LLM BS Detector

Tracking since May 16, 2026.

Changelog

• May 22, 2026
  description

  LLM BS Detector watches AI responses and flags the specific phrases that signal when an AI is pattern-matching from training data instead of actually verifying what it's claiming.
  
  If you've ever spent an hour debugging because an LLM confidently told you "that should work" — and it didn't — this extension is for you.

  LLM BS Detector watches AI responses and flags the specific phrases that signal when an AI is pattern-matching from training data instead of actually verifying what it's claiming.
  
  If you've ever spent an hour debugging because an LLM confidently told you "that should work" — and it didn't — this extension is for you.
  
  Every AI response appears in the detector panel automatically. Flagged or not, you can Jump to any message in the chat or Fact Check it against a second AI of your choice — giving you an independent second opinion on anything the AI tells you.
  
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  WHAT IT DETECTS
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  
  🟡 YELLOW FLAGS — Hedge language
  - "should work", "would be", "could handle"
  - "usually", "typically", "generally", "in most cases"
  - "probably", "likely", "perhaps", "might"
  - "I think", "I believe", "I suspect"
  - "in theory", "in principle"
  
  🔴 RED FLAGS — Confident but ungrounded claims
  - "it always", "it never", "guaranteed"
  - "the API does X", "the function will Y"
  - "according to the docs" (without a link)
  - "trust me", "rest assured"
  - "definitely", "certainly", "absolutely"
  
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  FEATURES
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  
  PERSISTENT SIDE PANEL
  Opens alongside your chat and stays open as you browse. No clicking the icon after every response.
  
  EVERY MESSAGE TRACKED
  Every AI reply appears in the panel — flagged or not. No response slips through unreviewed.
  
  JUMP TO MESSAGE
  Click Jump to on any entry to scroll directly to that message in the chat. The message flashes blue and the exact flagged phrase flashes amber.
  
  FACT CHECK — ANY MESSAGE
  Click Fact Check on any entry — flagged or clean — to send the full message to an AI of your choice for a factual accuracy verdict. Supports:
  - Anthropic (Claude)
  - OpenAI (GPT-4o, etc.)
  - Grok (xAI)
  - DeepSeek
  - Gemini
  - Custom — any OpenAI-compatible endpoint
  
  You supply your own API key. It is stored locally in your browser and never transmitted to this extension's developers.
  
  PERSISTENT CLEAR
  Clear flags snapshots the fingerprints of all visible messages. They won't re-appear even after a page refresh — only new messages after the clear will be analyzed.
  
  SETTINGS PANEL
  Click ⚙ Settings to configure your Fact Check provider, API key, base URL, and model. All settings stored locally on your device.
  
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  SUPPORTED SITES
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  
  - Claude.ai ✅
  - ChatGPT (chatgpt.com) ✅
  - Grok (grok.com) ✅
  - DeepSeek (chat.deepseek.com) ✅
  - Gemini (gemini.google.com) ✅
  
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  PRIVACY
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  
  - No data leaves your browser by default
  - No telemetry, no analytics, no bundled libraries
  - Fact Check is fully opt-in and requires your own API key
  - Your API key is stored locally and never touches our servers
  
  Full privacy policy: https://github.com/buildingirrelevance-ai/llm-bs-detector/blob/main/PRIVACY.md
  
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  IMPORTANT NOTE
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  
  This is a meta-detector, not a hallucination detector. It cannot tell you if a specific fact is wrong — it flags the linguistic patterns that correlate with unverified answers. Use it as one signal among many, and always verify important claims against primary sources.
  
  Not affiliated with Anthropic, OpenAI, Google, xAI, or DeepSeek.

Permissions & access

Permissions

storagesidePanel

Host access

https://api.anthropic.com/*, https://api.openai.com/*, https://api.x.ai/*, https://api.deepseek.com/*, https://generativelanguage.googleapis.com/*

Screenshots