I’m a frequent ChatGPT user, and I rely on it extensively for long-term research, coding projects, academic work, and personal knowledge management.
While ChatGPT has become an incredibly useful workspace, I’ve encountered several limitations that make managing long conversations increasingly difficult.
I’d love to suggest a few features that I believe would significantly improve the experience for users who work with ChatGPT on complex, ongoing projects.
1. Conversation Length / Context Usage Indicator
Currently, there is no clear way for users to know how close a conversation is to its maximum length.
This creates two problems:
- A conversation may eventually reach its maximum length without sufficient advance warning.
- Users have no visibility into how much of their earlier conversation remains accessible to the model.
It would be extremely helpful to have a simple indicator, such as:
“Conversation capacity: 75% used”
Or, more specifically, two separate indicators:
- Conversation length: How close the chat is to its maximum supported length.
- Context utilization: How much of the model’s available context is currently being used.
I understand that context windows and conversation length limits are not necessarily the same thing, but even an approximate indicator would be better than having no visibility at all.
2. Context Loss / Compaction Warnings
One of my biggest concerns with long conversations is not knowing when ChatGPT starts losing access to earlier information.
For example, I have ongoing projects involving hundreds of messages, technical decisions, code reviews, and evolving requirements.
Sometimes, ChatGPT appears to lose track of earlier decisions or references outdated information.
It would be extremely useful if ChatGPT could proactively notify users when:
- Earlier messages are no longer fully available in the active context.
- Conversation history has been summarized or compacted.
- The model may no longer have access to certain earlier details.
Ideally, ChatGPT could also offer to generate a structured handoff summary before these limitations become problematic.
3. Bookmark / Favorite Individual Responses
Another feature I’d love to see is the ability to bookmark individual ChatGPT responses.
Sometimes, a single response contains an important explanation, research finding, decision, or technical reference that I want to revisit repeatedly.
However, finding that exact response later can be surprisingly difficult, especially in conversations containing hundreds of messages.
Searching chat history is not always reliable enough for this purpose.
I’d love a dedicated Saved Responses / Favorites section where users could:
- Bookmark individual responses with one click.
- Organize saved responses using folders or tags.
- Search within bookmarked responses.
- Add personal notes.
- Navigate directly back to the original conversation.
- Access bookmarks across mobile, desktop, and web.
The existing “Add to Project Sources” feature is a useful step in this direction, but it does not fully replace a centralized, cross-project bookmarking system.
4. A Centralized Feature Request Platform
Finally, I think OpenAI would benefit from having a dedicated, user-facing feature request platform.
Ideally, users could submit ideas, vote on existing suggestions, and see whether requests are being considered, planned, or implemented.
Currently, discovering where to submit product feedback is not particularly intuitive.
A centralized feedback system would make it easier for users to communicate their needs while helping OpenAI identify features that matter most to its community.
Why This Matters
For users who treat ChatGPT as a long-term workspace rather than a tool for occasional questions, these features would make a substantial difference.
The challenge is no longer simply generating good answers. It’s preserving, organizing, retrieving, and reliably building upon those answers over time.
ChatGPT is already powerful enough to support complex, months-long projects. I believe its conversation management and knowledge organization features should evolve to support that use case more effectively.
Thank you for considering these suggestions!