Our Methodology

The Core Principle

Every candidate position on Vote26 must be traceable to a primary source — a direct quote, a recorded vote, or a verified public statement. We never infer positions from party affiliation.

Five-Stage Pipeline

1. Race Discovery. We query the FEC API and state Secretary of State databases to build a candidate master list for each race. No AI involvement at this stage.
2. Position Extraction. For each candidate, an AI agent performs targeted searches on 8–10 core issues and extracts only direct quotes or verifiable actions — with source URLs. No position is published without a URL.
3. Comparison Content Generation. AI generates side-by-side comparison content. We enforce equal word count (within 10%) across all candidates in a race and run a loaded-vocabulary filter that blocks terms like "extreme," "radical," "dangerous," and "flip-flop."
4. Human Review. A human editor reviews every stage's output before it is published. No AI-generated political content goes live without human sign-off.
5. Ongoing Refresh. A weekly automated job monitors for candidate updates and flags them for human review before any changes publish.

Confidence Scoring

Every candidate position carries a confidence score:

• High — Direct quote or recorded vote with primary source URL
• Medium — Two or more corroborating secondary sources
• Low — Single source or inferred position; excluded from published pages or labeled "Unverified"

What We Don't Do

• Infer positions from party affiliation
• Publish AI-generated content without human review
• Use loaded or evaluative vocabulary
• Share individual voter data with campaigns or parties
• Accept sponsored or paid candidate content