# 竞品内容 Pattern 雷达 · AI 答案里的重复说法

Examines AI answers and rival copy to detect repeated phrases, quoted sources, and formulaic text. Helpful for auditing LLM response uniformity.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed
- Author: savantcat
- License: MIT · policy=link-only · description only; the file stays with the author
- Source: https://github.com/savantcat/pattern-radar
- Updated: 2026-10-21
- HTML: https://affpapa.org/en/nevibecode/mcp/pattern-ai

## What it does

The `pattern-radar` server processes raw text answers from JSON arrays or `geo-monitor` records. It relies on character-level n-grams, expanding high-frequency fragments outward into full terms. Alongside repeated wording, the engine tracks brand share, extracts cited web domains, and calculates pairwise response similarity.

Thresholds derive from a null-hypothesis simulation. The tool scrambles characters across the corpus while preserving length and character counts, picking the p95 baseline to eliminate accidental patterns. It runs with zero external Python dependencies in read-only mode, ignoring template placeholder domains automatically.

Corpora under 12 samples trigger an immediate rejection, preventing skewed metrics from tiny datasets. The logic does not perform semantic clustering, meaning synonymous expressions with different characters remain unmerged.

## Requirements
- Needs: Python and uv (commands run via uvx)

## Install

### Claude Code (native platform)

**Install**
- The MCP registry has no launch command for it: see the repository README

## When to use it
- Find recurring formulaic phrases across generated model answers
- Extract cited domains to determine primary information sources
- Measure duplicate rate and wording fixation for identical prompts
- Track competitor brand visibility and positioning within LLM output

## Example prompts
- Use pattern-radar to analyze recurring phrasing in our collected answer dataset
- Run self_check via pattern-radar before running the main analysis
- Call explain_method to view the statistical thresholds and corpus boundaries

## Similar
- [optimize-for-ai](https://affpapa.org/en/nevibecode/skills/optimize-for-ai.md) — skill · 6600 installs
- [geo-content-optimizer](https://affpapa.org/en/nevibecode/skills/aaron-he-zhu-geo-content-optimizer.md) — skill · 1700 installs
- [best-ai-marketing-platform-benchmark](https://affpapa.org/en/nevibecode/skills/best-ai-marketing-platform-benchmark.md) — skill · 903 installs
- [geo-brand-mentions](https://affpapa.org/en/nevibecode/skills/geo-brand-mentions.md) — skill · 822 installs
- [geo-fundamentals](https://affpapa.org/en/nevibecode/skills/geo-fundamentals.md) — skill · 168 installs
- [Sleepwalker](https://affpapa.org/en/nevibecode/mcp/sleepwalker.md) — MCP server · ★ 25

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md