Robotic AI tone
Removes polished but empty prose, mechanical transitions, repetitive sentence shapes, and language that sounds generated rather than written.
A free set of writing rules for Claude, ChatGPT, Gemini, Copilot, and more
Make AI writing sound precise, direct, and natural.
Clean Language removes the habits that make AI writing irritating: robotic tone, filler, fake enthusiasm, and repeated conclusions. It also cuts contrast framing, business jargon, excessive bullets, em and en dashes, and predictable vocabulary such as delve, unlock, and transformative.
Clean Language is a small file of written rules you add to your AI once. It tells your AI to keep your facts, figures, names, quotations, and technical terms exactly as you wrote them; when a style rule and the meaning conflict, the meaning wins.
Free. No account beyond the AI you already use. Pick your AI:
Set up in ClaudeDownload 1 file and upload it to Claude
Set up in ChatGPTDownload 1 file and upload it to ChatGPT
Set up in GeminiCreate a reusable Clean Language Gem
Set up in CopilotCreate a Clean Language agent
Detailed instructionsClick-by-click setup for every platform
View source on GitHubReview the rules, history, and licence
(opens in a new tab)
No dash theatreConvert every em and en dash to the punctuation the sentence needs; those long marks (— and –) are what AI leans on.
No fake contrastStop the repeated “not X, but Y” pattern.
No corporate fogReplace filler and vague jargon with the actual point.
Meaning over styleFacts, figures, obligations, and technical meaning outrank any style change.
What Clean Language fixes
The writing patterns people notice.
Contrast framing
Rewrites formulas such as “it is not X, it is Y,” “not because X, but because Y,” and “this is not merely X.”
Em dash and en dash overuse
Uses commas, colons, parentheses, or complete sentences where they communicate the idea more cleanly.
AI filler words
Challenges phrases such as “it is important to note,” “at its core,” “in today’s landscape,” “delve,” “tapestry,” “leverage,” and “unlock.”
Fake enthusiasm
Removes automatic praise, reassurance, validation, inspirational framing, and unsupported claims that something is critical or transformative.
Over-explaining
Cuts repeated conclusions, unnecessary summaries, obvious implications, long introductions, and paragraphs that announce what the next paragraph will say.
Bullet list abuse
Collapses the bullet staircase, where every sentence becomes a bullet without adding structure, and keeps connected reasoning in paragraphs.
Hidden ownership
Names the person, team, system, control, consequence, or required action when vague language would conceal responsibility.
Artificial balance
Stops forcing a positive counterpoint after every criticism. Findings remain clear unless mitigating evidence changes the conclusion.
Designed for real work
Writing that stays precise.
Clean Language focuses on meaning, clarity, accountability, and natural cadence. It improves the writing itself; it does not try to fool AI detectors.
It supports email, Teams, Slack, LinkedIn, reports, executive communication, technical documentation, governance material, policies, incident updates, and general prose.
Example prompt
Apply the Clean Language skill to this text.
Preserve every fact and technical term.
Remove filler, contrast framing,
dashes, and generic AI tone.This prompt presumes that Clean Language is installed. Set it up, or try it first without installing.
Before and after
The same information in fewer words.
Generic AI prose
It is important to note that the implementation requires additional testing before it can be considered fully ready for production. This is not merely a technical exercise, but a critical step that will help ensure a robust and seamless deployment.
Clean Language
The implementation requires additional testing before production deployment.
Questions
What Clean Language is for.
Is Clean Language an AI humanizer?
Clean Language makes generated text more natural by improving the writing itself. It instructs the AI to preserve correctness and does not target AI-detector scores.
Does it ban every em dash, en dash, adverb, or passive sentence?
For adverbs and passive sentences, no: constructions that carry precise meaning stay. Em and en dashes are converted to the punctuation the sentence needs, except inside quoted or verbatim material; numeric and date ranges keep a hyphen or “to”. You can still instruct it to prohibit dashes absolutely.
Can it clean writing produced by Claude, ChatGPT, Gemini, Copilot, or another model?
Yes. The rules work with any text-generating AI that accepts persistent instructions or supplied reference text. Use them as an installed skill, paste them as instructions, or follow them yourself as a checklist.
Is it suitable for technical and policy writing?
Yes. The precedence rules protect commands, identifiers, legal language, obligations, standards terminology, factual qualifications, and domain-specific vocabulary.
Is it free?
Yes. It is free for personal and commercial use, under the Creative Commons Attribution-ShareAlike 4.0 International licence.
What exactly am I adding to my AI?
A small skill file of Markdown instructions, configuration, icons, and the licence and notice files that your AI reads. It is not a program, nothing is installed on your computer, and it does not connect to anything. You can switch it off or remove it at any time, from the same place you added it; removing it changes nothing else in your account.
Why did Jeff Posluns create Clean Language?
Jeff Posluns, an IT and cybersecurity executive, created Clean Language after one too many AI-generated documents arrived copy-pasted and unread by their own senders. He wrote the first version in about an hour, drawing on his GRC Library (opens in a new tab) project (source on GitHub (opens in a new tab)), and it has been revised steadily since. It exists to make everyone's AI output a little less infuriating to read.
Can I use Oxford, Canadian, British, or American spelling?
Clean Language uses Oxford English spelling with -ize endings by default: colour, centre, organize. Canadian, British, and American each take one line pasted into your instructions. See the install guide for the exact line and a word-by-word comparison.
Stop publishing generic AI prose.
Set it up once, apply it to your next draft, and keep the meaning intact. It is free and takes a few minutes.
Pick your AI and set it up