Summary:
Introduce a simple way for users to react to individual parts (paragraphs, bullet points, steps) of an AI’s response using lightweight reactions such as Like 
, Agree 
, Disagree 
, Confused 
, and Want More 
.
This would dramatically improve feedback quality, model training precision, and user engagement.
Benefits:
Richer, more detailed feedback per response.
Smarter model improvement without intrusive surveys.
Higher user satisfaction — feeling heard at a fine level.
Easier discovery of strengths and weaknesses inside responses.
(Drafted by collaboration between ChatGPT and user proposal, 2025.)
a first step would be to allow to reference multiple quotes from previous responses
because reactions (upvote/downvote/explain) can be expressed in text (yes/no/what)
like here in discourse:
i can select a range of text, and click “quote” or “copy quote”
in chatGPT:
currently i can select a range of text and click “ask chatGPT”
to quote a previous response, but that works only once
so a workaround is to use markdown quotes (> …)
to give feedback on parts of a response