“Error in message stream” appears in ChatGPT when an answer starts to arrive and then stops: the reply is cut off partway, and the error takes its place. ChatGPT sends answers to your screen piece by piece as they are generated, so anything that interrupts that delivery, on OpenAI’s side or yours, can end a reply this way.

The good news is that it is usually temporary. OpenAI’s troubleshooting guide for error messages treats a failed response as something that “may be a one-time error”, and its steps for failed and stuck replies work for this message too.

On this page

Quick fix: try these in order

  1. Regenerate the answer. Click Regenerate (or resend the message). OpenAI lists this as the first step when ChatGPT cannot produce an answer.
  2. Check whether ChatGPT is having an outage. Look at status.openai.com. If there is an incident, nothing on your side will fix it; wait it out.
  3. Start a new chat. OpenAI recommends this when “the conversation is long or has many turns”. Copy your last prompt and send it in a fresh conversation.
  4. Hard refresh or restart the app. Press Cmd/Ctrl + Shift + R in the browser, or close and reopen the desktop or mobile app.
  5. Turn off VPNs, proxies and secure DNS tools. OpenAI names VPNs, proxies and “Web Protect or other security filters” as common causes of connection errors.
  6. Try an incognito window with extensions off. Ad blockers, script blockers and privacy extensions can interfere with the page.
  7. Switch network or device. Moving from office Wi-Fi to a phone hotspot is a quick way to tell whether the network is the problem.

Why the stream breaks

OpenAI’s guide groups the causes of ChatGPT errors into three families: “network configuration, browser extensions, or transient server-side problems”. For a cut-off answer, the likely suspects are:

  • A short-lived problem on OpenAI’s side. These are the most common and clear up on their own, which is why Regenerate usually works.
  • An unstable connection. ChatGPT uses long-running connections to stream replies. OpenAI’s network recommendations note that if a network blocks or times out WebSocket traffic, “some features may stall, disconnect, or fail to stream updates correctly”. Company networks with TLS inspection or strict firewalls are the usual culprits.
  • A very long conversation. Long threads are heavier to process, and OpenAI’s first advice for stuck or slow replies is to start a new chat.
  • Something in the browser. Extensions that block scripts or modify pages, or stale site data, can break the page mid-reply.

If you are on a work network

If the error only appears on your office network, OpenAI’s network article lists the domains ChatGPT needs and says WebSocket traffic over TCP port 443 must be allowed. Ask your IT team to check that ChatGPT’s domains are not filtered and that long connections are not closed early. A quick test: if ChatGPT works on your phone’s mobile data but not on the office Wi-Fi, the network is the cause.

When to contact OpenAI

If the error keeps appearing across different browsers, devices and networks, and status.openai.com shows no incident, OpenAI asks you to contact Support with details: a HAR file and browser console errors with timestamps, the model you used, and the conversation URL. You can reach Support through the chat widget at help.openai.com.