First, check whether it is actually down
When ChatGPT stops responding, the fastest diagnosis is OpenAI’s official status page at status.openai.com. It lists live incidents and degradations affecting ChatGPT, the API, and related services, and posts updates as engineers work a problem. If that page shows a red or orange incident, the outage is on OpenAI’s side and there is nothing to fix on your end except wait. If it is all green, the trouble is probably local to you.
Confirming a global outage
The status page is authoritative, but a second opinion helps when an incident is just starting and hasn’t been posted yet. Public outage trackers such as Downdetector aggregate user reports, so a sudden spike there is a strong sign the problem is widespread. Searching social platforms for “ChatGPT down” will also surface a wave of posts within minutes of a real outage. Between the official status page and a tracker, you can usually tell within a minute whether to wait it out or troubleshoot your own setup.
What the common errors mean
ChatGPT’s failure messages map to predictable causes:
- 503 / “at capacity” — servers are overloaded. Transient; wait a few minutes and retry. Plus users typically get priority during peaks.
- 429 / “too many requests” — you hit a rate limit or usage cap. Slow down, and on the API add backoff and check quota.
- “A network error occurred” / spinning response — often a client-side or connectivity issue rather than a true outage.
- 5xx (500, 502, 504) — server-side errors or timeouts; retryable.
Recognising which bucket you are in tells you whether to wait, slow down, or fix your own connection.
When it’s just you
If the status page is green but ChatGPT still fails, work through quick local fixes: try an incognito/private window (rules out a bad extension or cached session), clear cookies for the OpenAI domain, switch from Wi-Fi to mobile data or vice versa, disable a VPN that might be geo-blocked, and test on another device. Logging out and back in clears many stale-session problems. These steps take a minute and resolve the majority of “down for me only” cases.
Routing around the outage
The best defence against a ChatGPT outage is having an alternative ready before you need it. Claude, Google Gemini, and Microsoft Copilot cover almost all everyday writing, coding, and reasoning tasks, and Perplexity is strong for cited research. Set up at least one account in advance so a disruption costs you minutes rather than derailing your day. If you run a product on the OpenAI API, the durable fix is provider failover — detect 5xx/429 responses and automatically reroute to a backup model — so an OpenAI incident never becomes your incident.