Every Side Of The Story

The challenge

What needed solving

Most people only ever see the version of a story that agrees with them. When one side covers something heavily and the other barely touches it, that gap is itself information — and so is a change in framing between two accounts of the same council vote, the same conscription bill, the same set of numbers. The tools that make those differences legible exist, but they are split in two and neither half is free. Ground News does the story-level comparison — how the left, the centre and the right each covered one event — behind a monthly subscription. Media Bias/Fact Check does the outlet-level lookup — is this publication biased, and is it factually reliable — as a reference site with its own paid tier and no story comparison at all. A reader who wants both answers is paying twice, or doing the work manually.

We took the position that media literacy is a bad thing to paywall. If knowing how coverage splits is genuinely useful — and it is — then the barrier to seeing it should be a browser tab, not a card on file. So the brief we set ourselves was to build the comparison product and give it away, which is a much harder engineering problem than it sounds: a subscription business can afford a beautiful feed nobody finds, because it lives on retention. A free one lives on organic search, which means every page has to answer a question somebody is actually typing into Google, and the whole thing has to be cheap enough per story to run without a revenue floor underneath it.

The second constraint was credibility. A site whose entire pitch is "we show you what other people leave out" cannot ship placeholder widgets, invented ratings or a coverage percentage with no stated denominator. Every number on the page had to be traceable to something, or it could not be on the page.

The solution

What we built

mediabias.news is live. It gives every major story a single page holding four things: a neutral summary written from the source reporting, three sections showing what the left, the centre and the right each emphasised and each left out, a panel showing exactly which outlets ran it grouped by leaning, and a full list of the original articles with direct links out. A reader who lands from search gets in one screen what would otherwise take them six tabs and a working knowledge of who owns what.

The editorial pipeline runs in a fixed order and never grades its own homework. A first pass reads only the original source articles and produces three published scores — Trust, Craft and Hype — built by counting specific visible evidence in the text rather than asking a model for an impression. A second pass writes an original article from the isolated facts. A third writes the headline, constrained so our own headline can never read as more sensational than the Hype score we just published for the story. Every draft is checked mechanically against its sources on overlapping ten-word runs before it can publish, and anything over a 12% match is discarded rather than edited. Nothing goes out unsupervised: editors write and version the instructions the models follow, review the drafts, and are named as accountable on the site's own masthead.

The bias leanings themselves are deliberately not ours to invent. The outlet ratings come from the established third-party organisations readers already trust — AllSides, Ad Fontes Media, Media Bias/Fact Check — and are shown side by side, including where those organisations disagree with each other. Our own scores measure the reporting in front of us; the political placement of an outlet is somebody else's published judgment, credited as such.

That combination is what makes the site a genuine free alternative to both incumbents at once. The story pages do what a Ground News subscription does — the coverage split, the blindspots, what each side omitted — and the outlet profiles answer the question people currently go to Media Bias/Fact Check for, with the added benefit of showing all three raters together instead of one verdict. Same two jobs, one site, nothing to pay and nothing to sign up for.

Because the product had to earn its traffic, search architecture was designed in from the first commit rather than added later. Alongside the story pages there are 2,400+ individual outlet profiles answering the single highest-intent question in this category — "is [outlet] biased" — plus hub pages for left-leaning, centrist and right-leaning sources, a free standalone bias checker tool, and a Blindspots section that indexes the stories one side of the spectrum has almost entirely ignored. That is 1,600 URLs in the live sitemap at launch, each one aimed at a real query rather than generated for volume.

Key features

Under the hood