Methodology
How Polyview works
1. Collecting the news
Every 10 minutes we read the public RSS and Atom feeds of every outlet in our index. We keep the headline, a short excerpt, the publication time and a link — never the full article. Reading the story always means visiting the publisher.
Articles older than 72 hours drop out of the live index.
2. Grouping articles into stories
Articles are grouped into a story when they describe the same event. We represent each article as a weighted bag of words (headline words count more than excerpt words, and rare words count more than common ones), then build clusters around the articles most similar to their neighbours. An article joins a story only if it is similar to the story as a whole — not just to one other article — which stops loosely related stories from chaining together.
A story needs at least two different outlets. Each outlet counts once per story, so a publisher that files ten updates doesn't outweigh one that files one. The story's headline is taken from the most representative article, preferring centre-rated outlets with strong factuality.
3. Bias ratings
Each outlet is placed on a seven-point scale. Ratings reflect the outlet's news coverage — story selection, framing, word choice and sourcing — plus the weight of its opinion section. Lean is relative to the outlet's home country: a “Lean Right” paper in the UK sits against British politics, not American.
- Far LeftOpenly ideological left: advocacy or movement media whose story selection and framing consistently serve a left-wing agenda.
- LeftReliably progressive editorial line and framing, often with prominent opinion content.
- Lean LeftMainstream outlets whose story choices and wording tilt modestly left of their country's centre.
- CenterWire services, public broadcasters and outlets with balanced sourcing, neutral language and little opinion in news coverage.
- Lean RightMainstream outlets whose story choices and wording tilt modestly right of their country's centre.
- RightReliably conservative editorial line and framing, often with prominent opinion content.
- Far RightOpenly ideological right: advocacy or movement media whose story selection and framing consistently serve a right-wing agenda.
For coverage bars we fold the scale into three groups: Left (Far Left, Left, Lean Left), Center and Right (Lean Right, Right, Far Right). A story's bar shows the share of distinct outlets in each group.
4. Factuality ratings
Separately from lean, every outlet gets a reliability rating based on its sourcing, corrections practice and fact-check record.
- Very High
- Rigorous sourcing, transparent corrections, almost never subject to failed fact checks.
- High
- Generally well sourced with a corrections policy; rare factual problems.
- Mostly Factual
- Usually accurate but with occasional misleading headlines, thin sourcing or uncorrected errors.
- Mixed
- A pattern of misleading framing, selective facts or failed fact checks alongside accurate reporting.
- Low
- Frequent misinformation, conspiracy content or fabricated claims.
5. Ownership
Who pays for the newsroom shapes what it covers. We classify every outlet by ownership type and name the owner on its profile page.
- Independent
- Media Conglomerate
- Private Equity
- Billionaire-owned
- Family-owned
- Non-profit
- Public Broadcaster
- State-funded
- Intergovernmental
State-funded outlets are flagged as such regardless of how independent their editorial charter claims to be; public broadcasters with statutory editorial independence are listed separately.
6. Blindspots
A story is a blindspot for the Left when at least four left- or right-rated outlets have covered it and 20% or fewer of those are left-leaning; the reverse makes it a blindspot for the Right. Centre-rated outlets don't count towards the calculation, because they cover most stories regardless of which side cares.
7. AI briefings
Stories covered by five or more outlets get a short AI-written briefing: a neutral summary, how left, center and right outlets framed the story, the facts they agree on, and where they differ. The model sees only the headlines and excerpts shown on the story page, is instructed not to add outside facts, and its output is checked against a strict format before publishing. Briefings are refreshed as coverage grows.
AI can still get things wrong or miss nuance. Every briefing is labelled, names the AI service that wrote it, and sits above the original reporting so you can check it yourself.
8. Limitations
Ratings are judgements, not measurements, and outlets change over time. Our index over-represents English-language and Western media, and an outlet with a small RSS feed can be under-counted. Automatic grouping occasionally merges two related events or splits one. We would rather tell you this than pretend otherwise.
9. Disagree with a rating?
Good. Write to ratings@polyview.news with examples of coverage that supports a different rating. We review every submission and publish changes on the outlet's profile.