I’m a relatively new ChatGPT Plus user, and I’ve been using ChatGPT heavily for several large, long-running personal projects. Overall, I find the product extremely useful, but in a fairly short period of time I’ve encountered several recurring UX issues that I think have relatively straightforward solutions.
1. Usage limits need visible, understandable meters.
I understand that different models and tasks consume different amounts of compute, so I don’t expect every request to have the same cost. However, users should be able to see what resources they have and approximately how much remains.
For example:
-
5-hour allowance: 63% remaining / resets or replenishes at X time
-
Weekly allowance: 41% remaining / resets on X date
-
After a substantial Work task: “This task used approximately 6% of your weekly allowance.”
The units can be normalized; users do not need to understand the underlying GPU time, tokens, bandwidth, etc. We just need a consistent gauge so we can plan our usage. Currently, the limits can feel arbitrary because the user sees percentages or warnings without enough information about what they represent or how quickly different activities consume them.
2. Clearly distinguish the different limits.
Chat/model limits, Work limits, short-term usage windows, weekly allowances, context windows, and maximum conversation size are different things, but that distinction is not obvious in the UI. A simple “Usage” page showing each resource separately would eliminate a great deal of confusion.
3. Warn users before a conversation reaches its maximum size.
I have now had two long-running conversations unexpectedly reach their maximum size with no advance warning. Long conversations are particularly valuable because they accumulate context over time.
Even approximate warnings would help:
“Conversation capacity: ~75%”
“Your conversation is becoming large: ~90%. Consider starting a continuation.”
Better still, provide a Start Continuation button that creates a new conversation and automatically carries forward an appropriate summary of the important context, decisions, and unfinished work.
The exact token count is not important. What matters is preventing the user from unexpectedly hitting a wall.
4. Chat, Projects, Work, Library, browser, and app behavior need better integration and clearer boundaries.
I have repeatedly had difficulty understanding which information or files are available in Chat versus Work, when I need to switch modes, and what context follows me when I do. I’ve also encountered differences between the browser and app that make the overall system feel like several related products rather than one integrated workspace.
Transitions should make it explicit: “Work will have access to X, Y, and Z from this conversation/project” and “The following will not carry over.”
5. File/Library operations need better progress and completion reporting.
During a bulk file operation involving 105 items, the interface appeared to stop after 40. Because additional items were loaded dynamically as I scrolled, it was difficult to determine whether the operation was still running, had failed, or which files had actually been processed.
For bulk operations, users need reliable progress such as:
“Moving 63 of 105 files…”
“105 of 105 completed.”
“3 failed — view failures.”
The same principle applies to downloads, OCR, indexing, and other long-running operations: clearly show what is happening, whether it is still running, and when it has actually finished.
6. Approval/confirmation workflows need clearer status.
I have encountered workflows where I approved or confirmed an action and was returned to what appeared to be the same screen, leaving me unsure whether anything had actually happened. Every approval should end with an unambiguous state such as “Approved,” “Processing,” “Completed,” or “Failed,” rather than relying on the user to infer the result from the UI.
The common theme behind all of these issues is visibility.
I don’t expect AI workloads to be perfectly predictable, nor do I expect unlimited usage. I simply want enough information to understand the current state of the system and plan around its limitations.
ChatGPT is capable of doing remarkably complex work, but sometimes the interface provides less information about what it is doing than much simpler software. Exposing resource meters, progress indicators, conversation-capacity warnings, clearer cross-mode boundaries, and explicit completion states would make the product feel substantially more predictable and trustworthy.
None of these suggestions require exposing complicated internal infrastructure to the user. They simply require translating that complexity into understandable user-facing information.
In short: I don’t need to know how the engine works. I need a dashboard.