Wikipedia's AI-spotting playbook just became a tool for human-sounding prose
A new tool called Humanizer taps Wikipedia's AI-detection guide to strip out common "tells" in machine-written text. It's a custom skill for Claude Code that edits phrasing so AI drafts read more like a human wrote them, as spotted by Ars Technica.
For working writers, this isn't just a curiosity. It's a signal: editors are watching for patterns. Your workflow should be, too.
What Humanizer does
Developer Siqi Chen built Humanizer by feeding Anthropic's Claude the list of signals Wikipedia editors use to flag weak AI content. The skill then rewrites or removes those signals before text lands in an editor's lap.
Examples from the project's page show the shift. "Nestled within the breathtaking region" becomes "a town in the Gonder region." "Experts believe it plays a crucial role" tightens to "according to a 2019 survey by…". It also trims collaborative sign-offs like "I hope this helps!" that scream chatbot.
Chen says the tool will auto-update as Wikipedia's guidance evolves. Expect model makers to keep adjusting their defaults as well; OpenAI has already toned down ChatGPT's overuse of em dashes after that quirk became a tell.
Why this matters to writers
Publications and platforms are training moderators to spot AI patterns. If your pitch, draft, or client deliverable triggers those signals, you create work for the editor and risk a quick rejection.
Use this shift to your advantage. Treat the "AI tells" list as an editing checklist for clarity, specificity, and credibility-things readers value anyway.
Common AI tells you should avoid
- Vague sourcing: "Experts say…" without names, dates, or reports.
- Hype adjectives: "breathtaking," "groundbreaking," "transformative."
- Chatty helper phrases: "I hope this helps," "As an AI language model," "Here's a step-by-step guide…"
- Generic filler: sweeping claims with no data, soft hedges that say nothing.
- Mechanical rhythm: uniform sentence length and structure.
A practical rewrite checklist for your AI-assisted drafts
- Replace vibes with facts. Add names, dates, sources, and numbers.
- Cut promo adjectives. Use concrete descriptors and comparisons.
- Delete helper-talk. Remove meta-comments and friendly sign-offs.
- Vary the cadence. Mix short punchy lines with longer, specific ones.
- Swap abstracts for specifics. Places, people, examples, quotes.
- Trim hedges. Keep "may," "might," and "could" only where they truly matter.
- Proof for tone drift. Aim for confident, neutral, and useful.
- Run a "tell sweep." Search for clichés you overuse and kill them.
Ethics and positioning
Use tools like Humanizer as quality control, not camouflage. If a piece is AI-assisted, your judgment, reporting, and editing should carry it across the finish line.
The win isn't "beating detectors." It's producing clear work that reads clean, cites sources, and stands up to scrutiny.
What's next
As Wikipedia updates its guidelines and AI providers tweak outputs, the signals will shift. Your best hedge is timeless: specificity, evidence, and voice.
If you want more structured ways to sharpen your AI writing workflow, explore curated tools and training built for copy-driven work at AI tools for copywriting.
Write like you respect the reader's time. The rest follows.
Your membership also unlocks:
