Let ChatGPT Propose and Set Up Projects From Existing Work

The problem

Long-running use of ChatGPT can naturally generate:

  • multiple related chats;

  • spreadsheets and documents;

  • accumulated operating rules;

  • research and decisions from previous conversations;

  • handover or context documents;

  • recurring tasks that would benefit from ChatGPT Work.

Over time, these materials may clearly form part of the same underlying project or workflow, even though they were originally created across separate chats. This isn’t as much of a problem when I am aware I am starting a large project and set it up from the start. But if a project starts as a chat that develops into something bigger, that is harder to re-structure.

ChatGPT is increasingly capable of understanding these relationships through conversation history, memory and existing context. From use, I find it is already able to recognize that several conversations and files logically belong together and would be better organized inside a Project.

The current limitation is that, even when ChatGPT can identify this structure when prompted about it, it cannot help the user perform a reorganization, nor does it suggest it unless explicitly prompted to analize it.

The user still needs to manually realize a setting a project would be benefitial, create the Project, identify and move each relevant conversation, locate the appropriate files, decide which accumulated rules should become Project instructions, and reconstruct an organizational structure that ChatGPT may already understand. It fact, ChatGPT can walk the user through each of those steps. But it cannot perform the process.

As ChatGPT becomes better at understanding long-term user context, this creates an unnecessary gap between understanding a user’s work and helping the user organize that work.

Proposed feature: Assistant-proposed Project setup

ChatGPT could be able to propose a Project structure based on existing chats, files and recurring workflows. This could be started my either the user through a prompt, or by ChatGPT itself.

For example, ChatGPT could notify the user (a setting that could be toggled off):

“You appear to have an ongoing workflow spread across several conversations and files. I think this would work better as a Project. Here is how I would organize it.”

It could then show a reviewable proposal containing:

Project name

A suggested name based on the shared purpose of the relevant conversations.

Chats to include

  • Initial planning or research

  • Execution or implementation

  • Follow-up work

  • Related analysis

  • Supporting discussions

  • Exceptional cases or decisions

Files / sources

  • Current working files

  • Reference documents

  • Handover or context material

  • Other relevant sources

Proposed Project instructions

A summarized version of stable rules ChatGPT has learned across the relevant conversations, such as research methodology, workflow rules, formatting preferences, decision criteria and exception handling.

The user could then:

Approve · Edit · Remove items · Cancel

Only after explicit approval would ChatGPT make any changes.

Why the approval step matters

ChatGPT should not silently reorganize a user’s workspace.

The useful model is, in my opinion:

ChatGPT proposes; the user authorizes.

The assistant can use its understanding of context to infer structure, while the user remains the final authority over what belongs together and which information should become persistent Project context, or if it should be a Project at all.

This same pattern could later support smaller organizational suggestions, for example:

  • “This rule has now been used consistently across several conversations. Add it to the Project instructions?”

  • “This file appears to have become the authoritative source for this workflow. Make it a Project source?”

  • “These new chats appear to belong to the same existing Project. Move them?”

  • “This conversation has developed into a distinct sub-workstream. Create a separate Project chat for it?”

Again, every structural change would require user approval.

Why this becomes more valuable as ChatGPT improves

Projects solve the problem of persistent workspace organization.

ChatGPT Work increasingly solves the problem of delegating larger pieces of work.

Memory and conversation-history understanding increasingly allow ChatGPT to understand what the user is doing over time.

What remains relatively manual is the layer connecting those capabilities.

The user still has to act as the workspace administrator even when ChatGPT already understands:

  • which conversations belong together;

  • which files matter;

  • which instructions are stable;

  • which source appears authoritative;

  • and what the logical structure of the work has become.

For users who accumulate substantial history, the organizational burden can actually increase as they get more value from ChatGPT.

Allowing ChatGPT to help organize its own working context would reduce that burden.

Concrete user benefit

Today, if a user using ChatGPT reaches this conclussion:

“You should probably create a Project for this workflow.”

that creates another task for the user.

The ideal experience would be for ChatGPT itself to flag it:

“You should probably create a Project for this workflow. I have prepared a proposed structure below. Review it and, if you approve, I can set it up for you.”

This would make Projects much easier to adopt for users who begin in normal Chat, accumulate substantial useful history, and only later realize that their conversations have evolved into persistent workflows.

It could also make the transition between Chat, Projects and ChatGPT Work much more natural and increase usage/adoption rates of those features: instead of requiring users to recognize in advance which organizational structure they should use, ChatGPT could help them adopt the appropriate structure as their work develops.

The broader principle could also be useful beyond Projects:

ChatGPT should be able to propose improvements to its own working environment when it has enough context to recognize them, while requiring explicit user approval before making structural changes.

This progression feels very familiar to me — especially the part where something starts as a normal chat and only later becomes obvious that it has turned into a real project.

I went through something similar with my research. At first I moved the work into Projects and had ChatGPT maintain a kind of master ledger in the Project files, so different chats could share the same current state.

That worked fairly well, but eventually keeping the shared file updated manually became another job of its own. That was what pushed me toward keeping the evolving state outside the individual chats instead, while still having the AI help decide what should be carried forward.

I also really like your “ChatGPT proposes; the user authorizes” distinction.

Do you currently use any kind of handover/master document as a workaround when one of your chats grows into a project, or are you mostly reorganizing the chats/files manually?

Yes, that is almost exactly the workaround I use.

I rely quite heavily on handover/master documents once a chat gets too long or the work starts branching across several chats. I usually ask ChatGPT to capture the relevant context, current state, specific decisions, working rules, instructions, open questions, and sometimes the files/data the next chat will need.

It works surprisingly well. In some cases I’ve effectively built small project-management systems around those handovers. But, as you said, maintaining that shared state becomes a task in itself: deciding when the handover is outdated, updating it, making sure new chats inherit the latest version, and avoiding losing decisions made in between.

That is a big part of what motivated my post.

ChatGPT already seems to have many of the pieces needed to recognize that several chats belong to the same evolving body of work and that some information should become persistent Project context.

Instead of periodically asking it to create another handover, I’d like ChatGPT to propose: “These seem to be the current project instructions/state/files. Should I create or update a Project with them?” Then the user reviews and authorizes it. In fact, the very message about creating a handover within the chat could easily become a trigger for ChatGPT to analyze if it may be time to propose a Project.

So I think we’ve arrived at essentially the same “problem”. The handover approach works, but it also demonstrates why an extra layer would be useful. Most of the functionality already exists. It feels like connecting a few dots is the only missing step.

Yes — I think we’re looking at almost exactly the same problem.

I started with handover/master documents too, then moved the shared project state outside the chats into Git. That gives each chat a common durable source instead of relying on whichever handover happens to be current.

But Git only solves the storage/versioning part. The harder question is still what should become durable state, and when. I like your idea of ChatGPT recognizing that boundary, proposing an update, and letting the user review and authorize it.

A handover request itself may be a very good trigger for that.