Score any text for political bias
Paste an X post, a headline, or a few paragraphs of an article. Two AI models — Claude and Grok — read it independently and score its political framing on the same −5 (far left) to +5 (far right) scale we use to track 56 news outlets every day. Where they disagree is often the interesting part.
What is actually being scored
The models score language, not politics. They look for the linguistic signals that mark a piece of writing as left- or right-framed: loaded word choice (“undocumented” vs “illegal alien”), charged verbs, who is cast as victim and who as threat, which sources are treated as trustworthy, and what gets emphasised versus buried. Subject matter alone moves nothing — a post about immigration is not left or right because of its topic, only because of its wording.
Why two models
Claude and Grok are given an identical rubric and never see each other's answers. Two independent readings make a single model's quirks visible: when both land in the same place, the framing is unambiguous; when they diverge by more than a point or two, the text is genuinely ambiguous — or one model is reacting to something the other ignored. The Model Wars tab tracks the same disagreement across every outlet we follow.
Limits worth knowing
A score is a computational estimate of linguistic patterns, not a fact-check and not a verdict on whether the text is true, fair, or worth reading. Short inputs are harder to score than long ones — a six-word headline gives the models very little to work with, which is why each verdict carries a confidence rating. Nothing you paste is stored.