Hi team,
I’ve noticed that when attaching files in ChatGPT, the system starts uploading them right away and immediately counts their sizes toward my total attachment quota—even if I later decide not to send the message or remove the file.
From what I’ve observed, this behavior happens because the upload completes before message submission, and the server adds that size to the quota as soon as it receives the file. While this makes uploads feel faster, it also means that users lose part of their quota on files that were never actually sent.
I think a more user-aligned approach would be to implement deferred quota accounting, where uploaded files are flagged as pending until the user commits the message.
Proposed Flow:
-
When a file is attached, it uploads as normal but is marked as
pending. -
Once the user sends the message, those files are marked as
committed, and only then should their sizes count toward the quota. -
If the user cancels or removes the file before sending, the pending uploads are discarded and never counted.
This small change would make quota usage feel more transparent and fair, especially for users who test, draft, or handle large projects frequently. It also keeps performance intact since uploads still happen ahead of time—the only change is when they’re accounted for.
Benefits:
-
Prevents unintentional quota depletion.
-
Reflects user intent accurately.
-
Maintains upload speed and UX.
-
Simple to implement (just a
pending→committedflag switch and timed cleanup for abandoned uploads).
This would be a small but meaningful quality-of-life improvement that respects user control and system efficiency at the same time.
Thanks,
Orland Benniedict D. Sayson