Expected Parrot

Expected Parrot is a research platform for running surveys and experiments with AI respondents alongside real people. At its center is an open source Python package that lets a researcher define questions, attach agent personas with demographic traits, choose which language models answer, and run the whole instrument in parallel, then analyze the output as a structured dataset. The intended use is testing assumptions and question wording cheaply before, or alongside, fieldwork with humans. It is written for academic and commercial researchers, survey designers and teams doing large scale text labeling. The package provides many question types, scenarios for parameterized tasks such as labeling a large corpus, surveys with rules and conditional logic, comparison of answers across models, and built in results analysis with cost estimation, token tracking and exception handling. The hosted service adds remote inference, automatic caching of results and API calls so anyone given the code reproduces the exact outputs without keys or setup, notebook sharing, a file store and a no code survey builder. For human validation it can turn a survey into a web form for real respondents and launch studies on Prolific through an integration. No prices are published. The pricing page describes free accounts for eligible researchers, credits bought as you go, the option to use your own model provider keys instead, and credits earned through referrals. Enterprise terms cover shared workspaces, shared credits with usage controls, dedicated support and custom integrations, and are arranged by contacting the company. Worth knowing before adopting it: the package runs entirely on your own provider keys without the hosted service, but caching, sharing, the survey builder and the human respondent features depend on the platform account.