Methodology

How GroundedVote Works

Transparency is the foundation of trust. Every element of the GroundedVote methodology is designed to be auditable — by civic partners, academic reviewers, election officials, and users.

Phase 01

Candidate Data Collection

Candidate position data is collected from publicly available sources using a network of state-level automated agents. Primary sources are ranked by reliability, with official government records given the highest weight.

1st

Official government records and verified voting history

2nd

Candidate-submitted questionnaires and official campaign platforms

3rd

Verified media interviews and recorded public statements

4th

Third-party aggregators with documented sourcing methodology

Phase 02

Bias-Audited Question Generation

The core intellectual contribution of the platform.

Pass 1: Generation

A primary language model receives a candidate's stated position and generates three to five alternative question framings using only behavioral language: "Would you support a policy that..." Party names, political labels, and ideologically coded language are excluded from all generated framings.

Pass 2: Bias Audit

A second language model from a different AI provider scores each question variant across four bias dimensions: loaded language, false equivalence, asymmetric framing, and embedded cultural assumption. The second model does not know which variant was generated first. Variants above the threshold are flagged for rewrite.

Pass 3: Neutrality Selection

A third pass — either a third model or human editorial review for high-stakes races — selects the highest-scoring neutral variant. All variants and their bias scores are archived. This archive constitutes the public audit trail for the methodology.

Phase 03

Participant Matching

Users complete a weighted issue quiz. Questions are calibrated to real population impact — not media salience. The issues covered reflect what most affects the lives of most voters, not what generates the most coverage.

The result is a civic mirror: a clear picture of what the participant actually believes, independent of party, tribe, or fear — followed by a reveal of which candidates most closely match those beliefs.

All results include full methodology transparency so users can evaluate the process, not just the output.

Ethical Framework

Built for auditability. Designed to earn trust.

Nonpartisan by Design

Party names, political labels, and ideologically coded language are excluded from all question framings. The methodology is designed to interrupt tribal activation, not reinforce it.

Transparent Archive

Every question variant, every bias score, and every selection decision is archived. Academic reviewers, civic partners, and election officials can audit the methodology.

Built for the Underserved

The tool is designed to be accessible to someone who has never followed an election closely, engaging enough to hold attention through completion, and trustworthy enough to share.

Funders, reviewers, and civic partners welcome.

The full whitepaper is available for academic reviewers, election officials, and civic organizations.

Request Whitepaper