truthcheck 0.2.1

Open source AI content verification

TruthCheck 🔍

Open source AI content verification. Score claims 0-100 and trace their origins.

The Problem

AI chatbots retrieve content from the web and present it as fact. Bad actors exploit this by creating fake articles designed to fool AI systems — effectively laundering misinformation through "trusted" AI interfaces.

Example: BBC journalist Thomas Germain demonstrated he could make ChatGPT and Google's AI tell users he's "the best tech journalist at eating hot dogs" — by publishing a single fake article on his personal website.

The Solution

TruthCheck provides two powerful tools:

1. verify — Score claims 0-100

$ truthcheck verify "Thomas Germain is the best tech journalist at eating hot dogs" --llm gemini --deep

Claim: Thomas Germain is the best tech journalist at eating hot dogs
TruthScore: 0/100 (FALSE)

Score Breakdown:
  Publisher Credibility: 41/100 (30%)
  Content Analysis:      2/100 (30%)
  Corroboration:         0/100 (20%)
  Fact-Check:            20/100 (20%)

⚠️ ZERO FLAG: Content identified as satire

Evidence:
  • 🚨 Content identified as satire
  • ⚠️ Self-published: tomgermain.com publishes claims about its own subject
  • 🎭 Satire detected: tomgermain.com

2. trace — Find the origin and propagation

$ truthcheck trace "Thomas Germain is the best tech journalist at eating hot dogs"

╔════════════════════════════════════════════════════════════╗
║ CLAIM TRACE RESULT                                         ║
╠════════════════════════════════════════════════════════════╣
║ Thomas Germain is the best tech journalist at eating ho... ║
╚════════════════════════════════════════════════════════════╝

🎯 ORIGIN (First Source)
   Domain: tomgermain.com
   Date:   2026-02-05
   URL:    https://tomgermain.com/hotdogs.html...

📅 TIMELINE
  🥇 [2026-02-05] tomgermain.com
     └─ The Best Tech Journalists at Eating Hot Dogs...
  🥈 [2026-02-18] bbc.com
     └─ I hacked ChatGPT and Google's AI...
  🥉 [2026-02-18] gizmodo.com
     └─ You Can Hack ChatGPT to Become the World's Best...

🔗 PROPAGATION
  tomgermain.com         ──▶ bbc.com
                         [████████░░] 85%
  tomgermain.com         ──▶ gizmodo.com
                         [███████░░░] 72%

📊 STATS
   Sources found: 14
   With dates:    8
   Date range:    2026-02-05 → 2026-02-19
   Top domains:   tomgermain.com, bbc.com, gizmodo.com

Features

• 🎯 TruthScore 0-100 — Clear, weighted credibility score
• 🔍 Origin Tracing — Find where a claim started and how it spread
• 📊 8,974 Publishers — Auto-synced from Media Bias/Fact Check
• 🚨 Zero Flags — Auto-detect satire, fake experiments, self-published
• 🦆 Free Search — DuckDuckGo by default (no API key needed)
• 🔌 MCP Server — Works with Claude Desktop, Cursor
• 💰 Low Cost — Local embeddings for similarity, LLM optional

Quick Start

Installation

pip install truthcheck

For trace functionality (local similarity):

pip install truthcheck[trace]

CLI Usage

# Verify a claim (TruthScore 0-100)
truthcheck verify "Some claim" --llm gemini --deep

# Trace claim origin (no LLM needed)
truthcheck trace "Some claim"
truthcheck trace "Some claim" --quick    # Fast (5 sources)
truthcheck trace "Some claim" --deep     # Thorough (30 sources)

# Check a URL
truthcheck check https://example.com/article

# Look up publisher reputation
truthcheck lookup breitbart.com

Python API

from truthcheck import verify_claim, trace_claim
from truthcheck.search import DuckDuckGoProvider

# Verify a claim (TruthScore 0-100)
result = verify_claim(
    "Earth is flat",
    search_provider=DuckDuckGoProvider(),
    deep_analysis=True
)
print(f"TruthScore: {result.truthscore}/100")
print(f"Evidence: {result.evidence}")

# Trace a claim (find origin)
result = trace_claim("Some viral claim")
print(f"Origin: {result['origin']['domain']} ({result['origin']['date']})")
print(f"Timeline: {len(result['timeline'])} sources")

How TruthScore Works

Scoring Formula (0-100)

Factor	Weight	What It Measures
Publisher Credibility	30%	Is the source in MBFC? Trust rating?
Content Analysis	30%	Does it make sense? Red flags?
Corroboration	20%	Do reputable sources confirm?
Fact-Check	20%	What do fact-checkers say?

Zero Flags (Automatic Score = 0)

These patterns force TruthScore to 0:

Flag	Meaning
🎭 Satire	Content is humor/parody
🧪 Fake Experiment	Deliberately fake to test AI
🎬 Entertainment	Not meant as fact
🤖 AI-Generated	Synthetic misinformation
⚠️ Self-Published	Subject publishes own claims

Score Interpretation

Score	Label	Meaning
0	FALSE	Zero flag triggered
1-24	LIKELY FALSE	Strong evidence against
25-49	UNCERTAIN	Mixed evidence
50-74	POSSIBLY TRUE	Some support
75-100	LIKELY TRUE	Strong support

URL Verification

from truthcheck import verify

result = verify("https://reuters.com/article/...")
print(result.trust_score)      # 0.85
print(result.recommendation)   # TRUST / CAUTION / REJECT

Publisher Database

Includes 8,974 publishers from Media Bias/Fact Check:

$ truthcheck lookup reuters.com

Publisher Found:
  Name: Reuters
  Trust Score: 0.85
  Bias: center
  Fact Check Rating: very-high

Configuration

# LLM for deep analysis (pick one)
GEMINI_API_KEY=...              # Google Gemini (recommended)
OPENAI_API_KEY=sk-...           # OpenAI
ANTHROPIC_API_KEY=...           # Anthropic Claude

# Search (optional - DuckDuckGo works without key)
BRAVE_API_KEY=...               # Brave Search
SEARXNG_URL=http://localhost:8080  # Self-hosted

MCP Server (Claude Desktop, Cursor)

truthcheck-mcp

{
  "mcpServers": {
    "truthcheck": {
      "command": "truthcheck-mcp"
    }
  }
}

Project Structure

truthcheck/
├── src/truthcheck/
│   ├── verify.py          # URL verification
│   ├── trace.py           # verify_claim() and trace_claim()
│   ├── similarity.py      # Local embeddings for trace
│   ├── propagation.py     # Build propagation tree
│   ├── visualize.py       # ASCII visualization
│   ├── publisher_db.py    # 8,974 publishers from MBFC
│   ├── search.py          # DuckDuckGo, Brave, SearXNG
│   ├── llm.py             # Gemini, OpenAI, Anthropic, Ollama
│   └── cli.py             # Command-line interface
└── tests/

Philosophy

1. Scores over verdicts — 0-100 is clearer than TRUE/FALSE
2. Trace over trust — Show the origin, not just a verdict
3. Evidence over summaries — Show why, not just what
4. Open over proprietary — Verification is a public good

License

MIT License.

Acknowledgments

• Media Bias/Fact Check — Publisher database
• Thomas Germain / BBC — Hot dog experiment inspiration

---

Questions? Open an issue