AI Methodology

How Bill Classification Works

Citizen uses two AI models in our classification pipeline: xAI Grok for initial bill classification, and Anthropic (Claude) for an independent audit of xAI's work. This two-model approach is our automated quality check. Claude does not re-read every bill — it re-reads bills xAI flagged as uncertain, bills whose answers looked internally inconsistent, and a random sample of the rest, against the full bill text, and flags disagreements. Bills where the two models disagree persistently are labeled as disputed in the app.

Classification Labels

• AI-classified

  Standard AI classification with full bill text analysis.

• AI-classified (partial analysis)

  Bill text exceeded processing limits. Classification based on available text.

• Independently audited

  Classification verified by a second AI model.

• Classification under review

  Two AI models disagreed. Under manual review.

• Historical data

  Pre-AI data not yet classified.

Why Your Governor Has No Score

We do not score governors, and we would rather explain why than publish a number we cannot stand behind. The only executive record we hold is which bills a governor signed and which they vetoed — and across every state we track, governors sign about 95% of what reaches them. A score built on that would mostly measure what the legislature chose to send, not what the governor decided.

The veto is where a governor acts on their own, and in most states there are too few of them to score fairly — a typical governor vetoes only a handful of bills in a term. So your governor's profile shows their veto record as a factual list rather than a rating.

One thing to read carefully there: if it shows no vetoes, that means none are recorded in our data for that period. It does not mean the governor vetoed nothing. Some states are less completely covered than others, and we will not render a gap in our data as a fact about a person.