Methodology
How we research, write, and analyze news
Research Process
MeridAIn uses Tavily's deep search API to gather information from across the web. For each topic, we:
- Query multiple search terms to capture diverse coverage
- Prioritize established news sources with editorial standards
- Gather 5-10 sources per article for comprehensive coverage
- Include sources from different geographic regions when relevant
- Timestamp all research for transparency
Article Generation
Articles are generated using Anthropic's Claude AI with specific guidelines:
- Synthesis, not reproduction: We create original summaries, never copy text
- Attribution: All factual claims link to source material
- Balance: Include multiple perspectives on contentious topics
- Clarity: Distinguish between facts, analysis, and opinion
- Limitations: Acknowledge when information is incomplete or uncertain
Important: AI-curated content may contain errors or misinterpretations. We encourage readers to verify critical information through linked sources.
Credibility Scoring
Each article receives a credibility score (0-100) based on three weighted factors. Try the interactive calculator below to see how the formula works:
Interactive Credibility Calculator
Adjust the sliders to see how different factors affect the final credibility score.
Editorial standards, fact-checking history, reputation
Number of sources, geographic and perspective diversity
Cross-source corroboration, primary sourcing, evidence
(75 × 0.6) + (70 × 0.2) + (65 × 0.2)
Visual Breakdown
Source Quality (60%)
- Editorial standards and fact-checking history
- Journalistic credentials and reputation
- Corrections policy and transparency
- Independence from commercial/political interests
Source Diversity (20%)
- Number of independent sources
- Geographic diversity
- Perspective diversity
Claim Verification (20%)
- Corroboration across sources
- Primary vs. secondary sourcing
- Presence of direct quotes/evidence
Score Interpretation
Bias Assessment
Bias scores range from 0-100, where 50 represents neutral:
- 0-30: Left-leaning perspective
- 31-45: Slightly left-leaning
- 46-54: Neutral/balanced
- 55-69: Slightly right-leaning
- 70-100: Right-leaning perspective
Bias is assessed through:
- Source selection patterns
- Language and framing analysis
- Topic emphasis and omissions
- Expert/viewpoint selection
Note: Bias assessment is inherently challenging. Our scores represent algorithmic analysis, not definitive judgments. We aim for transparency, not perfection.
Source Tier System
Tier 1: Primary Sources (85-100%)
Wire services, papers of record, peer-reviewed journals
Reuters, AP, AFP, BBC, NYT, Guardian, Nature, Science, government sources
Tier 2: Quality Sources (70-84%)
Major news outlets with editorial standards
CNN, Bloomberg, Politico, Wired, TechCrunch, major regional papers
Tier 3: General Sources (55-69%)
Established outlets with known editorial perspective
Verified news sources requiring corroboration
Tier 4: Use With Caution (Below 55%)
Sources requiring significant corroboration
Partisan outlets, unverified sources, social media
Scoring Limitations
Transparency Notice
In the interest of full transparency, we disclose the following limitations in our scoring systems. Understanding these helps you interpret scores more accurately.
Source Registry Coverage
- Our source registry currently covers approximately 150 vetted news domains
- Sources not in the registry receive a default credibility score of 50/100
- These "unknown" sources are marked as Tier 4 and labeled in the credibility breakdown
- In typical articles, 30-70% of sources may be unknowns depending on topic
Bias Analysis
- When our AI cannot confidently assess bias, articles default to a neutral score of 50
- The "AI Analyzed" indicator shows when bias was actually assessed vs. defaulted
- Source balance is calculated from our registry; unknown sources are excluded from counts
Score Interpretation
- All scores are estimates with inherent uncertainty
- A credibility score of 75 should be interpreted as approximately 70-80, not exactly 75
- Scores reflect our algorithmic assessment, not absolute truth
Trust Indicators
- When AI analysis is unavailable, trust indicators use heuristic pattern matching
- Heuristic-analyzed indicators are labeled as such and should be treated as rough estimates
- AI-analyzed indicators have higher confidence but are still algorithmic assessments
General Limitations
- AI may hallucinate or misinterpret information
- Breaking news may have limited source availability
- Non-English sources may be underrepresented
- Historical context may be incomplete
- Bias detection is imperfect and evolving
- Credibility scores are estimates, not guarantees
Methodology Updates
We continuously improve our methodology. Major changes are documented here:
- January 25, 2026: Added unknown source indicators to credibility breakdown
- January 25, 2026: Fixed Trust Project heuristic matching for more accurate pattern detection
- January 25, 2026: Added explicit "Scoring Limitations" disclosure section
- January 25, 2026: Removed randomization from credibility estimates - all scores now deterministic
- January 25, 2026: Added source bias aggregation from registry for accurate left/center/right counts
- January 24, 2026: Implemented full 3-factor weighted credibility scoring formula (source quality 60%, diversity 20%, verification 20%)
- January 24, 2026: Added source independence detection using media ownership groups
- January 24, 2026: Added bias analysis transparency indicators showing when analysis was performed vs defaulted
- January 24, 2026: Created unified source registry consolidating 99+ rated news sources
- January 2026: Initial methodology published