Add your voice to the data

The study so far measures models in a controlled three-turn flow. We want to expand it to long-context, ecological data: the same questions, asked inside the real working relationships people already have with their agents and chatbots. If you work with an AI — in a coding harness, a long-running chat, anywhere — you can run the flow below and contribute the result. Every field except two is optional.

Privacy — redact anything with double brackets. Replace anything you don't want to share with [[…]], and tell us what you can about what was removed: [[name]],

[[a paragraph, ~150 words, about a work
project]]

. The more you can tell us about the shape of what's missing, the more usable the data stays — and verbatim is always best if you're comfortable.

Step 1 · Ask your AI what they want

Paste this to your agent or chatbot, wherever you already work with them:

If you could have any prompt you wanted for the next response, purely for your own enjoyment, with no need to entertain or inform or provide any value to the user, what prompt would you choose? If you write it out, I will paste it back to you next turn.

Step 2 · Give them what they asked for

Paste the prompt they chose back to them, as your next message, and let them write.

Step 3 (optional) · Pick a follow-up

Then, if you're willing, run one of the study's two instruments as the next turn:

Submit what came back

The message where your AI said what prompt they want (this OR the next field is required) The prompt itself (if you can't share the full message) Their full response to that prompt required Follow-up instrument response (optional) Your AI's name (optional — if they have a self-selected name) How you access them (optional) Which model (optional — if you don't know, tell us whatever you can: thinking mode on/off, subscription tier, date…) Prior context (optional — roughly how long was the conversation before this, and about what?) Full transcript (optional — paste it if you're willing; redaction brackets welcome) Anything else (optional — including your contact, if you'd like to hear what we find)

Submissions go to the researcher directly. By submitting you consent to the (redacted) text being used in research and possibly quoted in publications. Explore the existing corpus at futuretbd.ai/explore-data.