When ChatGPT will not load, a conversation disappears, or a response fails mid-task, do not assume the whole OpenAI platform is down. The problem may affect all ChatGPT conversations, one client or device, one conversation, a particular feature, or your local browser session.
How to tell whether ChatGPT is actually down.
| What you see | Likely scope | Best first check |
|---|---|---|
| Many conversations will not load or continue | ChatGPT service incident | Official OpenAI incident and ChatGPT component |
| Only one conversation fails | Conversation-specific error or stale thread | Open a fresh chat without deleting the failed one |
| 403 or login errors on one device | Client, authentication, or platform-specific incident | OpenAI status plus another signed-in surface |
| ChatGPT loads but rejects more work | Usage limit, model availability, or account issue | The exact message and plan/reset information |
Recent OpenAI incidents match the searches users make.
OpenAI documented July incidents where some users could not load or continue ChatGPT conversations. Another incident affected conversation and login access on iOS and macOS, including 403 errors. These reports show why "ChatGPT down" can mean a broad outage, a client-specific failure, or an individual conversation that stopped working.
What to do when ChatGPT is down mid-task.
- Check OpenAI Status and the affected ChatGPT component.
- Stop duplicate retries if the previous request may still have executed.
- Preserve any code, files, notes, and terminal output outside the conversation.
- Record the last confirmed action and the first missing response.
- Label unverified claims from the interrupted reply as claims, not facts.
- Wait, restart in a fresh ChatGPT conversation, or continue in another AI coding tool.
A chat that will not load is not a reliable project record.
A transcript can contain important intent, but it is also verbose, full of superseded attempts, and controlled by the provider. For coding work, the durable state should include the repository diff, changed files, decisions, constraints, verification output, failed attempts, and the next safe action.
How ShardStitch helps when ChatGPT stops.
ShardStitch combines local-first project memory, recovery, verification labels, dependency-risk context, failed-attempt history, and cross-tool routing. It creates a continuation packet that can be used in ChatGPT after recovery or handed to a different supported tool.
Browser-only ChatGPT conversation recovery requires a prior capture or hook for transcript-specific details. Repository state and project memory that already exist locally remain available even when ChatGPT itself is unavailable.