Clean Language

Clean Language is a cross-platform editorial standard for AI writing. It supports drafting, rewriting, proofreading, adapting, and reviewing precise, direct, natural prose.

It removes common AI-writing patterns such as robotic tone, filler, contrast framing, unnecessary em dashes and en dashes, fake enthusiasm, repeated conclusions, business jargon, and vague claims while preserving factual, technical, legal, policy, contractual, and standards meaning.

Detailed installation instructions · Project website · Repository installation guide

What it does

• Leads with the answer, decision, request, finding, or required action.
• Produces concise, information-dense Oxford English.
• Removes generic AI language without imposing arbitrary grammatical bans.
• Preserves names, dates, figures, commitments, qualifications, quotations, and domain terminology.
• Protects legal, technical, contractual, policy, and standards meaning.
• Supports executive communication, email, Teams, Slack, LinkedIn, reports, policies, incident updates, technical documentation, governance material, and general prose.
• Distinguishes verified facts, reasonable inferences, estimates, and speculation.
• Avoids fake enthusiasm, unrequested reassurance, ceremonial praise, and generic engagement prompts.

Supported platforms

Clean Language uses one canonical skill definition for Claude, ChatGPT, Gemini, Copilot, and other compatible AI systems. The core skill lives in clean-language/.

• Claude: download the packaged Skill, upload it through Claude's Skills interface, and enable it.
• ChatGPT: create or upload a Skill using the Clean Language instructions.
• Gemini: create a reusable Gem or install the Agent Skill through Gemini CLI.
• Copilot: create a Copilot agent with Agent Builder and add the Clean Language instructions as its knowledge.
• Other AI systems: use the portable instructions as custom instructions, project knowledge, or a system prompt.

See INSTALL.md for complete setup instructions.

Usage examples

• "Rewrite this email using Clean Language."
• "Audit this report for generic AI language without changing the technical meaning."
• "Draft a CIO-level Teams message from these notes."
• "Proofread this policy clause with minimal intervention."
• "Adapt this email into a LinkedIn-compatible response."

Distribution and assurance

Latest release: . The full changelog is on the releases page.

Tagged releases provide:

• clean-language.zip, containing the upload-ready Skill;
• clean-language.zip.sha256, containing its SHA-256 checksum.

The package contains prose instructions, metadata, and reference material. It contains no executable scripts, software dependencies, or instructions that require external network access.

Licence

Except where otherwise noted, all material in this repository, including the Clean Language skill and the cleanlanguage.ai website content, markup, and styles, is licensed under Creative Commons Attribution-ShareAlike 4.0 International (SPDX: CC-BY-SA-4.0).

Suggested attribution:

Clean Language by Jeff Posluns, https://cleanlanguage.ai, licensed under CC BY-SA 4.0. Changes were made.

Third-party material retains its original licence. See NOTICE.md.