Methodology · Bias Scoring

How We Score Bias

Every article on Updat3 comes from a real outlet with a real editorial lean. We don't hide that — we show it. Here's exactly how the scoring works.

Step 1 — Outlet lean score

Each outlet has a lean score from −1.0 (hard left) to +1.0 (hard right), with 0.0 representing wire-service neutrality (Reuters, AP, AFP). Each rating is an estimate of an outlet's observed editorial orientation and framing, produced by Updat3. It is not a measure of accuracy or reliability, and it is not a third-party rating. Earlier versions of this page cited external media-bias research; ratings derived from that research are no longer shown.

A rating describes an outlet's established editorial orientation. It is resolved once per outlet and is the same on every story. How a particular article frames a story is not blended into the outlet's rating: a single article is not evidence of where an outlet sits.

Ratings describe where an outlet usually sits politically. They say nothing about whether a particular report is accurate, and Center is not the same as correct: every rating, including Center, is a description of the outlet, not a verdict on the story.

Political orientation is a separate question from state affiliation, government ownership or editorial independence. Outlets controlled or directed by a government — any government, including US-funded international broadcasters — and official government or agency sites are not placed on the Left–Right axis, because there is no defensible way to map them onto it. They are shown as Unrated and still count as outlets covering the story.

Ratings are Updat3's own estimates, produced by an automated classifier that is reviewed against a sample of the outlet's coverage in our own corpus. No third-party bias rating is used as an input. A rating is used only when the classifier's confidence is at least 60%; otherwise the outlet stays Unrated. Ratings are re-reviewed periodically.

Example outlet scores

A representative sample, resolved live exactly as a story page resolves it:

State-affiliated examples — Unrated

Outlets we have not rated are Unrated and left out of the spectrum (never counted as Center). New outlets are reviewed as they appear in our coverage.

Step 2 — One vote per outlet: Left, Center or Right

Each distinct outlet covering a story is counted once, however many articles it published, and placed in a bucket by its rating. Outlets we have not rated are left out rather than counted as Center.

Too few rated outlets to judge: a coverage mix is only described once at least 3 of the outlets covering a story carry a rating.

Outlet rating grades

Coverage mix labels (per story)

Step 3 — Where the marker sits

Each unique outlet is placed Left, Center (within ±0.12 of 0) or Right from its rating, and counts once — no matter how many articles it published or how strongly it is rated. The marker shows the balance of those outlets: dead centre when Left and Right are equally represented, fully left or right when only one side covered it. Outlets we have not rated are left out rather than counted as Center.

A story covered only by right-leaning outlets does not mean the event is false — it means our current coverage of it skews right. More sources improve accuracy.

Step 4 — Bias Gap

Wide spread of viewpoints means the rated outlets covering this story genuinely span the spectrum: at least 2 rated outlets sit Left and at least 2 sit Right, and the outermost outlets' ratings differ by at least 0.5 on our −1 (left) to +1 (right) scale. It highlights divergent source perspectives, where framing may differ. A story with fewer than 3 rated outlets reads "Too few rated" instead of a coverage mix label, and gets no spread note: that is too few ratings to judge. The coverage bar beside it shows how many of the rated outlets sit Left, Center and Right.

When all sources cluster on one side of the spectrum, the story summary may reflect the framing assumptions of that side. We aim to add sources from the other side over time.

Unrated outlets

Outlets we cannot place on the Left–Right axis: ones we have not rated yet, state-affiliated and government sources, and automated ratings below our confidence threshold. They count toward the number of outlets covering a story but are left out of the Left/Center/Right split — never counted as Center. Unrated is preferable to false precision.

Methodology changelog

Version 2026.10.1 · Last updated October 14, 2026

• Oct 14, 2026A coverage mix label (Left Only, Center Only, Right Only, Mostly Left, Mostly Center, Mostly Right, or Mixed; the thresholds are in the label table above) is only shown once at least 3 outlets are rated; below that the story reads "Too few rated". "Wide spread of viewpoints" now requires at least 2 rated outlets on each of the Left and Right, not just two ratings 0.5 apart.
• Sep 28, 2026All outlet ratings are now Updat3's own estimates from its automated classifier, sampled against the outlet's coverage in our corpus. A legacy table derived from third-party media-bias research was retired; outlets it alone covered read Unrated until reviewed.
• Sep 26, 2026State-affiliated and government sources are Unrated rather than scored on the Left–Right axis. Outlet ratings no longer blend in how individual articles are framed. Automated ratings below the confidence threshold are not used.
• Sep 26, 2026Outlets we have not rated are shown as a separate “Unrated” group and never counted as Center. Outlets whose rating could not be determined are no longer stored with a default neutral score.
• Sep 26, 2026The marker shows coverage balance (each rated outlet counts once) instead of the average outlet rating.
• Sep 25, 2026One Center band (within ±0.12 of 0) is used on every page, card, share image and API. Publishers linked through Google News are identified individually.

What bias scoring does not do