Skip to content
TruthVouch Docs

Getting Started with Shield

Getting Shield running takes just three steps. You’ll start detecting hallucinations within 15 minutes and see first results within 24-48 hours.

Prerequisites

Before launching Shield, you need:

  • Truth Nuggets (at least 5-10) — your ground truth about your organization
  • Shield enabled on your subscription tier (Starter or higher)

If you haven’t created Truth Nuggets yet, start here. The onboarding wizard can create them automatically in 10 minutes.

Step 1: Create Your Truth Nuggets (10 minutes)

Shield only detects hallucinations against facts you’ve explicitly defined. Start with essentials:

Minimum 5 nuggets:

  1. Company founding date or stage (e.g., “Series B, founded 2024”)
  2. CEO/founder name (e.g., “CEO: Sarah Chen”)
  3. Main product name and purpose
  4. Pricing or availability (if public)
  5. Company location or size

Go to: Knowledge Base → Truth Nuggets → Add Truth Nugget

Or use the quick import to load pre-prepared facts via CSV.

Examples:

  • “TruthVouch was founded in 2024”
  • “Headquarters: San Francisco, California”
  • “Shield detects hallucinations with 94% accuracy”
  • “Shield starts at $349/month”
  • “Trusted by 500+ enterprises”

Once created, they’re active immediately. Shield uses them to score incoming AI responses.

Step 2: Schedule Monitoring (5 minutes)

Tell Shield which AI engines to monitor and how often:

Shield → Settings → Monitoring Schedule → Add Schedule

Or click the Setup Monitoring button on the Shield dashboard.

Choose AI Engines

Select which engines to monitor. Available options:

EngineSupported?Notes
ChatGPT (GPT-4, 4o, 4.5)YesMost popular; highest priority
Claude (3.5, 4)YesFast growing; technical audience
Gemini (Advanced, 2.0)YesIncreasingly used for brand mentions
Perplexity (Pro)YesAI search engine; critical for research
Copilot (Windows, web)YesEnterprise audience
OthersComing soonLet us know which you want monitored

Pro tip: Start with ChatGPT and Claude (cover 60%+ of usage), add Gemini/Perplexity based on your customer base.

Set Polling Frequency

Choose how often Shield queries each engine:

FrequencyWhen to UseCost
HourlyCritical facts (product name, CEO, pricing)Higher volume
Every 6 hoursImportant facts (company mission, features)Medium volume
DailyGeneral brand monitoringLow volume
WeeklyHistorical facts, evergreen claimsVery low volume
MonthlyCompliance facts, legal positioningMinimal

Recommended starting point:

  • ChatGPT/Claude: Daily (covers most queries)
  • Gemini/Perplexity: Every 6 hours (fewer total queries but growing)
  • Copilot: Weekly (lower volume, enterprise-only)

Create Template (Optional)

Organize monitoring by category:

  • Template 1: “Product Monitoring” (all product facts, hourly)
  • Template 2: “Brand Monitoring” (company/leadership facts, daily)
  • Template 3: “Competitive Monitoring” (competitive facts, weekly)

Templates let you adjust schedules by category as you scale.

Save & Activate

Click Save Schedule. Monitoring starts immediately.

You’ll see a confirmation: “Monitoring active for 5 AI engines, 3 schedules, 25 cross-checks/day”

Set up notifications so you know when Shield detects hallucinations:

Settings → Notifications → Shield Alerts

Choose Channels

  • Email (always available): Get a digest with summary and links
  • Slack (Professional+): Real-time alerts in your workspace
  • Teams (Professional+): Direct messages to your team
  • PagerDuty (Enterprise): Escalate critical issues to on-call
  • Webhooks (Business+): Send alerts to custom systems

Set Severity Thresholds

  • Critical (Score >80): Immediate notification
  • High (60-80): Morning digest
  • Medium (40-60): Weekly digest
  • Low (<40): Dashboard only (no notification)

Recommended: Get critical + high in real-time, medium/low in weekly digest.

