CredLens NarrativeAI is a Chrome Extension that analyzes YouTube Shorts and evaluates the overall narrative presented in a video using evidence retrieved from trusted sources.
Unlike traditional fact-checking systems that verify isolated statements or rely heavily on keyword matching, CredLens waits until sufficient context is available, constructs a narrative understanding of the video, retrieves supporting evidence, and generates an evidence-based assessment.
Claims that are identified as Insufficient Evidence, Misleading, or Exaggerated are automatically archived into a public knowledge base, allowing users to explore and review previously analyzed content.
Most misinformation is communicated through narratives rather than individual statements.
Traditional verification systems often:
- Analyze incomplete transcripts
- Trigger verification too early
- Depend heavily on keyword matching
- Miss the creator's actual message
- Fail to evaluate the overall context of a video
As a result, many misleading narratives can appear credible despite individual statements being technically true.
CredLens addresses this by evaluating the complete narrative instead of isolated transcript fragments.
YouTube Short
↓
Transcript Collection
↓
Dynamic Watch Gate
↓
Transcript Stabilization
↓
Claim Worthiness Filter
↓
Narrative Synthesis
↓
Evidence Retrieval
↓
Evidence Evaluation
↓
Verdict Generation
↓
Knowledge Base Archive
Rather than fact-checking individual sentences, CredLens analyzes the creator's complete narrative.
Example:
Transcript:
- "Leaky gut causes eczema."
- "Leaky gut causes rosacea."
- "Ghee repairs gut lining."
Narrative Understanding:
The creator argues that gut dysfunction contributes to skin disorders and that dietary interventions such as ghee may improve gut health.
Verification is performed on this narrative representation.
Verification begins only after sufficient context has been collected.
The system dynamically adapts to:
- Video length
- Transcript completeness
- Caption stability
This prevents premature verification while still allowing short videos to be analyzed.
The transcript pipeline continuously cleans and stabilizes captions before analysis.
Features include:
- Duplicate caption removal
- Active caption tracking
- Video-to-video transcript isolation
- Transcript contamination prevention
Not every video should be analyzed.
CredLens automatically filters out:
- Music videos
- Entertainment clips
- Vlogs
- Reaction content
- Low-information transcripts
- Non-factual conversations
Only videos containing meaningful factual assertions proceed to verification.
Evidence retrieval is tailored to the type of claim being evaluated.
Sources include:
- PubMed
- Google Fact Check Tools
- Google News RSS
Examples:
- Health and nutrition claims prioritize scientific literature.
- Current events and political claims prioritize recent reporting.
- Technology and science claims use both academic and news sources.
Retrieved evidence is analyzed against the video's narrative.
The system determines whether evidence:
- Supports the claim
- Contradicts the claim
- Provides mixed evidence
- Is insufficient to verify the claim
A structured assessment explains the reasoning behind the verdict.
Possible outcomes include:
- Supported
- Misleading
- Exaggerated
- Insufficient Evidence
Each verdict contains:
- Narrative Summary
- Central Claim
- Supporting Claims
- Evidence Assessment
- Evidence Strength
- Supporting Sources
Claims classified as:
- Insufficient Evidence
- Misleading
- Exaggerated
are automatically archived in a Supabase-backed public knowledge base.
Each archived entry stores:
- Video metadata
- Transcript
- Narrative summary
- Central claim
- Supporting claims
- Retrieved evidence
- Assessment reasoning
- Evidence strength
Users can browse previously analyzed content and review supporting evidence.
Each analyzed video is cached using its YouTube Video ID.
Cached data includes:
- Narrative summary
- Claims
- Evidence
- Verdict
- Analysis metadata
Benefits:
- Faster revisits
- Reduced API usage
- Lower verification cost
Verification continues even if users scroll away from a Short.
Completed analyses remain available through caching and are immediately displayed when revisiting the same video.
Each YouTube Short maintains an independent analysis state.
Switching videos automatically resets:
- Transcript buffers
- Caption history
- Verification state
- Progress tracking
This prevents cross-video contamination.
- React
- TypeScript
- Vite
- Tailwind CSS
- Manifest V3
- Chrome Storage API
- OpenRouter
- Claude 3.5 Haiku
- PubMed API
- Google Fact Check Tools API
- Google News RSS
- Supabase
- PostgreSQL
- Row Level Security (RLS)
npm install
npm run devCreate a .env file:
VITE_SUPABASE_URL=your_supabase_url
VITE_SUPABASE_ANON_KEY=your_supabase_anon_keynpm install
npm run buildLoad the extension:
- Open Chrome
- Navigate to
chrome://extensions - Enable Developer Mode
- Click Load Unpacked
- Select the extension build directory
Configure API keys and view the retrieval pipeline used for narrative verification.
CredLens analyzes the narrative of a YouTube Short and generates an evidence-based assessment.
Public archive of analyzed claims with verdicts, confidence levels, and evidence summaries.
Detailed claim analysis including supporting evidence, assessment reasoning, and verification results.
CredLens aims to improve transparency in short-form content by helping users understand whether the overall narrative presented in a video is supported by credible evidence.
Rather than replacing human judgment, CredLens provides context, evidence, and structured assessments that enable users to make more informed decisions about the information they consume.