Skip to content
TruthVouch Docs

Shield Overview

Shield is TruthVouch’s hallucination detection and correction engine. It monitors what 9+ AI engines say about your organization, automatically detects inaccuracies, and deploys corrections in under 4 seconds.

Shield Dashboard - Real-time hallucination detection and accuracy monitoring

The Problem

AI engines hallucinate facts about organizations:

  • ChatGPT claims you shut down after a failed funding round (from 2021 news)
  • Claude says your CEO is someone who left 3 years ago
  • Gemini lists a discontinued product as your main offering
  • Perplexity misquotes your pricing or features

These hallucinations damage your brand, confuse customers, and create regulatory risk.

The Solution: Shield

Shield solves this with three core capabilities:

1. Continuous Monitoring

Shield automatically queries 9+ AI engines (ChatGPT, Claude, Gemini, Perplexity, Copilot, and more) on a schedule you set:

  • Hourly polling for critical facts
  • Daily polling for general brand monitoring
  • Custom schedules per category

Monitoring starts within minutes of setup and requires no manual intervention.

2. Accurate Detection

Shield compares AI responses against your Truth Nuggets using Natural Language Inference (NLI):

  • 94%+ detection accuracy
  • Distinguishes facts from opinions
  • Handles paraphrases and reformulations
  • Learns from your feedback

When Shield detects a hallucination, it calculates a severity score (0-100) and alerts your team.

3. Automated Corrections

Shield automatically deploys corrections via:

  • Neural Fact Sheets: AI-optimized structured data published to your domain
  • JSON-LD: Schema.org markup crawled by AI engines
  • Direct feedback: Submission to OpenAI, Anthropic, Google (where available)

Corrections deploy in under 4 seconds with automatic verification. Most hallucinations are corrected within 24 hours.

Key Features

FeatureBenefit
9+ AI enginesCoverage of all major LLMs your customers use
94% detection accuracyNLI-based inference beats simple keyword matching
Sub-4-second correctionsFastest correction in the industry
Custom schedulesMonitor critical facts hourly, others weekly
Competitive intelligenceSee what AI says about competitors
Trust Score badgeEmbeddable widget showing your accuracy
REST APIIntegrate detection into your stack
Audit trailSHA-256 keyed history for compliance

Availability & Pricing

Available from: Starter tier ($349/month)

Includes:

  • Monitoring up to 5 AI engines (ChatGPT, Claude, Gemini, Perplexity, Copilot)
  • Up to 100 cross-checks per day
  • Basic alerts (email + dashboard)
  • Truth Score badge

Professional ($1,199/mo)

  • All Starter features
  • 9+ AI engine monitoring
  • Unlimited cross-checks
  • Slack/Teams notifications
  • Competitive intelligence (5 competitors)

Business ($2,499/mo)

  • All Professional features
  • Webhook integrations
  • Custom AI engine monitoring
  • Unlimited competitors
  • 99.9% uptime SLA

Getting Started

Three Steps

  1. Create Truth Nuggets → Define your ground truth (15 min)
  2. Schedule Monitoring → Select AI engines and frequency (5 min)
  3. Review Detections → Monitor dashboard and alerts (ongoing)

Get started →

How It Works (High Level)

Your Truth Nuggets
        ↓
    Query AI Engines
        ↓
    Extract Entities (NER)
        ↓
    Compare with Truth (NLI)
        ↓
    Score & Alert (if mismatch)
        ↓
    Deploy Correction (auto or manual)
        ↓
    Verify Success (24-72h)

Detailed explanation →

Real-World Example

Scenario: You launch a new product called “TruthVouch Governance”

Day 1 (Launch day)

  • Create 5 Truth Nuggets about the product
  • Set hourly polling on ChatGPT, Claude, Gemini
  • Dashboard shows “Not yet detected” for all engines

Day 3 (First detection)

  • ChatGPT mentions your product in a response, but calls it “TruthVouch Control” (confuses it with older name)
  • Shield flags: “ChatGPT hallucinates about product name”
  • Severity: Medium (70/100) — important for brand but not critical

You Review

  • Click “Fix” on the alert
  • Shield generates correction: “TruthVouch Governance” (correct name, features, pricing)
  • You approve the correction

Correction Deploys

  • Neural Fact Sheet published to your domain (2 seconds)
  • JSON-LD added to your website footer (1 second)
  • Feedback submitted to Anthropic (1 second)

Day 4 (Verification)

  • Shield re-polls ChatGPT
  • ChatGPT now says “TruthVouch Governance” (correct)
  • Alert auto-resolves as “Corrected”
  • Dashboard shows: “8/8 engines aware of TruthVouch Governance”

Impact: In 24 hours, all major AI engines know about your product correctly.

Dashboard at a Glance

Your Shield dashboard shows:

  • Health Score (0-100): Aggregate accuracy across all monitored engines
  • Recent Alerts: Critical issues requiring attention
  • Top Issues: Most-cited hallucinations
  • Engine Comparison: Which engines are most accurate about you
  • Correction Status: Which corrections are deployed and verified
  • Trends: How accuracy improves over time

Learn more →

Common Use Cases

Protect Your Brand

Monitor what AI says about you, detect inaccuracies, deploy corrections before customers see them.

Launch Products

When you launch a new product, Shield monitors all AI engines until they’re all aware and accurate.

Regulatory Compliance

Audit trail proves you monitored AI output and corrected hallucinations. Required for EU AI Act transparency.

Investor Due Diligence

Show investors a dashboard of AI accuracy over time. Demonstrates governance rigor.

Competitive Advantage

Monitor what AI says about 5 competitors. Identify competitive narratives you’re losing.

Next Steps