Feature Request: Prune intermediate conversation branches while preserving the final result

[caroline.detry]

As a user working iteratively on long-form projects — such as fiction, role-playing scenarios, books, specifications, research, or software design —

I want to be able to remove a selected range of intermediate messages from an existing conversation

so that I can keep only the final, validated version of an idea without carrying hundreds of obsolete iterations throughout the rest of the conversation.

Current problem

Many creative and analytical workflows are iterative.

A typical workflow looks like this:

1. I propose an idea.
2. ChatGPT responds.
3. I modify the idea.
4. ChatGPT gives feedback.
5. We refine it several times.
6. Eventually, we reach a final version.
7. I then move on to another part of the project.

Later, a new idea may affect something created earlier, so we return to that section and iterate again.

After weeks or months, the conversation becomes extremely long, even though a large part of it consists of intermediate drafts, rejected ideas, corrections, and obsolete versions that are no longer useful.

For example, ten or twenty exchanges may have been necessary to produce one final scene or concept, but once that scene is validated I may only want to keep the final version.

Starting a new conversation is not a good solution because the broader project context is still useful.

Proposed feature

Allow the user to select a start message and an end message inside a conversation and choose something like:

“Prune this section”

The selected messages would be removed from the active conversation history.

The user could therefore reduce:

Idea → response → correction → response → correction → response → final version

to simply:

Final version

while keeping the rest of the conversation intact.

Optional but valuable behavior

Before removing the selected section, ChatGPT could offer:

• Delete completely
• Replace with a short summary
• Keep only the final selected message

The “replace with summary” option would be particularly useful when some information from the iterative process still matters to the later conversation.

For example:

Pruned section summary:
After several iterations, the characters’ motivation was changed from X to Y, and the final scene is the version below.

This would preserve semantic continuity without preserving every intermediate exchange.

Important requirement

The pruned messages should no longer be part of the active context used by ChatGPT.

Otherwise the feature would only improve visual readability without solving the underlying context-management problem.

The remaining conversation — and optionally the generated summary — should become the effective context going forward.

Example use case

I am writing a role-playing campaign.

I spend 25 messages developing a scene involving an NPC.

During those messages I explore several possibilities, reject most of them, change motivations, rewrite dialogue, and eventually settle on the definitive version.

Once that version is accepted, the earlier 24 messages are no longer useful.

I want to select those messages and prune them, leaving only the final version in the conversation.

I can then continue developing the campaign without carrying every discarded iteration with me.

Acceptance criteria

1. A user can select a contiguous range of messages in an existing conversation.
2. The user can remove that range without deleting the entire conversation.
3. The user can preview what will be removed.
4. The user can optionally replace the removed section with an AI-generated summary.
5. The user can choose to keep one or more final messages from the selected range.
6. Pruned content is removed from the active model context.
7. The rest of the conversation remains unchanged.
8. The operation can ideally be undone for a limited period.

Why this matters

ChatGPT is increasingly used for projects that span days, weeks, or months.

In these workflows, conversation length is often caused not by the amount of final information, but by the number of iterations required to reach that information.

Giving users the ability to prune obsolete iterations would make long-running conversations:

• easier to navigate,
• more coherent,
• less cluttered,
• more efficient in their use of context,
• and much more suitable for serious long-term creative and analytical work.

Conceptually, this would turn a ChatGPT conversation from a permanent transcript into something closer to a living working document whose history can be consolidated as the project evolves.