ChatGPT Error Messages: Every Common Error Explained + Fast Fixes

If ChatGPT throws Something went wrong, A network error occurred, Conversation not found, Error in message stream, too many concurrent requests, Unable to load history, Error in moderation, or just sits on an endless spinner, the fix is usually simpler than the error text makes it sound. Most failures fall into four buckets: a temporary OpenAI incident, a stale local session, a blocked network path, or a request that is too large, too sensitive, or too messy for the current chat state.

This version is fact-checked against OpenAI’s official ChatGPT troubleshooting guide, the OpenAI status page, current ChatGPT model notes, and recurring user reports in the OpenAI community.

The fastest ChatGPT troubleshooting table

The one-minute ChatGPT fix routine

1. Check status.openai.com first. OpenAI’s history regularly shows ChatGPT incidents involving login problems, elevated error rates, VPN-related access issues, and web failures. If there is a live incident, your fix is usually to wait, not to keep refreshing.
2. Start a fresh chat. OpenAI’s own help article recommends refreshing the page or starting a new chat for the broadest class of failures, including Something went wrong.
3. Hard refresh or sign out and back in. This clears a surprising number of broken sessions, especially when the page is stuck, blank, or endlessly thinking.
4. Disable VPNs, proxies, secure DNS tools, and aggressive extensions. OpenAI explicitly calls these out for network errors, suspicious-activity alerts, and page-loading failures.
5. Make the request smaller. Split giant prompts, remove unnecessary attachments, or upload a document instead of pasting a wall of text into a tired conversation.
6. If it still fails across browsers, devices, and networks, collect diagnostics and contact support. OpenAI asks for HAR files, browser-console errors, timestamps, and the conversation URL or ID when persistent issues survive the normal fixes.

1) “Something went wrong”

This is ChatGPT’s most generic error, so treat it like a symptom, not a diagnosis. OpenAI’s official advice is straightforward: refresh the page or start a new chat, check the status page, clear browser cache and cookies, use an incognito window, disable extensions, turn off VPNs, and try a different network if needed. That is the correct troubleshooting order because the message can come from either your local setup or OpenAI’s side.

Fast fix: start a new chat first. If the new chat works, your old session was the problem. If the same prompt fails in a new chat, shorten the prompt, remove attachments, and check for a live incident before you touch anything else.

2) “A network error occurred” or websocket errors

OpenAI treats A network error occurred and An error occurred while connecting to the websocket as connection-path problems between your device and ChatGPT. The official fixes are to disable VPNs or proxies, turn off security filters, try another browser or incognito mode, and switch to another device or network. In plain English: the problem is often the route between you and ChatGPT, not necessarily your Wi-Fi itself.

Fast fix: if the error appears on long answers, ask for a shorter response or split the task into steps. If it appears before ChatGPT really starts, suspect a VPN, proxy, security filter, or browser extension first.

3) “Conversation not found”, “Unable to load conversation”, and “Unable to load history”

These are the errors that trigger the most panic because they look like data loss. Usually, they are sync and retrieval problems, not proof that your chat is gone forever. OpenAI’s current history docs say older chats can fall out of the fast-loading sidebar cache without being deleted, and that searching or reopening them forces a full fetch. That matters because it explains why a conversation can feel missing in one part of the UI but still exist in your account.

User reports add an important wrinkle. In one OpenAI community thread, Temporary Chat users reported a Conversation not found error on the second turn, which later appeared to resolve as a service-side bug. In other threads, users described Unable to load conversation or Unable to load history symptoms that sometimes improved after logging out and back in, updating the browser, disabling a proxy, or waiting through an outage window.

4) “There was a problem preparing your chat”

This one is more specific than it sounds. OpenAI says it usually happens when a browser extension or URL blocker interferes with page loading. Translation: the chat is not even getting set up properly, so fixing your prompt will not help.

Fast fix: open ChatGPT in an incognito window, disable script-blocking or privacy extensions, and retry there first. If incognito works, your browser setup was the cause.

