How to Make AI Writing Sound Like You
A text can be well written and still sound like nobody. People who know you notice right away. Here is how to spot AI-written text and how to make your emails, proposals and posts sound like you anyway.
The short version
- AI sounds like AI when it has to guess. Then it fills the gaps with the average.
- What matters most are texts you wrote yourself. They show the AI how you typically write.
- Start with your own rough draft, dictated or in bullet points, instead of an empty prompt.
- Read it yourself at the end: Are the facts right, are the commitments right, does it sound like you?
How to spot AI-written text
Three things stand out to me. The texts are built the same way every time. They sound generic, as if no human were behind them. And the same stock phrases keep coming back. Above all, they don't sound like your own emails. People who know you notice.
Wikipedia editors who hunt for AI-generated text have collected typical patterns in their own field guide. They wrote it for Wikipedia articles and note that some signs may not apply elsewhere. Among the patterns:
- Small details puffed up into a “pivotal moment” or “a testament” to something bigger.
- Ad-copy language, even where it does not belong: “vibrant”, “groundbreaking”, “showcasing”.
- Vague attributions such as “Experts argue” or “Industry reports”, with no name behind them.
- Contrasts that read as if they correct a misconception nobody had: “Not only … but …” or “It is not just …, it's …”.
- The rule of three, over and over: three adjectives or three short phrases in a row.
- A stock ending along the lines of “Despite these challenges …”.
- Heavy boldface, lists that start every line with a bold label, and emoji in front of headings or bullet points (less common now).
The em dash has long been a notorious tell, too. The guide now lists heavy use of it among its historical signs: the newest chatbots are increasingly built to avoid it altogether. I cut it from my own texts as a rule.
Important: No single sign proves anything. People write like this, too, and the guide itself says not all text with these signs is AI-generated. It also warns against relying solely on AI detection tools, which it says have “non-trivial error rates”, and against relying too much on your own judgment. For your texts, something else counts anyway: they should sound like you.
Why AI sounds like AI
An AI writes its answer piece by piece, token by token, and picks whatever is most likely to come next based on its training. If it lacks your facts and your examples, it lands on the average of all the texts it knows. That is exactly what you then read: smooth, polite, interchangeable.
Wikipedia's guide describes the same effect: the output tends to “regress to the mean”, and specific details get replaced by generic, positive descriptions.
A cleverer magic prompt won't help much here. What helps: letting the AI guess less.
How to make it sound like you
This is how I work with AI myself when I write emails, proposals or posts:
Give the AI texts you wrote yourself
For me this is the most important step. In real emails, proposals or posts of your own, the AI can see how you typically write: how you open, how long your sentences are, how directly you phrase requests. Ideally, give it three texts from different situations.
Start with your own rough draft
I often just dictate or jot down bullet points. The AI turns that into a clean text, but the thoughts, facts and order come from me. Clarify beforehand: What can you commit to, what do you need to ask, what must you not promise? An empty prompt like “Write an email to …” gets you the average instead.
Save your style
So you don't have to explain it every time, turn your examples into a short style brief and save it as a standing instruction, for example in a project or in your tool's saved instructions. You'll find the prompt for that right below.
Read it yourself before it goes out
Are the facts right? Are the commitments right? Does it sound like you? The second question matters most: a single phrase can commit you to something you haven't decided yet. And anything that could just as well appear in any other text gets cut.
The prompt for your style brief
Paste in three of your own texts and have the AI pull out the style rules. You add the facts of the new email separately afterwards:
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.
Names, dates and commitments don't belong in the brief; otherwise, a deadline from one example suddenly turns up in an unrelated reply. Read the brief yourself once before you save it.
For emails, proposals and posts
What to give the AI, depending on the text:
- Emails: the facts, meaning date, deadline and open points, plus two or three real emails of yours as a template.
- Proposals: service, scope, price, timeline and what is not included. Your earlier proposals serve as the template. When you check the draft, the commitments count double.
- Posts: your point, one concrete example from your work, and posts you wrote yourself.
A quick test
In a video I tested this with ChatGPT on fictional client emails. Even the first answer was usable, because the facts were in the prompt. With a style brief built from three sample emails, the second version came out shorter and more specific. And a German sample was enough for a usable English email.
Frequently asked questions
By patterns: a template-like structure, generic wording, the same stock phrases, and contrasts like “not only … but …”. None of them proves anything, because people write like this, too. The clearest signal comes from people who know you: the text doesn't sound like you.
With texts you wrote yourself. Three examples from different situations are enough for a style brief. Save it as a standing instruction, for example in a project, so you don't have to explain it every time.
By letting the AI guess less: your own rough draft, real facts, your own texts as a template. At the end you read it yourself and cut anything that could just as well appear in any other text.
Yes, if you supply the facts: service, scope, price, timeline and what is not included. Use earlier proposals as a template and check the draft above all for what it commits you to.
Not as proof. Wikipedia's guide to spotting AI writing warns against relying solely on AI detection tools, which it says have non-trivial error rates. More important than any detector result is whether the text sounds like you.
Sources
Retrieved on October 10, 2026.
Related
Want texts that sound like you, without starting from scratch every time?
I work 1:1 with freelancers, consultants, coaches and small teams on their real tasks. The intro call is free, takes 15 minutes and is not a sales pitch.
Book Free Consultation