Skill: Editorial Voice Configuration

This is a skill file, not a one-off prompt. Load it as context before any content creation task to ensure the AI writes in a consistent voice. Think of it as a portable style guide that travels with you across conversations.

Works in: Claude Projects (as a project instruction), custom GPTs (as a knowledge file), system prompts (paste the content), AI coding tools like Cursor or Windsurf (as a skill file), or any tool that accepts persistent context.

Use it for: blog posts, social media posts, newsletters, marketing copy, documentation, emails, or any content where voice consistency matters.

Skills differ from prompts in persistence and scope. A prompt is a single instruction for a single task. A skill is a reusable configuration that shapes all tasks within a session. This editorial voice skill does not tell the AI what to write. It tells the AI how to write, regardless of the task.

The banned patterns list is the most effective section. LLMs have strong default tendencies. They "delve into" things, they "harness the power" of tools, they love em dashes. These patterns are so consistent that they serve as AI detection signals. Banning them forces the model to find better alternatives, which makes the output sound more human.

Sentence structure rules create rhythm. Professional writing has cadence. Short sentences create urgency. Longer sentences provide context. The rules here create that variation automatically. Without them, AI output tends toward uniform sentence length, which reads as flat and monotonous.

"Knowledgeable peer, not teacher or salesperson" is the most important line. It sets the fundamental relationship between writer and reader. "Teacher" voice is condescending. "Salesperson" voice is pushy. "Peer" voice is respectful and direct. This single instruction shifts the entire tone.

Replace the banned patterns with your own. The list provided targets common AI writing patterns. But every writer has personal pet hates. Add yours. If "synergy" makes you cringe, ban it. If your brand uses "leverage" deliberately, remove it from the ban list.

Adjust the tone. The default voice is direct and slightly informal. For a more formal brand, change "Fragments are fine" to "Use complete sentences" and "Dry wit is welcome" to "Maintain a professional, measured tone throughout."

Add brand-specific vocabulary. If your company has preferred terms (e.g. "clients" not "customers", "platform" not "tool"), add a "Preferred Vocabulary" section with a substitution table.

Add audience context. The skill works broadly but becomes more effective with audience specificity. Add: "Primary audience: [description]. They care about [X, Y, Z]. They are already familiar with [concepts]. Do not explain [basic things]."

A voice configuration sets the floor, not the ceiling. It prevents the worst AI writing habits but does not guarantee great writing. The quality of the output still depends on the quality of the task instruction.

Loading this skill consumes context window space. On models with smaller context windows, this may reduce the space available for the actual task. For quick, informal tasks, the overhead may not be worth it. For published content, it always is.

The banned patterns list needs periodic updating. As models improve, some patterns become less common and new ones emerge. Review the list quarterly and adjust based on what you are actually seeing in outputs.

Claude: Follows skill files exceptionally well, especially when loaded as project instructions. The banned patterns are respected consistently. Claude also follows the sentence structure rules (length limits, paragraph limits) more reliably than other models.

GPT: Respects the skill when loaded as a Custom GPT instruction. May occasionally slip on banned patterns in longer outputs. If this happens, add: "Adherence to the banned patterns list is mandatory, not aspirational."

Gemini: Works but may require the skill content to be reinforced in individual task prompts for longer sessions. Gemini's context attention can drift over very long conversations.