Humanize AI Writing: Remove AI Tells, Keep Your Meaning

Unlike the free humanize-writing skill, which rebuilds AI drafts from scratch for a "smell score", this one edits conservatively and proves nothing was lost: every rewrite ships with a fact and meaning check table (numbers, names, dates, links, CTAs, caveats) and a change log, and it flags any line that needs the author's sign-off. An editor, not an AI-detector bypass.

Sample run (real data, 23 Oct 2026): a 513-word AI-sounding news article about Casepoint Public Records (third-news.com, 19 Oct 2026). We ran it through the skill's 7-step method, then counted before and after against a fixed list of AI-tell phrases.

1. Words 513 → 254 · sentences 21 → 18 · average sentence 24.4 → 14.1 words · longest 40 → 24
2. AI-tell phrases 27 → 0 (seamlessly ×2, enhance ×3, ensuring ×2, cutting-edge, revolutionize, navigating, daunting, landscape, realm, "in summary" and more)
3. Trailing "-ing" clauses (", ensuring…", ", facilitating…") 7 → 0 · em dashes 4 → 0
4. Fact check: 21 of 21 protected items kept (October 19, 2026 · SLED · 30 years · FOIA · eDiscovery · Casepoint IQ™ · Microsoft 365, Teams, Slack, Google Workspace · FedRAMP® High · DOD Impact Level 5 and 6 · audit trails · the demo CTA …)
5. Before: "Navigating the complexities of public records requests can be daunting, especially as requests grow in volume and sensitivity." After: "Requests keep getting bigger and more sensitive."
6. Before: "…ensures that human oversight is maintained for crucial decisions, thereby balancing efficiency with accountability." After: "People still make the key decisions."
7. Cut: "Casepoint is positioned to revolutionize how…". Change log: hype with no fact in it
8. Flagged for the author: "with fewer misses" replaces "more accurate identification". Confirm the wording
9. Dropped 8 section headings that each wrapped a single paragraph ("Robust Security Measures", "Conclusion")
10. No detector score is claimed. The goal is copy people want to read

What it does

• Notes what must not change first (facts, caveats, links, CTA), then diagnoses filler, buzzwords, repeated sentence shapes and stiff rhythm
• Turns your voice brief ("direct, warm, no hype") into concrete choices: sentence length, contractions, directness
• Rewrites conservatively. It never adds facts, anecdotes, testimonials or statistics that aren't in your source
• Returns: edit diagnosis, clean rewrite, what changed, a fact and meaning check table, and an author review checklist
• Takes must_keep / must_avoid lists (e.g. keep the pricing date, avoid "leverage")

Use cases

• Emails, landing-page sections, product updates, press releases and internal notes drafted with AI
• Cleaning up a teammate's AI draft without losing claims, links or legal caveats
• Agency or client copy that needs a reviewable change log

Works in Claude Code, Cursor and Codex · tested on real data · v1.0 updated Oct 2026 · questions answered within 24h More Cygnix tools (Apify actors, MCP server): github.com/cygnixlabs-cmd

Pairs well with (Cygnix Labs, on Agensi):

• Source-Backed Research Brief: Facts, Citations + Angles: gather the facts before you draft (agensi.io/skills/source-backed-research-brief-facts-citations-angles)
• SEO Content Brief (free): a writer-ready outline for the post you'll humanize (agensi.io/skills/seo-content-brief-writer-ready-outline)
• Turn Any Offer into Landing Page Copy That Converts: write the page, then humanize it (agensi.io/skills/turn-any-offer-into-landing-page-copy-that-converts)
• AEO Audit: Get Cited in ChatGPT + AI Overviews: make the cleaned-up page quotable by AI answers (agensi.io/skills/aeo-audit-get-cited-in-chatgpt-ai-overviews)
• Find Decaying Pages: GSC Triage + Content Refresh Plan: find which old posts to rewrite first (agensi.io/skills/find-decaying-pages-gsc-triage-content-refresh-plan)