Clean User-Facing Text
A final text-hygiene pass for authorized prose — audits and strips invisible Unicode artifacts and homoglyphs, then reduces mechanical AI-writing signals with a measure-before/measure-after check, while preserving every fact and the writer's actual voice.
Copy-pasted AI output can carry invisible Unicode characters that look identical to a normal space or letter but aren't — a hygiene problem entirely separate from writing style, and one that most 'sound less like AI' guides never even check for.
Who it's for: writers and editors finalizing AI-assisted articles, manuscripts, and reports they're authorized to polish, documentation teams cleaning up prose before publishing, anyone who's had AI-generated text carry stray Unicode artifacts into a final document, teams wanting an honest, non-evasion-framed approach to reducing AI writing tells
Example
"Clean up this draft before I publish it" → A deterministic pass stripping invisible Unicode and homoglyphs first, a before/after AI-cadence density check that skips the rewrite entirely if the text already reads naturally, a single rewrite pass that kills stock AI vocabulary and injects real sentence-length variation without inventing a single new fact, and an honest report distinguishing what was verifiably fixed from what's only a best-effort stylistic pass
CLAUDE.md Template
New here? 3-minute setup guide → | Already set up? Copy the template below.
npx ccplaybooks add clean-user-facing-text# Clean User-Facing Text
A final text-hygiene pass for prose you own or are authorized to process: audit for suspicious invisible Unicode, then rewrite to reduce mechanical AI-writing signals — while preserving every fact, claim, and the writer's actual voice. Two honest boundaries up front: Unicode cleanup is deterministic; statistical-signal reduction is best-effort. Never claim a rewrite proves human authorship or is undetectable, and never use this for undisclosed-authorship evasion (academic integrity submissions, and similar).
## When to Use
Use when asked to clean, humanize, polish, or finalize articles, manuscripts, reports, documentation, emails, product copy, UI text, Markdown, or HTML prose. Don't use for code-only tasks, and don't use to help someone misrepresent AI-assisted work as undisclosed solo human authorship where that matters (an academic submission under an integrity policy, for instance).
## Workflow
### 1. Identify the Prose
Get new playbooks like this one
One email a week with new Claude Code workflows. Free, like everything here.
No spam. Unsubscribe anytime.
What This Does
A two-layer text-hygiene pass distinct from a typical "sound less like AI" rewrite. Layer A is fully deterministic: strip invisible Unicode characters and homoglyphs — artifacts that look identical to a normal space or letter but aren't, and that often ride along silently when text is copy-pasted from certain tools. Layer B is a measured, honest rewrite pass: score the text's AI-cadence density first, skip the rewrite entirely if it's already low or medium, and only when density is genuinely high, kill stock AI vocabulary, inject real sentence-length variation, and flatten formulaic structure — all while a hard rule holds throughout: never invent a detail, name, number, quote, or source to make the prose easier to write or more varied. A missing fact gets flagged, never filled.
The whole approach is explicit about its own limits: Unicode cleanup is deterministic and verifiable; statistical-signal reduction is best-effort against perplexity/burstiness-style detectors specifically, and the skill refuses to claim detector evasion, proof of human authorship, or removal of any vendor's secret-key watermark. It's built for text the user owns or is authorized to process, explicitly not for disguising AI involvement where authorship disclosure genuinely matters.
Quick Start
Step 1: Create a Project Folder
mkdir clean-text && cd clean-text
Step 2: Download the Template
Click Download above, then:
mv ~/Downloads/CLAUDE.md ./
Step 3: Clean a Draft
claude
Paste in the prose you want finalized and ask Claude to clean it up. It will strip invisible artifacts, measure AI-cadence density, and rewrite only if the density check says it's actually needed — preserving every fact and the writer's own voice throughout.
Tips & Best Practices
- Run the artifact-stripping pass unconditionally, even on text that scores low on AI-density — invisible Unicode characters are a real, separate hygiene problem from writing style, and they're free to fix.
- The measure-before step matters as much as the rewrite itself — rewriting text that's already fine risks introducing errors or flattening a deliberate voice for zero real benefit.
- When using a voice sample, confirm it's actually owned or authorized before applying it — this preserves the line between "sound like my own established voice" and impersonating someone else's.
Limitations
- Only for text the user owns or is authorized to process — never use to help disguise AI involvement where authorship disclosure genuinely matters (a graded academic submission, a professional certification, and similar contexts).
- The statistical-detector levers are honestly limited to perplexity/burstiness-style detectors — trained neural classifiers are adversarially trained against exactly this kind of edit, and no claim here should be read as guaranteed evasion of any specific tool.
- Never a substitute for actually disclosing AI assistance where a policy or relationship requires it — this tool improves prose quality and strips accidental artifacts; it makes no authenticity claims on its own.