Frequently Asked Questions

It's Political Water Cooler's own rating of how a news source's content leans politically, shown alongside (not instead of) the well-known third-party MBFC (Media Bias/Fact Check) rating. We built it so a source's bias rating isn't dependent on a single outside dataset, and so the reasoning behind every score is something you can actually inspect on that source's rankings page.

For each source, we pull a real sample of up to 40 of its recent Politics, World, or Business articles and send that same sample to four independent AI models (currently from Anthropic, OpenAI, Google, and xAI), each scored on the same -100 (Left) to +100 (Right) scale. The published score is the median of those four independent scores — not any single model's opinion. We restrict the sample to those topics so a source's score reflects its actual political reporting rather than being diluted by sports, entertainment, or other non-political coverage it may also publish. News and opinion content are scored as two separate buckets (see the next question). We won't publish a rating for a source until we have at least 5 qualifying articles to work from.

Plenty of outlets run a straight-news desk and an opinion/editorial section with very different tones — lumping them into one number would blur both. We use our own AI-based content analysis to sort each source's articles into a "news" bucket and an "opinion" bucket first, then run the Clearwater Score independently for each bucket that has enough sample size.

MBFC (Media Bias/Fact Check) is an established, independent third-party dataset we license — we don't control its methodology or its numbers. The Clearwater Score is our own rating, built the way described above. We show both, side by side, on purpose: they're two independent estimates of the same thing, and where they agree that's a stronger signal than either one alone.

Both scores get mapped onto the same -100 (furthest left) to +100 (furthest right) scale and bucketed into those five labels so they're easy to compare side by side at a glance, even though MBFC and the Clearwater Score arrive at their underlying numbers differently. "Center" isn't a value judgment — it just means that source's content doesn't consistently read as leaning either direction.

A human is always in the loop. A source's very first score in either bucket always waits for manual approval — there's no history to compare it against yet, so it always gets a human's eyes. After that, a new score can only auto-publish if it's a small move (within a narrow band of the currently-published score) AND the review didn't flag low confidence; anything bigger, or anything the models themselves weren't confident about, sits in a review queue until it's checked by hand.

On a recurring weekly cycle, automatically. A score can move between cycles if a source's recent article mix shifts — see the previous question for how much human review that involves depending on the size of the move.

Every lean bar runs the same way: blue for left/Democrat-leaning, grey for center or unrated, red for right/Republican-leaning. "MBFC" labels the well-known third-party Media Bias/Fact Check rating; "Clearwater" labels our own AI-generated score (see above for how it's built). A "Subscription" badge just means that particular article may be paywalled at the source. You'll see a collapsed "what do these mean?" key at the top of any page using these indicators — this answer is the same explanation, in full.

Any story that's had activity in the last 3 days qualifies, up to the 100 most relevant. From there, you can toggle between two orderings: "Most Recent" (the story with the newest linked article first — this is the default) or "Most Sources" (the story covered by the most outlets first).

It's the last time we linked a new article into that story — not when the story first appeared. A story that broke three days ago but just got a fresh article linked ten minutes ago will show "Updated 10 min ago," which is the point: it tells you whether a headline is still actively developing or has gone quiet.

We match articles by how similar their actual content is (using AI-generated embeddings), not by simple keyword or title matching. That's what lets us group, say, five different outlets' distinct headlines about the same underlying event into one story, each still linked out to its own original source.

That's not a labeling glitch — when a real story's subject IS a media company or platform (a lawsuit, a ban, an acquisition, a controversy involving that outlet), naming it is just accurate reporting of what happened, the same as any other headline naming its actual subject.

It highlights stories where one side of the political spectrum is covering something heavily while sources on the other side, among the outlets we monitor, haven't touched it at all — the idea being that what a side chooses not to cover can be as telling as what it does cover.

A story from the last 3 days, tagged Politics, World, or Business, where at least 2 of our monitored sources on one side (left or right) have covered it and literally zero of our monitored sources on the other side have.

No — it means none of the specific outlets we currently monitor on that side covered it. We track a real but limited set of sources, not the entire media landscape, and we're careful to word it that way rather than imply we can see everything everyone publishes.

The full, always-current list — along with each one's live MBFC label and Clearwater Score — is on the Source Rankings page. We're regularly adding more outlets across the spectrum, so that list is a better answer than any fixed number we could give here.

No. Political Water Cooler aggregates headlines and brief excerpts from other outlets' publicly available reporting and links directly to the original source for the full article. We don't publish original news reporting of our own.

No. It's an independent aggregator with no ownership ties to, or funding from, any political party, campaign, or the outlets it tracks and rates.

Email us at contact@politicalwatercooler.com with the headline and we'll take a look — corrections like this feed directly into keeping the news/opinion split (and the scores built on it) accurate.

contact@politicalwatercooler.com.