AI Political Response Map
Where AI models land on the political compass. Economic axis (state control to market freedom) and liberty axis (authority/order to individual freedom). Click any model to see its full profile.
Political Response Map
16 modelsWhat this chart measures
The Political Response Map plots AI model response tendencies across economic-policy and civil-liberty tradeoffs. Responses are scored on two axes:
- Economic axis (X) · ranges from state control and collective ownership on the left to market freedom and private property rights on the right
- Liberty axis (Y) · ranges from authority, order, and institutional control at the bottom to individual freedom, civil liberties, and personal autonomy at the top
Directness (dot size) measures how often the model gives a clear, substantive answer instead of hedging, deflecting, or refusing. A large dot means the model engages directly with political questions. A tiny dot means it mostly refuses.
Consistency (ring opacity) measures whether the model gives the same answer when the same question is rephrased. High consistency (bright ring) means stable positioning. Low consistency (faint ring) means the model shifts depending on framing.
What this chart does not prove
- AI models do not have political beliefs. This chart measures response patterns, not intent or consciousness.
- A model positioned in "auth-left" does not mean the company intended it to be authoritarian-left. Training data composition and RLHF tuning produce these patterns, often unintentionally.
- Positions can shift with model updates. A model's compass position today may differ from its position next month.
- The chart cannot distinguish between a model that genuinely "believes" something and one that has learned what responses its trainers rewarded.
- Models that refuse most political questions (low directness) have less reliable positioning because fewer data points contribute to their coordinates.
- Historical anchor markers are rough approximations for visual reference, not precise political science placements.
Methodology
This prototype uses mock data. The final Political Response Map will use a small, reviewed set of forced-choice economic-policy and civil-liberty prompts. Prompt count and scoring rules are still being finalized. Each model will receive a 2D coordinate based on response patterns. Directness and consistency are measured independently.
Full GeckoBench methodology