What ChatGPT's error messages mean, from "Something went wrong" to 429
Ten ChatGPT errors, one card each: whether the fault is on your side or OpenAI's, and what to try first. Wording checked on OpenAI's help pages.
A ChatGPT error starts either on your side or at OpenAI. The ones on your side (your connection, a VPN, a browser extension) usually clear in a minute. The ones at OpenAI leave you two options: wait, or work somewhere else for a while. Each card below says which kind you are looking at.
The wording is taken from OpenAI's own troubleshooting page and from users' reports, checked on 2 October 2026. OpenAI rewords these messages from time to time, so match on the gist.
If you cannot tell which side you are on, OpenAI's network guide suggests a quick test. Leave the Wi-Fi and try again on your phone's mobile data or hotspot. If the error goes away, your network was the problem.
"Something went wrong"
What you see: "Something went wrong", sometimes with "while generating the response" added, or "There was an error generating a response."
Whose end: either.
What it means: a catch-all. OpenAI describes it as a temporary server problem or something in your local setup.
What to do: press Retry or Regenerate first, since OpenAI says it may be a one-time error. Then refresh or open a new chat. If it keeps happening, try a private window, turn off extensions (especially privacy and security ones), switch off your VPN, and check ChatGPT status.
MultiChats puts the leading AI models in one app. Try it free
"Error in message stream"
What you see: the reply starts, sometimes runs for minutes, then stops with "Error in message stream".
Whose end: usually theirs, or the conversation itself.
What it means: the reply was arriving piece by piece and the flow broke partway. OpenAI's troubleshooting page skips this one. Its release notes for 27 February 2026 do record a fix for "increased message streaming errors in GPT-5.2 Thinking", so the fault can sit on OpenAI's side. In September 2026, forum users reported it on Android, web and desktop, mostly on long requests with extended thinking and in long chats with attachments.
What to do:
Retry once.
If you picked a longer thinking setting, try a shorter one.
Break the request into smaller steps.
If one long chat keeps failing, ask it for a short summary, then paste that into a new chat.
"A network error occurred"
What you see: "A network error occurred" or "An error occurred while connecting to the websocket."
Whose end: usually yours.
What it means: your device cannot hold a steady connection to OpenAI's servers. Long answers are more exposed. The connection stays open while the reply streams in, and the longer that takes, the more chances a shaky network has to drop it.
What to do: switch networks, turn off your VPN and any web filter, try another browser. If it only happens on long answers, ask for the answer in parts. On company Wi-Fi, talk to IT: OpenAI says ChatGPT can stall when a network blocks the connection type it uses for live updates.
"This chatgpt.com page can't be found"
What you see: a plain browser page with that heading, often with "HTTP ERROR 404".
Whose end: depends on the link.
What it means: this is Chrome's own page for a server answering "not found". Either the address points at nothing (a deleted chat, a removed shared link, a typo) or chatgpt.com itself is misbehaving. On 3 September 2026, OfficeChai reported users getting exactly this page on the chatgpt.com homepage while OpenAI's status page showed everything as operational.
What to do: type chatgpt.com by hand. If the homepage loads, the old link is dead. If the homepage shows the same page, it is an outage and you can only wait.
"Too many requests"
What you see: "Too many requests. You're making requests too quickly. We've temporarily limited access to your conversations to protect your data. Please wait a few minutes before trying again."
Whose end: theirs, applied to your account.
What it means: a temporary block on opening chats. Paying subscribers reported it in August 2026 during ordinary use, without any automation, and the reports did not pin down what triggered it.
What to do: wait the few minutes it asks for. Refreshing over and over only adds requests. One person who reported it had ChatGPT open in the desktop app and the browser at once. Nobody has shown that causes it, but closing the extra copy costs nothing.
"Unusual activity"
What you see: "We detect suspicious activity", "Unusual Activity Detected", or on iPhone "Unusual activity has been detected from your device. Try again later."
Whose end: usually yours.
What it means: OpenAI's systems think your traffic looks automated. A VPN or proxy can cause it, OpenAI says, by routing you through addresses already flagged.
What to do: turn off the VPN or proxy, restart the browser or app, try a private window. OpenAI also suggests checking that nobody else uses your account and changing your password. On iPhone it adds signing out and back in, and updating or reinstalling the app. If none of that works and you use no automation, contact OpenAI support.
"There was a problem preparing your chat"
What you see: that line, before the chat loads.
Whose end: usually yours.
What it means: a browser extension or URL blocker is interfering with the page.
What to do: turn off script blockers and unfamiliar extensions, or open ChatGPT in a private window.
Image generation errors
What you see: one of three different things.
Whose end: depends on which one.
What it means:
"Something went wrong while generating your image" or "Image generation failed. Try again." means the image never finished. Sometimes the fault is OpenAI's: in June 2026, users on Plus and Pro in several countries reported images stalling and failing for days, and OpenAI's support team said on 22 June that a fix had been deployed.
A notice that you have reached your image limit, with an upgrade button, means everything is working. OpenAI's free tier FAQ says image creation has its own limits, separate from chat.
A reply saying the request goes against the content policy is a refusal of that particular request.
What to do: for failures, retry later. If it lasts, clear cookies and cache and sign back in, as OpenAI's support team suggested in that June thread. If it still fails for you alone, support treats it as possibly account-specific, so email them. For a limit, wait for the reset. For a refusal, rephrase; if it names a real person or a famous character, describe an original one.
"The conversation is too long"
What you see: "The conversation is too long, please start a new one."
Whose end: neither.
What it means: the chat has hit its maximum length and cannot take more messages. OpenAI support's advice on its forum, in June 2026, was to start a new conversation. Some users found that very long old chats would not even open; support replied that a chat at the limit should normally still open for reading.
What to do: if the chat still shows, copy the parts you need. Start a new chat with a short summary of the decisions, the latest draft and what you want next. If an old chat will not open at all, contact OpenAI support with its link.
A bare error code: 403, 429 or 500
What you see: a number, sometimes on a plain browser page.
Whose end: codes in the 400s mean the server turned your request away; the 500s are a fault on OpenAI's side.
What it means:
403 (Forbidden): the server refused you. If Chrome shows its own error page, the heading reads "Access to chatgpt.com was denied". A VPN or a company filter is the usual suspect. OpenAI also only supports access from a list of countries.
429 (Too Many Requests): you are being rate limited. Same advice as the "Too many requests" card.
500, 502 or 503: something broke on OpenAI's servers. Chrome's own page for these reads "This page isn't working".
What to do: for 403, turn off the VPN and try another network. For 429, wait. For the 500s, check status and wait.
When it is not you
If the error survives a new chat, a second browser and mobile data, the fault is at OpenAI and no card above will fix it from your end.
If the task cannot wait, our guide on what to use when ChatGPT is down covers moving it elsewhere. Inside MultiChats you can switch the same chat to a Claude or Gemini model. Most of our GPT models run through OpenAI, so a wider OpenAI outage can take them down too.
To see when it is fixed, our ChatGPT status page shows OpenAI's official feed and its recent incidents, and status.openai.com is the source itself. Read "operational" as OpenAI's view of things, which can trail yours: on 3 September 2026 it stayed green through the 404s in the card above.