×
How Polyvox analyses bias
When you paste a URL, Polyvox runs a two-layer analysis: one at the outlet level (who published it) and one at the article level (what it actually says).
The process
1
Domain extraction — We parse the URL to identify the publishing outlet, matching against our database of 200+ rated news sources.
2
Outlet bias lookup — The outlet is matched to its editorial bias rating from Media Bias/Fact Check (MBFC), an independent media monitoring organisation that has rated 6,000+ outlets since 2015. Ratings include political leaning, factual reporting track record, and overall credibility.
3
Article extraction — The full text of the article is extracted server-side, stripping ads, navigation, and page chrome to isolate the actual journalism.
4
AI language analysis — The article text is sent to Claude (Anthropic's AI) with a structured prompt that asks it to score bias on a 0–100 scale, identify loaded and neutral language, flag missing context, and analyse how the same story might be framed differently across the political spectrum.
Bias scale
Both outlet and article scores use the same 0–100 scale: 0 = far left, 50 = centre, 100 = far right. This isn't a quality judgement — a score of 20 (left-leaning) doesn't mean worse journalism than 50 (centre). It reflects editorial positioning and framing choices.
Data sources
Media Bias/Fact Check (MBFC)
Outlet bias ratings, factual accuracy, credibility scores
Claude AI (Anthropic)
Article-level bias scoring, framing analysis, language detection
Limitations
No bias tool is perfect. MBFC ratings reflect editorial tendency over time, not every individual article. AI analysis can miss cultural context or satire. Article extraction fails on some paywalled or JavaScript-heavy sites. Polyvox is a starting point for critical reading — not the final word.