Critical bug report: Data loss and project linking failures in large conversations. Issue occurs after reaching max chat size with no prior warning.

Summary:
I’m reporting a critical issue related to large conversation handling in ChatGPT.
As of April 2025, there appears to be no direct way to submit a bug report through the UI, so I am posting here.


Issue Description:
When a conversation in ChatGPT reaches its maximum allowable size — especially chats containing deep thematic discussions, embedded images, and charts — critical problems occur:

  • Several days’ worth of conversation disappears after leaving and reloading the chat.
  • The conversation can no longer be linked to a project after hitting the limit.
  • An error is shown once maximum size is reached, but no early warning is provided before reaching the limit.
  • Data loss occurs after a browser reload, not immediately.

Observed during the course of a long-running D&D worldbuilding conversation containing detailed lore, images, and structured data.


Severity:
Severity 1 – Critical Data Loss


Reproduction Steps:

  1. Create a large, complex conversation (text-heavy with images and tables).
  2. Continue adding content until the conversation hits maximum size.
  3. At maximum size, an error message will display stating that no further conversation can be added.
  4. Exit and later reload the conversation.
  5. Notice that multiple days’ worth of prior conversation is missing.
  6. Attempt to add the conversation to a project — it fails.

Observed Results:

  • Silent deletion of prior conversation content after browser reload.
  • Conversation cannot be added to projects after hitting max size.
  • Newly added content becomes possible after the loss — suggesting freed storage space.

Expected Results:

  • Proactive early warning when nearing the maximum chat size.
  • No loss of existing conversation content without explicit confirmation.
  • Ability to add large conversations to projects (or at minimum, ability to export them safely).
  • Reliable, non-destructive handling of maximum size limits.

Impact:

  • Data Loss: Several days of detailed project work permanently lost.
  • Workflow Disruption: Severely impacts long-term projects like collaborative writing, RPG campaign development, and research discussions.
  • Reduced User Trust: Users risk losing valuable content without warning.

Reproduction Rate:
Approximately 1 in 2 instances when maximum size is reached.


Environment:

  • ChatGPT Version: GPT-4o
  • Platform: Web (Desktop)
  • Browser: Chrome 135.0.0.0
  • OS: Mac OS X 10.15.7
  • Dark Mode: Enabled
  • Account Type: ChatGPT Plus
  • Timezone: US -0600

Additional Notes:

  • The “Help” (?) button only links to FAQs now — no “send us a message” option currently available.
  • No official bug reporting mechanism currently exists in the ChatGPT web interface.
  • “Thumbs down” feedback appears to be the only direct reporting option from within ChatGPT at this time.

My Background:
I am a professional Quality Assurance Manager with over 30 years of experience in software testing, including working at companies such as Microsoft and Apple.
I would be happy to assist further, provide additional diagnostic information, or reproduce the issue if needed.


Suggested Improvements:

  • Add a proactive early warning before users hit maximum conversation size.
  • Implement a safe archiving/splitting system for large conversations.
  • Allow conversations at maximum size to still be linked to projects.
  • Provide a temporary recovery window if data loss occurs after size limit is reached.

Request to OpenAI Team:
Please escalate this internally, as this constitutes critical data loss and represents a significant risk to long-term project work done through ChatGPT.

Thank you for your attention to this.

(edited to add a workaround section)

Mitigations / Workarounds

  • Manual Early Warning: I implemented a workaround where the assistant provides an alert when approaching large conversation sizes, allowing me to proactively move the discussion to a new thread and/or export history.
  • Export Before Reaching Limit: Manually exporting conversations before suspected limits is currently the only reliable way to mitigate loss.