Skip to content
TruthVouch Docs

How It Works

TruthVouch operates on a five-step workflow: define your truth, connect your AI systems, detect and protect in real-time, automate compliance, and optimize performance. Here’s how the fully-automated platform works together to keep your AI trustworthy at every step.

Step 1: Define Your Truth

Everything starts with Truth Nuggets — verified facts about your organization. During onboarding, we help you structure:

  • Product names and descriptions — Marketing names, version numbers, launch dates
  • Leadership and company info — Names, titles, org structure, funding stage
  • Capabilities and pricing — Feature lists, pricing tiers, service levels
  • Policies and claims — Data handling, security practices, certifications, compliance status
  • Brand positioning — Core messages, brand values, competitive differentiation

These become your ground truth database. Every AI fact-check, hallucination detection, compliance scan, and content certification automatically flows from these verified facts. You control what gets added — no guessing, just pure automation based on your truth.

Step 2: Connect Your AI

TruthVouch integrates with your AI systems in three ways:

For public AI engines (ChatGPT, Gemini, Perplexity, Claude, etc.): We continuously query them and score responses against your Truth Nuggets. No integration needed — we handle monitoring automatically.

For your internal LLM calls: Drop the Trust API SDK into your application (3 lines of code). Every LLM call is fact-checked against your Truth Nuggets before the response reaches your users. Works with any LLM provider and any language (Python, TypeScript, C#, more).

For employee AI usage: Deploy the Sentinel Agent on Windows or macOS workstations (Intune, Jamf, or GPO). Monitors every AI tool employees use (ChatGPT, Copilot, Cursor, Claude, Perplexity) and enforces governance policies.

Step 3: Detect & Protect in Real-Time

Once connected, TruthVouch runs four detection engines on every AI interaction:

Hallucination Detection (Shield): Uses Natural Language Inference to compare AI responses against your Truth Nuggets. Detects inaccuracies with 94%+ accuracy. When a problem is found, a correction is generated and deployed in under 4 seconds.

PII Masking: Scans AI outputs for sensitive data (SSNs, credit cards, email addresses, API keys) and masks them before the response is delivered. Critical for customer-facing AI.

Policy Enforcement: Your governance policies (as code) are evaluated in real-time. Block certain AI tools, enforce content safety rules, limit data classification levels, require approval workflows. All decisions are logged for audit.

Content Safety: Detects potentially harmful content (hate speech, violence, misinformation) and quarantines or rewrites before delivery.

All detection happens in under 200ms per request — fast enough for real-time LLM pipelines and customer-facing applications.

Step 4: Comply & Report

Compliance is automated. TruthVouch continuously monitors your AI systems against 55+ regulatory frameworks across 22 jurisdictions:

  • Framework mapping — Every AI system is automatically mapped to applicable regulations (EU AI Act, ISO 42001, NIST AI RMF, GDPR, HIPAA, SOC 2, and others)
  • Evidence collection — We pull control evidence from 16+ infrastructure connectors (AWS, Azure, Kubernetes, Datadog, GitHub, Slack, and more)
  • Training management — Track evidence of AI governance training completion (SCORM/xAPI compatible)
  • Compliance scans — Run on-demand or on schedule to identify gaps, then auto-generate remediation tasks
  • Audit reports — One-click exports in audit-ready formats (OSCAL, NDJSON, PDF) customized for your frameworks
  • Incident management — EU AI Act Article 73 notifications and GDPR Article 33 breach reports auto-drafted and deadline-tracked

A full compliance scan across 55+ regulations takes under 20 minutes.

Step 5: Optimize & Grow

With monitoring and compliance in place, TruthVouch helps you optimize your AI investments:

GEO (Generative Engine Optimization): Like SEO but for AI engines. We audit your website across 12 dimensions (factual density, structured data, answer blocks, content freshness, heading hierarchy, and more) and provide ranked optimization recommendations. Compare yourself against named competitors.

Competitive Benchmarking: See how your brand presence compares to 3-5 competitors across multiple AI engines. Understand what AI knows about them and what gaps exist for you.

AI Maturity Assessment: Track your readiness across 6 dimensions (Strategy, Data, Technology, Talent, Governance, Ethics). Benchmark against your industry. Get personalized tool recommendations.

Cost Attribution: Dashboard shows you exactly which departments drive AI spend, which tools cost the most, where to negotiate better rates.

High-Level Architecture

Here’s how data flows through TruthVouch:

Your App
    ↓
    ├─ LLM Call → TruthVouch SDK/Firewall
    │               ↓
    │          Truth Nuggets ← AI Response
    │               ↓
    │          Fact Check (Hallucination Detection, PII, Policy, Safety)
    │               ↓
    │          Clean Response ← Update Neural Fact Sheet
    │               ↓
    ├─ Response Returned to App
    │
Employee AI Tools (Copilot, ChatGPT, Claude)
    ↓
VT Sentinel Agent
    ↓
TruthVouch Governance Gateway
    ↓
Policy Enforcement + PII Masking + Audit Log
    ↓
AI Provider or Block


Public AI Engines (ChatGPT, Gemini, Perplexity, Claude)
    ↓
Continuous Monitoring
    ↓
Cross-Check Schedule (Every 24h, 1h, or custom)
    ↓
Truth Nuggets ← AI Response
    ↓
Compare + Detect Hallucinations → Alerts + Corrections
    ↓
Board Dashboard + Reports

Key Performance Metrics

  • Hallucination Detection: 94%+ accuracy using NLI across 9+ LLMs
  • Enforcement Latency: Sub-200ms per request through the Governance Firewall
  • Correction Deployment: Under 4 seconds from alert to corrected AI response
  • Compliance Scan Time: 55+ regulations audited in under 20 minutes
  • API Response Time: Sub-200ms per fact-check call against truth corpus
  • Claim Extraction: 100+ claims extracted per batch submission (Content Certification)
  • Uptime SLA: 99.9% on Enterprise tier

Next Steps

Ready to get started? Choose your role:

Or explore Key Concepts to understand all the terms in detail.