How it Works

How we do elections research

We use the latest deep research AI models to conduct a thorough, highly sourced synthesis of available sources related to communities, races, candidates and measures.
This research is evaluated both by additional AI models and a team of qualified human researchers with backgrounds in election research, electoral transparency, and journalism.
Researchers review, edit and validate our research before it becomes part of Change.vote’s underlying data corpus.
The research is also consistently updated (with continued AI + human review) as an election approaches.

Diagram: Data and Prompts feed a Deep Research AI step, then an Evaluation AI step, then a Human Review step, then a Diff AI step, then a final Human Review step

How we generate alignment scores

Scores & rationale are produced together to ensure consistent, aligned, informed deduction.
Scores are calibrated to user needs and the type of race to ensure maximum alignment.
Rationales have consistent structures and are optimized for relevance and candidate contrast to maximize voter utility.

Diagram: User Onboarding (Identities, Top Issues, Ideology) and Validated Deep Research (Race Type, Race Level, Experience, Endorsements, Positions) feed an AI step that runs multiple times to ensure accuracy and consistency, producing an Alignment Score and a Rationale covering candidate qualifications and voter relevance