How do people respond to an AI opponent, an AI advisor, or an AI member of their team?
Do they trust it, follow it, cooperate with it, or get persuaded by it?
To answer these questions you need experiments where real participants
interact with an LLM, live, inside a carefully controlled design.
oTree, with over
3,300 citations,
is already the standard platform for interactive experiments,
and LLMs fit right into it.
With oTree AI (OTAI),
you describe your study in plain English,
and an AI agent builds and tests it for you, usually in minutes.
There is no coding.
What you can build
- Games against AI: prisoner's dilemma, bargaining, public goods, auctions,
or any game of your own, where the AI player's moves and chat messages come from an LLM.
- AI advisors: participants make decisions with live advice from an AI,
to study trust, advice-taking, and overreliance.
- Conversations with AI: persuasion, deliberation, or customer-service chatbots,
with your own system prompt and your own choice of model.
- AI in live groups: several human participants plus an AI teammate, opponent, or moderator,
all in the same real-time chat or game.
- AI-generated content: stimuli, feedback, or follow-up questions generated on the fly
from each participant's earlier answers.
You're not limited to a menu of preset stages.
Your rules, payoffs, roles, branching, and treatments can be whatever your design requires.
See it for yourself
Our live demo site
has examples you can play right now
(or import into your own OTAI project and customize):
- Prisoner's dilemma vs AI: play several rounds against an LLM, and chat with it between decisions.
- AI advisor: an investment task with live advice from an AI.
- Chat with AI: a simple conversation, the starting point for persuasion and chatbot studies.
Why oTree?
LLM calls are built on top of a mature framework,
so the rest of your study comes for free:
- Any model: OpenAI, Anthropic, Google, DeepSeek, or a self-hosted open-weights model.
Anything with a web API works.
- Real-time multiplayer: lobbies that group participants as they arrive,
chat, live pages, timeouts, and dropout handling.
- Fully tested: bots play through your whole study automatically,
with the AI's replies simulated, so you can test every branch
(e.g. "what if the AI defects?") without waiting for or paying for real API calls.
- Every message saved: the full conversation between each participant and the AI
is stored with the rest of your experiment data.
- Live monitoring of every participant's progress and responses.
- Recruitment and payment through Prolific and other platforms.
- Hosting is built in, with no software to install.
Participant messages are sent only to the model provider you choose,
using your own API key.