Research Survey

How does phrasing change AI responses?

A blinded comparison of AI-generated outputs. Same task, two different ways of asking. You decide which response works better.

5
About 5 minutes
3-5 task ratings. Stop anytime.

What you'll do

  1. Read a short task description.
  2. Compare two AI-generated responses to that task.
  3. Rate them on three quick scales: which is more useful, which is more approachable, which you prefer overall.
  4. Repeat for as many tasks as you'd like.

What this is for

I'm submitting an essay to the John Locke Essay Competition (UK, judged by Oxford academics) on whether the way we phrase requests to AI changes the quality of what we get back. Your ratings are the empirical evidence in the essay.

The data is anonymous. Results appear in aggregate only ("X of N evaluators preferred…").

One submission per device. No personal information collected.
Estimated time per task: 60-90 seconds.

Task 1 0 rated
Task
A Response A
B Response B
Which is more useful for the task?
More accurate, complete, and actually addresses what was asked.
Which is more approachable?
Easier to read, better tone, more pleasant to receive.
Overall, which response do you prefer?
All things considered, which would you rather have received?

Thank you

Your ratings have been recorded. They go directly into the empirical section of the essay.

0
tasks rated
0s
avg time per task

If you opted into acknowledgments, your display name will appear in the essay if it's accepted by the competition. If you have questions about the study, ask Radek (G10).

Loading tasks…