Skip to main content
← Back to How It Works

TRUST & VERIFICATION

Trust & Citations

Trust, But Verify. We Make It Easy.

Sentence-level citations, confidence scores, and complete audit trails.

Citation System

Not "Somewhere in Document 3". Exactly Where.

Sentence-Level Precision

Every fact maps to a specific sentence in a specific document. Not just "Document 3" — but "Page 7, Paragraph 3 of the Q3 Security Audit Report." This level of granularity means you can verify any claim in seconds.

  • Every fact traced to its exact source sentence
  • Character offsets for precise highlighting
  • Page, paragraph, and sentence-level location
  • Context before and after included

Example:

Answer: "Revenue grew 23% in Q3 [Doc 1, Sentence 7]"

Click citation to see:

  • → Exact sentence highlighted
  • → Surrounding context
  • → Link to full document

In Practice

Citations in Action

Every answer includes inline citations with confidence scores. Click any source to see the exact passage highlighted in the original document.

AI Chat — Cited Answer with Source Documents
Courdx AI chat showing a detailed answer with inline citations, confidence percentages, and expandable source documents

Confidence Scoring

"Know How Sure We Are"

Every response includes:

  • Overall confidence score (0-100%)
  • Per-citation relevance scores
  • Retrieval quality indicator

Low confidence? We tell you.

Then we offer to search differently or check web sources.

Audit Trail

"Complete Transparency for Compliance"

Every query logged with:

User and timestamp
Documents retrieved
Relevance scores
Citations generated
Response delivered
Guardrail checks performed
  • Export for compliance audits
  • Search and filter by date, user, or topic

Compliance-Ready Architecture

SOC 2GDPRHIPAA

Full audit trails, data residency controls, and role-based access designed to meet the requirements of SOC 2, GDPR, and HIPAA compliance frameworks.

AI Guardrails

"We Catch Problems Before You See Them"

Hallucination Prevention

  • Every claim must have a citation
  • Claims without sources flagged
  • Confidence thresholds enforced

Prompt Injection Blocked

  • Malicious prompts detected
  • Jailbreak attempts logged
  • Admin alerts on suspicious activity

PII Protection

  • Sensitive data detected in inputs
  • PII scrubbed from outputs
  • Configurable sensitivity levels

Output Filtering

  • Toxicity detection
  • Blocked topic enforcement
  • Custom policy rules

Build AI You Can Trust

See how Courdx delivers trustworthy, verifiable answers.