Skip to main content

Fact-Checking Concepts

Core concepts and terminology for understanding the Fact-Checking API.

What is Fact-Checking?​

Fact-checking verifies the accuracy of claims in text. You submit content, the system analyzes it, identifies individual claims, verifies them against sources, and returns a determination with confidence scores.

Core Entities​

Fact Check​

A fact check is the primary resource. It represents the entire verification process for a piece of text.

Key properties:

  • Unique identifier (UUID)
  • Original text submitted
  • Status (pending, processing, completed, failed)
  • Final determination
  • Final confidence score
  • Source URLs (optional, for reference material)
  • Timestamps

Claim​

A claim is an individual statement extracted from your text that can be verified. Complex text may contain multiple claims.

Claim properties:

  • The claim text
  • Verification result
  • Confidence score
  • Supporting evidence

Status​

The status indicates where the fact check is in the processing lifecycle:

  • pending: Submitted, waiting to be processed
  • processing: Currently being analyzed
  • completed: Verification finished successfully
  • failed: Processing encountered an error

Determination​

The final determination is the overall verdict for the text:

  • accurate: Content is verified as factually correct
  • mostly_accurate: Largely accurate with minor issues
  • mixed: Contains both accurate and inaccurate elements
  • mostly_inaccurate: Largely inaccurate with some truth
  • inaccurate: Content is verified as false
  • unverifiable: Insufficient information to verify

Confidence Score​

A number between 0 and 1 indicating certainty in the determination:

  • 0.9 - 1.0: Very high confidence
  • 0.7 - 0.9: High confidence
  • 0.5 - 0.7: Moderate confidence
  • Below 0.5: Low confidence

Sources​

Sources are reference materials used for verification. You can provide specific URLs, or let the system discover sources automatically.

Source properties:

  • URL
  • Title
  • Reliability score
  • Access restrictions (paywall, login required, etc.)

How It Works​

  1. Submit text - Send content to verify
  2. Claim extraction - System identifies verifiable statements
  3. Source gathering - Reference materials are collected
  4. Verification - Each claim is checked against sources
  5. Determination - Overall verdict is calculated
  6. Results available - Full analysis returned

Processing Lifecycle​

Submit → Pending → Processing → Completed
                       ↓
                    Failed

Pending: Your request is queued for processing.

Processing: The system is actively analyzing the text, extracting claims, gathering sources, and verifying information.

Completed: Analysis finished. Results include the determination, confidence score, individual claims, and sources used.

Failed: An error occurred during processing. Check the error details for information.

Verification Results​

Individual claims receive verification results:

  • verified: Claim confirmed by sources
  • disputed: Sources contradict the claim
  • unverified: Insufficient evidence either way
  • partially_verified: Some aspects confirmed, others unclear

Source Discovery​

When you don't provide source URLs, the system automatically discovers relevant sources. When you do provide URLs, the system uses those as primary references while potentially supplementing with additional sources.

Confidence Factors​

Confidence scores are influenced by:

  • Number and quality of sources
  • Agreement between sources
  • Source reliability
  • Clarity of the claim
  • Availability of evidence

Use Cases​

Content Verification: Check articles or posts for accuracy.

Editorial Review: Flag claims that need human review before publication.

Research Validation: Verify factual statements in research content.

Automated Moderation: Filter content based on accuracy thresholds.