Beyond Binary User Response

ChatGPT Feature requests
1

TO: OpenAI Product Development Team
SUBJECT: Feature Request — Introduce a Sliding Scale for ChatGPT Response Feedback

Proposal: Graduated Response Feedback

The Problem

ChatGPT feedback mechanisms that rely on positive or negative ratings do not fully capture the range of user experiences. A response can be partially useful, nearly correct, or excellent in some respects but still require significant refinement. Binary feedback does not adequately communicate these distinctions.

The Proposed Solution

Introduce a graduated feedback scale that allows users to rate how effectively a response meets their needs.

For example, a seven-point scale could range from:

  1. Completely missed the mark

  2. Poor

  3. Partially useful

  4. Almost there

  5. Good

  6. Very good

  7. Excellent — exactly what I needed

An optional follow-up could allow users to identify the reason for their rating, such as accuracy, relevance, completeness, execution, clarity, or creativity.

Expected Benefits

  • More nuanced information about user satisfaction.

  • Better identification of specific shortcomings.

  • More efficient response refinement.

  • Improved evaluation of ChatGPT’s usefulness across different tasks.

  • Potentially more actionable data for product development and model improvement.

Implementation Consideration

The feedback interface should remain simple and optional. Users should be able to rate a response quickly and provide additional details only when they wish.

Feedback could be used to evaluate product performance, identify recurring issues, and inform future improvements. Any effect on model training or personalization would depend on the feedback systems and policies implemented by OpenAI.

Conclusion

I recommend that OpenAI consider replacing or supplementing binary response ratings with a graduated feedback mechanism.

The central principle is simple: Instead of asking only whether a response was good or bad, ask how well it accomplished the user’s objective and what could make it better.

This relatively small interface change could provide more meaningful user feedback and support a more effective cycle of evaluation and refinement.