How It Works
NewsDog uses AI to score every article from 13 global news channels on a 0–100 objectivity scale, updated every 6 hours. Here's exactly how each score is calculated.
The Pipeline
- 1
Fetch RSS feeds
Every 6 hours the pipeline pulls the latest articles from each channel's RSS feed. Duplicate URLs are skipped.
- 2
Analyse with AI
Each headline and lead image is sent to Claude (claude-haiku-4-5) with a structured prompt that returns four numeric scores and a short reasoning note.
- 3
Compute overall score
Headline scores (loaded language + framing + factuality) contribute 75%. Image sentiment contributes 25%. If no image is available, the headline score becomes 100%.
- 4
Aggregate daily scores
Per-channel, per-topic daily averages are written to the DailyScore table. The leaderboard and topic pages read from these aggregates.
The Four Dimensions
Each dimension is scored 0–100 independently, then combined into the overall score.
Does the headline use emotionally charged words ('slaughter', 'heroic', 'regime') rather than neutral ones? High scores reward plain, factual word choice.
Does the headline frame events in a way that favours one side, omits key context, or uses rhetorical questions designed to lead the reader? High scores reward balanced framing.
Are the claims in the headline verifiable and specific, or vague and speculative? High scores reward claims that are attributed, sourced, or independently verifiable.
Does the lead image amplify an emotional reaction (crowds, victims, dramatic faces) disproportionate to the story? High scores reward neutral, informative imagery.
Score Bands
Coverage Bias Alert
⚠ Coverage Bias
This alert fires when a channel publishes 2× or more articles on a topic compared to the peer average, and scores below 40 / 100 on that topic. The combination of heavy coverage and low objectivity suggests the topic is being used for editorial amplification rather than neutral reporting.
Directional Lean
On topic pages, each channel also shows a directional lean — the perspective its headlines most frequently favour when covering that topic, for example “pro-government”, “pro-Palestinian”, or “anti-establishment”. This is derived from the AI's bias label on each article and aggregated over the last 30 days. A label of neutral means the AI found no consistent directional pattern.
Caveats & Limitations
- Scores reflect headlines and images only — not the full article body.
- AI scoring can be inconsistent on nuanced language; treat scores as approximate signals, not ground truth.
- RSS feeds vary in freshness and article count; some channels update more frequently than others.
- Three channels (Reuters, AP News, RT) are currently unavailable due to network restrictions.
- The pipeline runs every 6 hours, so scores lag slightly behind breaking news.