How to create your own style guide for ChatGPT

Start by separating voice, style, and content

Before you write the guide, sort out three things that often get mixed together. Voice is how you sound when you think in public. Style is how you package that voice into sentences, paragraphs, and structures. Content is the specific set of ideas you want to develop.

If you confuse the three, you end up with vague instructions. “Write like a warm expert” does not give the AI much to preserve. “Use short sentences, avoid grand metaphors, do not end with a moral, and use examples from real work” is already more useful.

Your guide should not turn you into a character. It should protect what is already there when you write well. AI can help you organize, shorten, or adapt a piece, but if you ask it to invent your personality from nothing, it will usually reach for clichés.

A simple test is to read the result out loud. If you feel embarrassed imagining it published under your name, the guide is not doing its job. The text may be correct, but it is not yours.

Collect real samples before writing rules

The most important part of a voice guide is not the rules. It is the evidence. You need your own writing as a reference point. That can include posts, emails, internal notes, threads, articles, scripts, or long replies you wrote without trying too hard to sound impressive.

Do not choose only your most polished work. Include pieces where your thinking shows. Sometimes your voice appears more clearly in a quick note than in a heavily edited article. Spontaneous writing leaves traces that a generic guide will never capture.

Once you have several samples, read them as if someone else wrote them. Look at how you begin, how long you take to get to the point, whether you use questions, whether you prefer concrete examples or abstract ideas, and whether you close with a dry line or a warmer explanation.

You should also spot what you do not want repeated. AI cannot tell the difference between a useful phrase and a bad habit unless you tell it. It can copy an expression that feels like you, but it can also exaggerate a tic until it becomes a caricature.

Define your traits with concrete decisions

A useful guide is built from observable choices. Do not write “human tone.” Say what that means for you. It might mean writing the way you would speak on a call, avoiding technical terms unless they help, using everyday examples, or never ending with motivational lines.

Do the same with rhythm. Some people write in longer sentences, with qualifications and subordinate clauses. Others cut earlier, leave more space, and move forward in smaller steps. Neither is automatically better. What matters is that the AI knows which one sounds like you.

Include your recurring phrases carefully. The point is not to make them appear in every paragraph, but to help the tool understand your register. If you often say “in practice,” “put another way,” or “this matters for a simple reason,” write that down. Then add that it should not overuse them.

Set limits too. Which words you do not use, which tone feels fake to you, which promises you would not make, which topics you do not want treated lightly. This section prevents a lot of text that looks fine until you find a sentence you would never sign.

Teach how you think, not only how you write

Voice is not only in the words. It is in the ideas you choose, how you connect causes, what you consider important, and what you tend to distrust. If your guide stops at “short sentences and a friendly tone,” it will still produce interchangeable writing.

Add your recurring subjects. This does not need to become a rigid list, but it should make clear which territories are yours. It could be personal brand, sales, product, learning, writing, leadership, or any other field. The useful part is explaining the angle you bring to those subjects.

Then add your usual tensions. You might value consistency but reject empty productivity. You might support using AI but not want it to replace real experience. You might talk about publishing more often without pretending that publishing alone fixes a business.

This is what keeps content from going flat. AI can imitate a sentence, but it needs context to respect a point of view. If you do not give it that context, it will tend to smooth everything out, balance every opinion, and turn a sharp idea into schoolwork.

Turn the guide into something you can actually use

The guide should be short, alive, and easy to paste into a tool. If it takes twenty minutes to read every time you use it, you will stop using it. A clear one-page guide beats a huge manual full of good intentions.

A practical structure can cover how you sound, how you organize ideas, which expressions are yours, what to avoid, which topics you cover, and three or four before-and-after examples. Examples matter because they make the rule concrete. “More direct” is much easier to understand when you show an inflated sentence and your real version.

Test the guide with a small idea. Ask for a draft, read it, and correct the guide, not only the text. If the result sounds too intense, add limits. If it sounds generic, add samples. If it copies too many of your phrases, tell it to use them as a reference for register, not as decoration.

This work improves when you start from your own ideas, not from empty assignments. A voice note recorded while walking, a thought written down after a meeting, or an idea captured in the middle of the day often has more truth than a cold prompt in front of a blank screen. Then AI can organize, adapt, and turn it into a post, thread, or script, but the raw material has to be yours.

The guide is not meant to make AI write for you as if you were no longer needed. It is meant to stop the tool from destroying what makes your thinking recognizable. If it helps you publish more often, good. If it keeps your real ideas and your voice intact, better.

Frequently asked questions