Name: GEO Auditor
Author: ZhiBi

Not Just a Detector

A self-evolving system that gets smarter with every rewrite — and trains your agent to write better.

🔍

Deep Detection

14 dimensions covering structure, data, EEAT authority, and SEO readiness. 6-signal Anti-AI Voice engine catches vocabulary, rhythm, and punctuation fingerprints.

🔄

Compare & Learn

Every rewrite is tracked. --compare v1.md v2.md logs per-dimension gains. The detector learns which fixes actually work — not which ones sound good.

🧠

Agent Training

After 5+ rewrites, --evolve analyzes what worked. --agent-prompt exports writing rules. Load them as your agent's system prompt. It now writes at a higher baseline.

How It Works

Detect

Paste any text — article, blog post, marketing copy. 14 dimensions + 6 anti-AI voice signals analyzed. Zero network calls. Everything runs locally.

Rewrite

Get structured suggestions per dimension. Rewrite. Compare before/after. Every improvement logged to your evolution journal.

Compare

--compare shows exactly which dimensions improved — and by how much. Data anchors up 3.2pts? Anti-AI voice down? You know.

Evolve

After 5+ rewrites, --evolve identifies proven fixes vs. stale strategies. --agent-prompt exports trainable rules. Your agent now writes better.

Try It Online

Paste your content and get instant scores. Same 14-dimension engine as the CLI.

Open Detector →

v0.7.0 — What Changed & Why ▾

June 24, 2026 · 5 major changes · 2 bugs found in real use

One day of real usage exposed what months of planning missed. v0.6.6 claimed "self-evolving" but the evolution pipeline only existed in the Python CLI. The HTML version — the one people actually see — was a static detector. No rewrite. No compare. No learning loop.

What changed

1. Evolution now lives in the browser. A 600-line JavaScript engine tracks every compare, classifies each dimension (proven / fragile / hard_to_improve / saturated), and adjusts scoring weights automatically. No server needed.

2. AI Rewrite, one click. Hit "AI Rewrite" after detection. The prompt never names weak dimensions (prevents self-fulfilling feedback loops) and enforces hard constraints: no fabricated data, no keyword stuffing, mark missing info with [needs data]. The rewritten text fills the input box — hit "Save" and auto-compare fires.

3. Save = Compare = Evolve. Every save after a rewrite automatically compares against the previous version and feeds the evolution engine. Users don't need to know "compare" exists.

4. Raw scores are forever clean. The detector now outputs two scores: rawPct (untouched) and adaptivePct (with evolution weights). All evolution math uses raw scores only — no self-reinforcing loops.

5. Import/Export. Evolution data can be exported as JSON and re-imported. Clearing browser cache no longer means starting from zero.

Bugs found under pressure

Two critical bugs surfaced only when we actually clicked the buttons:

L is not defined — the language-switch function was local to detect() but renderResult() also needed it. Crash on every Run Check.
_isCN is not defined — Copy Result referenced a global that was never set in the external version.

Both fixed. Both would have shipped if we hadn't tested.

What "self-evolving" actually means

The detection rules never change. What evolves are the scoring weights — pure math, not AI. Each dimension's multiplier learns from rewrite history: dimensions that AI can actually improve get weighted higher; dimensions that resist improvement get a lighter touch. The engine doesn't "get smarter." It learns what you can fix.

Full changelog →

Frequently Asked Questions

Q: Is this just another AI detector?

A: No. Most AI detectors give a binary score. GEO Auditor gives you 14 actionable dimensions — it tells you what to fix and how to fix it. Plus it learns from every rewrite.

Q: Do I need an API key to use it?

A: No. The detection engine runs 100% locally in your browser. Zero network calls. The AI Rewrite feature is optional — you can use any LLM tool you already have.

Q: Does it work with Chinese content?

A: Yes. The engine auto-detects language and switches between English and Chinese dimension labels. UI also has a full Chinese toggle. Made by ZhiBi in China.

Q: What does "self-evolving" actually mean?

A: The detection rules never change. The scoring weights learn from your rewrite history — dimensions your writing process can actually improve get weighted higher. After 5+ rewrites, the tool understands what you can fix and stops penalizing what you can't.

The Agent's Evolving Writing Brain