AI Transparency

Where we use AI

AI helps us keep party positions current across ~100 policy questions. Specifically we use large language models to:

• Monitor sources. Detect meaningful changes in published party platforms, press releases, and policy documents.
• Map content to questions. Suggest which of our questions a policy update might affect.
• Draft proposed positions. Produce a structured proposal with the source quote, the suggested score, and a plain-language rationale.

Where we do NOT use AI

• We do not use AI to generate your personal report content beyond templated text.
• We do not sell, share, or train external AI models on your answers or email.
• We do not use AI to auto-publish any party position without human review.
• We do not use AI to profile you, target you, or infer sensitive attributes.

Human in the loop

Every AI-generated proposal enters a review queue. A human editor checks the source, the mapping, and the score before it goes live. Nothing an AI drafts affects your report until a human has approved it.

Models we use

We currently use OpenAI GPT-class models via the Lovable AI Gateway. Models may change over time as better options become available; the review process above stays the same.

Known limitations

• AI can misread ambiguous or satirical content and propose the wrong score.
• AI can miss nuance in complex or conditional policy positions.
• AI does not "understand" politics — it pattern-matches text.

The human review step is designed to catch these errors, but no process is perfect. If you spot something wrong, please tell us.

Report an AI error

Visit our contact page with a link to the source and the position you believe is inaccurate. We investigate every report.