Save

Click Save. You’re ready to go.

What Happens Next

First 24 Hours

  • Shield begins querying your selected AI engines
  • Queries appear in audit log (viewable in Settings → Audit)
  • No detections yet (baseline establishment)

24-48 Hours

  • First detection results arrive
  • You see alerts on the Shield dashboard
  • Some may be false positives — that’s normal

Day 3-7

  • Patterns emerge: which engines are most accurate about you
  • Health Score stabilizes
  • You identify which Truth Nuggets need refinement

Week 1 Review

Review your first detections:

  1. Open Shield dashboard
  2. Review each alert
  3. Decide: True positive (AI is wrong) or false positive (your nugget needs updating)
  4. Approve corrections for true positives
  5. Edit Truth Nuggets for false positives

This feedback trains Shield and improves future detection quality.

First Detection Example

Your Setup:

  • 8 Truth Nuggets (company info, products, pricing)
  • Monitoring: ChatGPT, Claude, Gemini (daily)
  • 24 cross-checks/day

Day 2, 3:45 PM: First alert appears

  • “Claude hallucinates about your pricing”
  • Claude response: “TruthVouch Shield costs $199/month”
  • Your truth: “Shield starts at $349/month”
  • Severity: High (75/100) — incorrect pricing damages sales
  • Status: New, unreviewed

You Take Action:

  • Click the alert to see full Claude response
  • Verify your truth nugget is correct (yes, $349 is right)
  • Click Fix to deploy correction
  • Choose correction method: Neural Fact Sheet (default)
  • System confirms: “Correction deploying to your domain…”

Correction Deploys (under 4 seconds):

  • Neural Fact Sheet published
  • JSON-LD added to your pricing page
  • Feedback logged for Claude trainer

24 Hours Later:

  • Shield re-polls Claude
  • Claude now says: “$349/month” (correct)
  • Alert auto-resolves: “Corrected”
  • Dashboard shows: “7/7 engines correct on Shield pricing”

Monitoring Your Progress

Shield Dashboard

Check anytime to see:

  • Health Score: Your current accuracy across all engines
  • Recent Alerts: New hallucinations (last 48 hours)
  • Top Issues: Most-cited hallucinations (all time)
  • Engine Status: How accurate each engine is about you
  • Correction History: Which corrections worked

Weekly Review

Spend 15 minutes/week:

  1. Check dashboard for new alerts
  2. Approve or dismiss corrections
  3. Update any Truth Nuggets that false positives revealed
  4. Note trends (which engines improve, which degrade)

Scaling Up

Week 2: Add More Engines

If you’ve solved initial issues, add Perplexity or Copilot to your monitoring.

Week 3: Add More Facts

Expand from 8 to 20+ Truth Nuggets covering:

  • All products (not just main one)
  • Financial facts (if public)
  • Leadership team
  • Competitive positioning

Month 2: Optimize Schedules

Based on detection patterns, adjust frequency:

  • Facts with frequent hallucinations → Hourly
  • Facts that are stable → Weekly

Month 3: Competitive Intelligence

(Professional+ tier) Start monitoring 5 competitors’ facts on your platform:

  • See what AI says about them
  • Spot competitive narrative gaps
  • Identify opportunities

Troubleshooting

No Alerts After 48 Hours

Check:

  • Monitoring is enabled: Shield → Settings → Monitoring Schedule
  • At least 5 Truth Nuggets created
  • Queries are running: Settings → Audit → Filter by ‘Cross-check’

If all look good, it may mean AI engines have accurate info — that’s a win!

Too Many Alerts (Alert Fatigue)

Adjust severity thresholds or pause low-value engines:

  • Increase minimum threshold from 60 to 70
  • Set daily/weekly frequency instead of hourly
  • Focus on 2-3 key engines

Can’t Approve Corrections

Some AI engines don’t accept direct feedback. Use Neural Fact Sheet method instead (automatic for most engines).

Need Help?

Contact support: help@truthvouch.com or chat in-app.

Next Steps