prompt. Availability, deadline and the open pricing question were all supplied, so ChatGPT had nothing to invent.
Style comes second. From three sample emails ChatGPT extracts a short style brief, and the second version of the same request comes back shorter, more specific and answers the pricing question more directly. It is a single comparison, not proof, but the second draft is clearly the better one.
The more interesting part is the awkward request: a client wants a 20 minute training video included in the existing fee. The draft stays friendly and still holds the boundary. What it does not remove is your final check, because a single phrase can commit you faster than the tone suggests.
| Round | Input | Result | My verdict |
|---|---|---|---|
| Plain prompt | The workshop facts, no style instruction | Confirms Friday, asks for the count by Tuesday, leaves pricing open | Already sendable, a little formal, one sentence repeats |
| Building the style brief | Three sample emails from three different situations | Observable features: formality, sentence length, how requests are phrased, openings and closings | Accurate, but you still have to read it yourself |
| Plain prompt plus brief | Same facts, same chat, now with the style brief | Shorter, names the six additional participants, answers the pricing question more directly | Clearly my pick of the two |
| The awkward request | Client wants a 20 minute video inside the existing fee | Boundary held, no price invented, no delivery date promised | Good, but one phrase commits too early |
| Language test | One fictional German sample, English as the target language | Usable English, very close to version two | Not a perfect style match, but not a blocker either |
This prompt turns your own sample emails into a short, reusable style brief. You then supply the facts of the actual email separately.
Analyze the email examples below. Describe their writing style: formality, sentence length, greeting, sign-off, directness, and how requests and boundaries are expressed. Give me a short reusable style brief. Separate recurring style habits from one-off details. Do not carry names, dates, or commitments into the style rules.
Not always. In this test the plain prompt already produced a usable reply, because the facts were supplied properly. Style samples made the second draft more specific and more consistent, not more correct.
Observable features only: how formal the writing is, sentence length, how requests are phrased, how emails open and close. Keep names, dates and commitments out of it, otherwise a deadline from one sample turns up in an unrelated reply.
In this test, yes. One fictional German sample was enough to produce a usable English reply. It was not a perfect style match, and one sample is thin evidence, but the language difference did not get in the way.
No. The boundary came from the facts, not from the style. The draft kept the boundary, invented no price and promised no delivery date, because the prompt stated what was inside the agreed scope.
Three passes. Are the facts correct? Does it make the right commitments? Does it sound like you? The second one matters most, because a phrase like before I scope it and provide a quote commits you to a step you have not decided on.
Three was enough here, and the important part is that they came from different situations: arranging a workshop, explaining a delay, and setting a boundary. If every sample does the same job, you cannot tell which features belong to your style and which belong to that task.
This page currently uses an editorial chapter summary. The subtitle files are prepared separately for YouTube.
Lukas sets out what this is not: a video where the first answer gets talked down so a magic prompt can save it. The first email was usable. The question is whether it can become more specific, more consistent and closer to a style you would actually send. Every email and the client are fictional and were used unchanged. So the test checks whether AI can follow a chosen writing style, not whether it has learned his personal voice.
Alex is organising a workshop. The client wants to move it to Friday, add six people and asks whether the price stays. Alex is free on Friday, needs the final count by Tuesday, and the price depends on scope. Those three points go into the prompt as facts. The most important step therefore happens before style: paste only the incoming email and say reply to this, and the answer has no idea about your availability, your deadline or the pricing agreement. A polished sentence does not fix missing information.
The first answer confirms Friday, asks for the final count by Tuesday and explains that additional participants may affect scope and price. It invents no number and offers no discount. Two small things stand out: the phrase any commercial change reads formal, and one sentence repeats points already made. Neither makes it a bad email, and a few edits would make it ready to send.
In a separate chat, ChatGPT gets three sample emails: one arranges a workshop, one explains a delay, one sets a boundary around extra work. Different situations are deliberate, otherwise you cannot separate what belongs to the style from what belongs to the task. Names, dates and commitments are explicitly kept out. The result matches the samples: professional but relaxed, short to medium sentences, clear and matter of fact without being blunt, a simple hi and a thanks at the end. But just because the AI describes a pattern does not mean it should apply to every email, so the brief deserves one careful read.
The second version opens straight with Friday works for me, shorter and enough for the situation. This time it names the six additional participants and closes with the condition before the price can be confirmed. Shorter overall, and it picks up the thanks sign-off from the samples. The detail about the six people was already in the request, so it makes the email more specific without inventing anything. It proves little though: both versions had the same information, the prompts differ slightly, and the same prompt sent twice gives different quality.
The client asks for a 20 minute training video after the workshop, included in the existing fee, because it should be quick. Anyone who freelances knows this request. The draft opens by saying the video can be discussed but would be a separate piece of work, explains the agreed scope and asks who the video is for and by when. No free work, no invented price, no promised delivery date.
The phrase before I scope it and provide a quote sounds like the decision to take the next step has already been made. More open would be: to work out whether I can help. That is the review worth doing. Read the draft for what it commits you to, not only for how it sounds.
Since most real emails here are in German, the question matters in practice: do the samples have to be in the language of the final email? A fresh chat, one fictional German sample, the workshop facts supplied separately, English as the target. Out comes a usable English reply, very close to version two. Not a perfect style match, and one sample is thin evidence, but the language difference was not a blocker.
Start with the facts: what can you confirm, what do you need to ask, what must you avoid promising? If consistency then matters, add a few writing samples and pull a short style brief from them, one that you read yourself. Finally the draft in three passes: are the facts right, are the commitments right, does it sound like you?
The plain prompt was already good in this test, not least because ChatGPT and Claude with memory switched on pick up your writing style anyway. The samples helped explore a more consistent style, and the second draft was clearly the better one. What they do not replace is your judgment about what the email should actually say. That is what makes the result yours.
I work 1:1 with freelancers, consultants, coaches and small teams on practical AI workflows, automation and custom tools.
Book a free strategy call