Building a Personal Style Reference for Your AI Prompts

A style reference is a document that tells the AI model what you sound like before you ask it to write anything. Without it, the model defaults to the statistical average of the writing it was trained on, which is a lot of content marketing from 2018 to 2022. The output sounds like that.

A style reference changes the starting point. The model writes toward your voice rather than toward the median. The editing pass becomes smaller because there is less to correct.

What a style reference contains

The style reference is not a list of rules. Rules like "be concise" and "write clearly" are too vague to change model behavior. The reference needs to be specific enough to be actionable.

Sentence length patterns. Pull three to five paragraphs from your best work. Note whether your sentences tend to be short and declarative, long and subordinated, or mixed. Note whether you use single-sentence paragraphs for emphasis. The model can imitate patterns it can see in examples better than patterns described in abstractions.

What the voice is not. This is the part that reduces hedging and generic register more than any other element of the reference. "The voice is not academic. It does not use phrases like 'it is important to note.' It does not present balanced views on questions where there is a right answer." Negative constraints prune more effectively than positive ones.

Topic handling. How do you typically open a piece? With a problem statement, an example, a counterintuitive claim? How do you handle transitions between sections? Do you use explicit transitions or let the logic carry without naming it? Do you include specific examples and, if so, what kind?

Register. What does the reader know? What does the reader want? What does the voice assume about the reader? "Writes to a reader who is already doing the work and does not need to be convinced it matters" tells the model something specific about how to pitch the level of detail.

What a style reference does not contain

It does not contain your full writing samples. Samples are useful in the system prompt to give the model concrete examples, but the style reference is a description, not a sample library. The description gives the model a frame for interpreting the samples.

It does not contain editorial preferences that are really just quality standards. "Write accurately" and "do not make things up" are not style preferences. They are baseline expectations. The style reference covers what makes your work sound like you, not what makes it acceptable.

How to build it

Take five pieces you have written that you think represent your voice well. Read them and note patterns. What words do you use that you would not expect in anyone else's work? What structures recur? What do you never do?

Write those patterns down in a page or less. Read it back. If it could describe any competent content writer, it is not specific enough. It should be specific enough that someone who has never read your work could produce something recognizable as yours.

Put the style reference in your system prompt, above the individual piece briefs. Every conversation inherits the reference automatically.

The maintenance problem

Style evolves. A reference built from work you did a year ago may describe a version of your voice that has developed since then. Read the reference every quarter and check it against recent work. Adjust where the description has drifted from what you are actually producing.

The gap between a stale style reference and your current voice is exactly the gap that appears in AI output and requires correction. Keeping the reference current reduces that gap. Not to zero, but enough to make the editing pass faster and the output more immediately usable.