Style Guides and Prompt Libraries

Introduction

AI output tends toward the generic. The cure is not cleverer one-off prompts. It is clear standards that you write down once and use every time: a style guide, tested prompts and a scoring rubric. This also makes teamwork, and switching tools, much easier.

The one-page style guide

A useful guide is short and concrete. Include:

Audience: who reads this, and what they already know
Voice traits: for example "plain, warm, practical, a little dry humour"
Reading level and sentence length: "short sentences, average reader of about age 14"
Words we use and avoid: "use 'start', not 'commence'; avoid hype such as 'revolutionary'"
Structure habits: answer first, short paragraphs, descriptive headings
Formatting: headings, lists, bold, numbers and units, dates, spelling variant (UK or US)
How we treat claims: sources, numbers, caveats, no guarantees
Inclusion and accessibility: respectful language, alt text, captions
Examples: two paragraphs that sound right, and two that do not, with a note on why

Feed the guide to AI tools as part of the instructions, and give examples of good writing.

Quick quiz

Why include examples of good and bad writing in a style guide?

A prompt library

A prompt library is a shared document of prompts that have worked, with notes. Each entry has:

Name and purpose: "Outline a how-to guide"
When to use it
The prompt text, with placeholders in square brackets
Required inputs: audience, sources, key points
An example of good output
Known problems and how to fix them
Version and date, and the tool or model it was tested on

An example entry:

Name: Draft one section from notes
Use when: you have verified notes and want a first draft in our voice.

Prompt:
You are helping write for [audience]. Follow the style guide below.
Write the section "[heading]" in about [length] words.
Use ONLY the facts in the notes. If something is missing, say so
instead of inventing it. Do not add statistics.

Style guide: [paste]
Notes: [paste]

Prompt libraries save time, and keep quality consistent. Review them every few months, because tools change.

Scoring with a rubric

A rubric turns "this feels off" into something checkable. Score each draft from 1 to 5 on criteria such as:

Criterion	Question
Accuracy	Is every claim verified and sourced?
Usefulness	Does it answer the reader's question?
Originality	Does it add our experience or examples?
Voice	Does it sound like our style guide?
Clarity	Is it easy to read and well structured?
Safety and compliance	Are claims, disclosures and rights in order?

Set a minimum standard, such as no criterion below 3, before publishing. Use the same rubric for AI-assisted and human-only work.

Test and improve prompts

Choose five typical tasks.
Run the prompt, and score the results with the rubric.
Change one thing in the prompt.
Run again and compare.
Save the better version and note why.

Keep failures too. They show what to avoid.

Keep your knowledge, not only your prompts

Store the style guide, prompts, rubric and examples in your own documents, not only inside a particular AI tool. That way you can switch tools or share them with colleagues.

Try it yourself

Write a one-page style guide for your project, with four example sentences. Create three prompt-library entries for your most common tasks, and a six-line rubric. Test one entry on three inputs.

Summary

A short, concrete style guide with examples keeps voice consistent across people and tools.
A prompt library with tested, versioned entries saves time and raises quality.
Score drafts against a rubric, improve prompts one change at a time, and keep everything in your own files.