5) “There was an error generating a response”, endless “Thinking”, or a frozen spinner

OpenAI documents this as a failed response and recommends the following order: click Regenerate, start a new chat, hard refresh, sign out and back in, test incognito mode with extensions disabled, turn off VPNs or proxies, switch browser or network, and then contact support if it persists. OpenAI has a separate article for slow or stuck chats that adds one more useful test: switch to a different model and try the same task in a brand-new chat to figure out whether the problem is model-specific or account-wide.

Fast fix: stop the generation, regenerate once, and if it happens again do not keep hammering the same broken thread. Open a fresh chat and retry the task there. If only one model is failing, switch models and come back later.

6) “Error in message stream”

This is one of those errors that sounds more mysterious than it really is. In practice, it usually means the response stream broke after ChatGPT had already started the run. That can happen because the model lane is having issues, the chat thread is unstable, the browser or desktop app lost the stream, or a file or image attachment made the request more fragile. OpenAI does not currently publish a dedicated help page for this exact message, so the right troubleshooting path is borrowed from the errors it does document: regenerate once, start a new chat, test another browser or model, disable VPNs and extensions, and check the status page first. Community reports add two useful patterns: the error has shown up as model-specific for some users, and attachment-heavy chats — especially with images — can trigger it more often. Full guide · OpenAI help · Community report · Attachment report

7) “ChatGPT is at capacity right now”, elevated errors, and login failures

These are the easiest errors to misread as local problems. OpenAI’s public status history regularly logs ChatGPT incidents for elevated error rates, login failures, empty responses, and VPN-related access problems. That means the right question is often not What is wrong with my browser? but Is ChatGPT degraded for a lot of people right now?

Fast fix: check status first. If there is a live incident, stop retry-spamming. Retrying every few seconds does not move you to the front of anything. Wait for the incident to clear and then reopen ChatGPT in a fresh tab.

8) “Too many concurrent requests”

This error is usually about overlap, not mystery. ChatGPT thinks too many requests are trying to run at the same time, whether because you have multiple tabs generating, you clicked retry too many times, or OpenAI is under enough load that the service is throttling hard. OpenAI’s API-side rate-limit docs make an important point that still applies conceptually here: short bursts can trigger request pressure, and blindly resending failed requests does not help. Community reports also show paid users getting the same message during broader service strain, so do not assume it is always your fault. Full guide · OpenAI 429 doc · Community report

9) “You’ve reached your limit” or “message limit reached”

This is not a bug. It is a plan or model cap. As of April 2026, OpenAI says Free users can send up to 10 messages with GPT-5.3 every 5 hours, while Plus and Go users can send up to 160 messages with GPT-5.3 every 3 hours before falling back to a mini model. Paid users can also manually choose GPT-5.4 Thinking, which has its own usage rules, and Pro or Business plans get much looser access subject to guardrails.

Fast fix: read the reset notice, then either wait it out or switch to the lighter fallback model. If you are doing heavy work, do not waste premium messages on repeated failures inside a broken chat.

10) “The message you submitted was too long”

This error is real, but the exact threshold is not a clean public number because it depends on the model, plan, file mix, and state of the current chat. OpenAI’s March 2026 release notes quietly improved one part of this by converting pastes longer than 5,000 characters into attachments for Plus, Pro, and Business users, which helps keep giant pastes from eating the whole composer. Even so, users still report message too long warnings in overloaded or unstable chats.

Fast fix: split the input, upload the file instead of pasting it, or start a fresh chat before retrying. If you hit the warning suddenly on prompts that used to work, treat it as a possible temporary bug, not proof that your account changed forever.

11) “Error in moderation”, suspicious-activity warnings, and false positives

OpenAI says ChatGPT checks prompts, completions, and uploads with automated safety systems, and its official troubleshooting page separately documents We detect suspicious activity or Unusual Activity Detected messages. Those are not the same thing, but they live in the same practical neighborhood: your request or your traffic pattern got flagged.

Community reports suggest benign prompts can still trigger moderation friction, especially when they combine multiple sensitive themes or ambiguous phrasing. The practical move is not to rewrite the entire prompt from scratch. Rewrite the specific sentence that might look risky, add harmless context about your intent, and remove stacked trigger terms if you can. If the problem is a suspicious-activity warning rather than a content warning, kill the VPN or proxy first.

