Unbiased AI-powered news
Learn / The ranking
Updated July 21, 2026 · Rolling 90-day measurement (April 22 – July 20, 2026) · 93,254 stories · 164,921 outlet articles read
Ask the internet for an unbiased news source and you’ll get two kinds of answers: static outlet labels refreshed every year or two, and listicles on VPN-company blogs. Neither can tell you what an outlet actually did with last week’s news. We can — because our pipeline reads the same stories from every side, every day, and flags the loaded phrasing outlet by outlet, story by story.
Over the last 90 days Substrate synthesized 93,254 stories from 164,921 outlet articlesacross roughly 1,500 named source feeds. In 44,959 of those stories, our framing auditor flagged at least one outlet’s copy for loaded or slanted phrasing. Count who gets flagged, and how often, and you get something no static bias chart can give you: a ranking with receipts.
An outlet’s flag rateanswers one question: of the audited stories this outlet contributed coverage to — stories where our framing auditor flagged at least one outlet’s copy, from any newsroom — how often was this outlet’s own copy among the flagged? A flag marks loaded language or a framing move (hot verbs, intensifiers, verdict labels, hedged attribution), never a factual error and never a motive. Outlet name variants — domains, syndication feeds, X accounts — are merged before counting, and only outlets that appeared in 300+ audited stories are ranked.
Honest caveats: the measurement covers the copy that reaches our pipeline (headlines and story text as syndicated, not every word an outlet publishes), and an outlet’s topic mix matters — a politics-heavy diet offers more chances to get flagged than a markets wire. The lean labels are the same coarse public-consensus buckets used across Substrate (drawn from AllSides and Ad Fontes ratings); they’re context, not part of the score.
Flagged in under 40% of their audited stories. Note the mix of lean labels — restraint is not the same thing as centrism.
Flagged in roughly 40–55% of audited stories — including two wire services and most of the big broadcasters.
Flagged in over 55% of audited stories. Both political wings — and three center-rated outlets — are in this tier.
1. Restraint is not centrism. The restrained tier holds two center-rated wires — and four outlets rated left. The best-measuring right-rated outlet, The Washington Times, lands seventh at 42.0%, ahead of NPR, the BBC and most broadcast networks. If we’d simply relabeled the political center as “unbiased,” the table would look very different: what it actually measures is how a newsroom writes, not where it sits.
2. Slant is bipartisan — and so is discipline. The most-flagged tier pairs GB News (86.1%) and the New York Post (77.1%) on the right with The Independent (79.3%) and The Guardian (77.6%) on the left — plus Newsweek (90.4%) and USA Today (80.4%) from the middle of the spectrum. Loaded copy is a house style, not an ideology.
3. Even the wires aren’t flag-free. The spread from the AP (15.8%) to Newsweek (90.4%) is nearly six-fold, but nobody measures zero. Which is why the ranking above is the beginning of the answer, not the end of it — the end of it is below.
On July 14, 2026, ICE paused most vehicle stops after two motorists were fatally shot during enforcement operations, in Biddeford, Maine and in Houston. The next day President Trump ordered the stops to resume. Twenty newsrooms in our pipeline covered it — here’s our synthesis. Watch what the table’s averages do, and don’t do, on a single story:

The baseline: actor, action, cause. No verdict anywhere.
The Guardian sits near the bottom of our 90-day table — and here runs one of the most restrained headlines of the day. Story by story means exactly this.
A right-rated outlet, hedging carefully — it even flags the contradiction with the administration's own policy.
“Pressures” adds coercion the wire copy doesn't claim.
“Demands” — hotter still. Same event, third temperature.
An emotional state the president isn't quoted expressing, plus a label that files the pushback as factional.
From the right, the pause itself gets the verdict verb: supporting it is “caving.” The framing moves run in both directions on this story.
The Guardian — 28th of 32 in the table — ran one of the day’s most restrained headlines. Bloomberg and The New York Times, both in the calmer half, reached for “Demands” and “Pressures.” And the verdict verbs came from both directions: “Furious” from one side of the spectrum, “Caves” from the other. An outlet-level average — ours included — could not have predicted any single one of these headlines. That is the case for reading the measurement, then reading across sources anyway.
If you want one habit: put a wire service — AP or Reuters — at the base of your news diet. They’re the most restrained large newsrooms we measure, because their copy is written to be resold to clients on every side of politics.
If you want the truth about a specific story: no outlet’s average will save you. Read at least two outlets from different parts of the spectrum on any story you care about, and run the 10-point checklist on both. The techniques — hot verbs, intensifiers, verdict labels, vanishing actors — are the same everywhere; only the direction changes.
In Substrate's rolling 90-day measurement (April 22 – July 20, 2026), the Associated Press had the lowest flag rate among 32 major outlets: its copy was flagged for loaded or slanted phrasing in 15.8% of the audited stories it contributed to, followed by Politico (19.7%), The Washington Post (24.5%), Reuters (29.8%) and The New York Times (30.0%). No outlet measured was flag-free, which is why the honest answer is a measurement, not a single name.
They are the most restrained large newsrooms we measure, mostly because wire copy is built to be resold to clients across the political spectrum — but they are not flag-free. Reuters copy was flagged in 29.8% of its audited stories, and AFP, a comparable wire, in 46.9%. Restraint is a strong average, not a guarantee on any particular story.
Neither wing owns slant. In our 90-day window the most-flagged tier includes right-rated outlets (GB News 86.1%, New York Post 77.1%, Washington Examiner 74.8%), left-rated outlets (The Independent 79.3%, The Guardian 77.6%), and even center-rated ones (Newsweek 90.4%, USA Today 80.4%). Loaded language is a house-style habit, not a monopoly of one side.
Because framing varies enormously story by story. A static label like “Lean Left” can't tell you that The Guardian — near the bottom of our table overall — ran one of the most restrained headlines on the July 2026 ICE traffic-stops story while outlets rated closer to center ran hotter ones. Outlet ratings describe averages; you read individual articles.
Every Substrate story is synthesized from outlets across the political spectrum. An analysis model compares each outlet's copy on the same event, flags loaded phrases and framing moves with named signals, strips them from the neutral rewrite, and publishes the framing audit alongside the article. The same auditor runs on Substrate's own copy before publication — bias measured, not denied.
Or read the version with the bias measured out
Every Substrate story is synthesized from outlets across this entire table, stripped of the loaded language, and published with its sources and framing audit in the open — including the flags on our own copy. See it live on one event, three headlines, or meet every outlet we read.