Brand Voice Linter

Why this exists

Em dashes are the most famous AI writing tell, but they are not the worst one. The worst ones are the words: "cutting-edge", "seamless", "unlock potential", "in today's fast-paced world". Copy full of them sounds like everyone and convinces no one. After 20 years writing for brands, I can tell you the pattern: weak copy hedges, buzzes, and exclaims. Strong copy states, counts, and stops.

The Em Dash Remover fixes one rule. This is the full editor. It reads your copy against a voice guide, the same kind of writing rules I keep for my own brand and apply to every client document, and reports every violation with a concrete rewrite. Your guide, your rules: the skill ships with a strong default set and you edit it to match your brand.

How it works

1. It loads your voice guide A voice-guide.md in your project root defines what is banned and what gets flagged. No guide yet? It creates one from a strong starter template you then edit to match your brand.
2. It scans every rule against your copy Em dashes, buzzword clusters, hedge language, vague superlatives, passive voice, exclamation marks, AI-tell phrases, and sentences running past 30 words. Code blocks, URLs, and real quotes are left alone.
3. It reports with rewrites, not just complaints Every violation comes with line number, severity, and a suggested replacement. "Industry-leading support" gets "replies within 4 hours (use your real number)", not just a red flag.
4. It fixes only when you say so Report-only by default. Say "fix it" and banned items are corrected in place, while fixes that need a real fact (your actual number, your actual outcome) come back as open questions instead of invented claims.

Step by step (for first-time users)

1. Write or collect the copy Landing page, proposal, About page, email sequence. Any text file works.
2. Open Claude Code in that project Navigate to the folder where the copy lives and launch Claude Code.
3. Type /brand-voice-linter path/to/copy.md First run, it offers to create your voice-guide.md. Accept it, then open the guide and make it yours: add your banned words, relax what does not apply.
4. Read the violation report A table per file: line, rule, what was found, the suggested rewrite, and severity. Plus a verdict: reads clean, light pass needed, or needs a rewrite.
5. Say "fix it" when you agree Banned items get corrected in place. Anything needing a real fact comes back to you as a question, because the linter never invents your numbers.

What it catches

Honest take

What it does well: Consistency at the moment you actually need it, which is right before publishing, not during a quarterly brand review. The voice guide as a file is the real idea here: your writing rules stop living in your head (or in a PDF nobody opens) and become something Claude enforces on every draft. The severity split matters too. Banned items get fixed mechanically; flagged items get a suggestion plus the reason, because turning "world-class service" into a real claim needs a fact only you have.

What it does not do: It will not make weak positioning strong. If the copy has nothing concrete to say, the linter will strip the buzzwords and what remains will be honestly empty, which is useful information but not a rewrite. It also does not check grammar or spelling; plenty of tools do that already. And in fix mode it deliberately refuses to invent numbers, so expect open questions back, not magic.

When to use it: Before publishing anything with your name on it. Landing pages, proposals, case studies, LinkedIn posts, email sequences. Run it after any AI-assisted writing session: it catches what slipped through, including the tells you have stopped noticing.