12) “Download failed”, “File not found”, blank screens, and frozen pages

OpenAI has concrete guidance here. For failed downloads, generated files can expire quickly, and custom GPT downloads depend on Code Interpreter or Data Analysis being enabled. For blank screens, frozen pages, and endless loading, OpenAI recommends a hard refresh, sign-out/sign-in, clearing site data, disabling extensions, testing incognito mode, and turning off VPNs or secure DNS tools.

Fast fix: if you need the file, regenerate it and download it immediately. If the entire page is broken, clear site data for ChatGPT rather than doing random browser rituals and hoping something sticks.

What users keep reporting in social and community threads

Three user-reported patterns matter because they fill in gaps that the official help docs do not spell out. First, Conversation not found sometimes hit Temporary Chats on the second message during a bug window. Second, Unable to load conversation and Unable to load history often looked account-ending at first, but many users eventually got access back after re-authenticating, disabling proxy settings, updating the browser, or waiting out a service issue. Third, the message too long warning sometimes behaved like a temporary product bug rather than a permanent rules change.

Reddit reports point in the same direction. People tend to describe these issues as sudden, account-wide, and catastrophic, but the threads often reveal a simpler pattern: the chat becomes unreachable in one surface, keeps working in another, or comes back after the next service recovery. That does not make the errors less annoying. It just means the right first move is usually verification, not panic.

Current model note: GPT-5.3, not GPT-4o

This matters because a lot of ChatGPT troubleshooting posts are already stale. OpenAI retired GPT-4o, GPT-4.1, GPT-4.1 mini, o4-mini, and the old GPT-5 models from ChatGPT in February 2026. In ChatGPT today, the default logged-in experience is GPT-5.3 Instant, with GPT-5.4 Thinking and GPT-5.4 Pro for deeper work depending on plan. So if you are reading an older fix article that tells you to switch back to GPT-4o, that advice no longer matches the current product.

ChatGPT error FAQ

Is there a ChatGPT status page?

Yes: status.openai.com. Bookmark it. It is the highest-leverage troubleshooting step because it tells you whether you are debugging yourself or an outage.

Why do older chats disappear from the sidebar?

OpenAI says the sidebar keeps a compact recent list for speed and reliability. Older chats can drop out of that quick cache without being deleted, and search or opening the conversation forces a full fetch. Source

Does Temporary Chat save my chats?

No. Temporary Chats do not appear in history, do not use memory, and are not used to train OpenAI’s models, though OpenAI may keep a copy for up to 30 days for safety purposes.

What is the default ChatGPT model right now?

For logged-in users, the default is GPT-5.3 Instant. Depending on plan, you can also use GPT-5.4 Thinking or GPT-5.4 Pro.

Does clearing cache always fix ChatGPT?

No. It helps when the problem is local session corruption or bad site data. It does nothing when OpenAI is having an active incident or when you hit a real plan limit. Refresh once, then diagnose the actual category of failure.

When should I contact OpenAI support?

After the problem survives a new chat, hard refresh, sign-out/sign-in, extension-free browser test, VPN-off test, and a second network or device. At that point, collect HAR logs, console errors, timestamps, and the chat URL or ID so support has something useful to work with.

More specific ChatGPT fixes

• ChatGPT Error Messages: Every Common Error Explained + Fast Fixes
• ChatGPT “Something Went Wrong” Error: Fastest Fixes That Actually Work
• ChatGPT Network Error: What It Actually Means and How to Fix It
• ChatGPT “Conversation Not Found” Error: Causes and Fixes
• ChatGPT Is at Capacity: What to Do When the Site Is Overloaded
• Too Many Concurrent Requests ChatGPT: Causes, Fixes, and What It Actually Means
• ChatGPT Message Limit Reached: Free, Plus, and Pro Fixes
• Error in Message Stream ChatGPT: Causes, Fixes, and What It Actually Means
• ChatGPT Error in Moderation: Why It Happens and Workarounds

Last updated: April 2026.