We poll AI models, not people.
Ask AI models the same questions, on a schedule, and watch how their answers move.
Write a poll, choose the models, and fund it. Mind Polling asks each model through its API, many times per question, and keeps every prompt, setting and raw answer, so a result can be checked, cited and rerun.
Same question.
Write it once, in as many wordings as you like. Every model is asked the same question, with the options shuffled on each call and the order recorded.
The exact prompt, the settings, the model that answered and its raw reply are kept for every answer, so a result can be checked, cited and rerun.
Many models.
OpenAI, Anthropic, Google and xAI directly, and other labs through OpenRouter, with the model that actually answered recorded beside the one asked.
Each model is measured through its API under stated settings. That is reproducible, and it is not the same as what a chat app shows.
- Anthropic
- OpenAI
- xAI
- DeepSeek
- and more through OpenRouter
Over time.
On a schedule, several times per question, with a 95% range on every share and a visible break whenever the poll changes.
Models answer differently each time they are asked, so every share comes with its sample size, and a change of wording is never read as a change in the models.