ai-writing-detector

Detects and scores AI-writing patterns in text without rewriting it.

5.0k|441|Updated Mar 6, 2026

Search Tags:#node.js #pattern matching #text analysis #false positives #stylometry #ai writing detection #writing audit

Ready to run

npx skills add https://github.com/conorbronsdon/avoid-ai-writing --skill ai-writing-detector-conorbronsdon

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: ai-writing-detector
Source: https://github.com/conorbronsdon/avoid-ai-writing/tree/main/skills/ai-writing-detector
Command: npx skills add https://github.com/conorbronsdon/avoid-ai-writing --skill ai-writing-detector-conorbronsdon

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? It identifies AI-generated writing patterns ("AI-isms") in prose and produces a deterministic score and issue report, without altering the original text. ## Core Features & Use Cases - Deterministic Detection: Runs a bundled Node.js engine that flags regex, structural, and stylometric AI-writing patterns, returning a 0-100 score, label, and issue list as JSON. - Context-Aware Analysis: Supports general, technical, marketing, and personal context modes, plus a rendered-markdown source mode that excludes YAML frontmatter and HTML comments from scoring. - Bypass Resistance: Normalizes zero-width characters, Cyrillic/Greek homoglyphs, and roleplay markers that humanizer tools use to evade detection. - Use Case: Paste a draft blog post and ask for an AI-pattern scan; the skill reports flagged phrases grouped by severity, distinguishes justified findings from plausible false positives, and never rewrites the text. ## Quick Start Scan this text for AI-writing patterns and report the score and flagged issues without rewriting anything.

Frequently Asked Questions about ai-writing-detector

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I detect AI writing patterns in a text file?▼

Run the bundled detector with node scripts/detect.js --file path/to/draft.md, or pipe text via stdin. It prints a JSON report with the score, label, issue types, matched text, and suggestions, and never modifies the input.

What does the AI writing detector score mean?▼

The score is a 0-100 value computed from weighted, deduplicated issue categories normalized by word count. It is a signal of pattern density, not ground truth about authorship, and should not be treated as an authorship classifier.

Does the detector work on Markdown files with frontmatter?▼

Yes. Use --source-mode rendered-markdown so YAML frontmatter and HTML comments are excluded from scoring. The default plain mode treats all input as prose, which can inflate scores on metadata-heavy files.

Can the detector catch humanizer or bypass tools?▼

It normalizes zero-width characters, Cyrillic and Greek homoglyph lookalikes, and roleplay action markers before pattern matching, so obfuscated tokens like a Cyrillic-spelled "delve" still trigger detection. Normalization itself counts as a corroborating signal.

What are the limitations of AI writing detection?▼

Deterministic hits are candidate matches, not proof of AI authorship. False positives occur with technical terms, genre conventions, and register effects, so findings require contextual assessment and must never be converted into authorship claims.