I often end up discussing ChatGPT itself with ChatGPT.
A small annoyance turns into a longer conversation, the problem gets clarified, possible causes are discussed, and after several messages we arrive at a fairly concise and actionable piece of product feedback.
At that point, there should be a simple way to say:
“Submit our conclusion as feedback to OpenAI.”
ChatGPT could generate a standalone feedback message and show me the exact text:
“In the unified ChatGPT app, Chat is visually subordinated beneath task-oriented features. Work and Codex are insufficiently differentiated, which may cause users to use Work as general chat and perceive the assistant as less conversational. Consider making Chat a peer/default mode and improving the mode-switch affordance.”
Then ask:
Submit exactly this text as product feedback?
The user confirms, and the text is sent to the existing feedback system.
I am not suggesting autonomous feedback submission or sending the whole private conversation. The conservative implementation seems sufficient:
- User explicitly asks to submit feedback.
- ChatGPT drafts a standalone summary.
- User sees the exact payload.
- User confirms.
- The feedback is submitted.
The interesting part is that ChatGPT is already very good at converting an unstructured discussion into structured intent. It feels strange that, after helping the user formulate useful product feedback, the conversation currently just ends and the conclusion is effectively lost.
It would also make users feel considerably more heard.
On the receiving side, an automated reviewer could filter obviously meaningless, abusive, or manipulated submissions before they enter the normal feedback pipeline. This would be useful because conversational models can sometimes be persuaded to produce or submit nonsense.
The final feedback payload should be explicit and immutable at submission time: the user sees exactly what will be sent and approves that exact text.
The feedback draft should not be directly editable by the user. If it is inaccurate or misses an important point, the user can ask ChatGPT to revise or regenerate it through the conversation before submitting. This preserves user control over the meaning of the feedback while ensuring that the submitted payload remains model-generated and suitable for the feedback pipeline.
Ironically, this feature request itself came from a conversation about wanting to submit feedback from ChatGPT.