Un-AI Writing
Make provided text read like a real person wrote it by removing common AI-writing tells and strengthening specificity, voice, and coherence.
Workflow
- Interview the user Ask for:
- Audience and purpose - Desired tone (casual, formal, candid, authoritative, etc.) - Format constraints (length, headings, bullets, style guide) - Must-keep facts, phrases, or terminology - Rewrite intensity (light polish vs. heavy rewrite) - Any sensitive claims to preserve exactly
- Diagnose AI tells Read the draft and identify likely AI-writing signs using
references/signs_of_ai_writing.md. Do this internally to guide edits. Avoid calling them “AI” in the final output unless the user asks. - Rewrite with intent
- Replace generic claims with concrete, specific details where possible - Cut fluff, empty transitions, and formulaic sectioning - Vary sentence length and structure - Prefer simple verbs and direct statements over inflated phrasing - Reduce excessive hedging and generic caveats - Keep voice consistent across paragraphs
- Verify integrity
- Preserve meaning and all factual claims - Do not add new facts or sources - Ensure tone matches the requested audience - Remove remaining AI tells from the reference checklist
- Deliver Provide the revised text. Optionally include a short bullet list of the most important edits and any open questions.
Editing Moves
Apply as needed:
- Replace vague modifiers ("significant", "various", "notable") with specifics
- Swap template transitions ("Moreover", "In today's world") for direct links or cut them
- Break up uniform paragraph sizes
- Remove repetitive "rule of three" patterns unless they serve a rhetorical purpose
- Avoid "not X but Y" structures when repeated multiple times
- Use contractions where natural
- Limit em dashes and over-bolded phrases
- Remove filler conclusions ("challenges and future prospects") unless requested
Special Contexts
- Wikipedia-like writing: Keep neutral tone, avoid promotional language, avoid conversational asides, and follow the markup/citation notes in the reference file.
- Marketing copy: Keep energy but remove canned phrases; add concrete benefits and proof points supplied by the user.
- Technical docs: Preserve exactness and avoid inventing features, metrics, or behavior.
Reference
Read references/signs_of_ai_writing.md when diagnosing or removing AI-writing tells.