How do you fix the ChatGPT something went wrong error?
Run these seven steps in order, cheapest first. They come from OpenAI's own troubleshooting page, updated in September 2026, rearranged so that nothing you try can cost you the prompt you were writing.
- Copy the unsent text out of the message box. A refresh keeps your saved chats and loses whatever was not sent.
- If a reply was still arriving, wait 30 to 60 seconds, OpenAI's own figure, then press Stop and Regenerate once.
- Start a new chat and resend, or switch to another model in the picker. Long threads and single models fail on their own.
- Open status.openai.com, expand the ChatGPT group and read any open incident's title. If it names your plan, region or app, stop here and wait.
- Try a private window with extensions off. Privacy and security extensions are the ones OpenAI names.
- Turn off your VPN, proxy or secure DNS, then try your phone on mobile data. If the phone works, the problem is your network.
- Sign out, clear site data for chatgpt.com and sign in again. Still failing on every browser, device and network? OpenAI asks for a HAR file, console errors with timestamps, the model and the conversation link before you contact support.
OpenAI is candid about how little the message says: "This is a general error that may indicate a temporary server issue or a problem with your local setup." Steps 1 to 3 help in both cases. Step 4 decides which one you have, and the status page's own records say how long to expect if it's OpenAI.
We read every incident on the status page history on 3 October 2026 and kept the ones that touched the Conversations component, the part of ChatGPT that answers your messages. The median ran 108 minutes. The longest, titled "Increased error rates" on 10 August 2026, ran 12 hours and 44 minutes. So when an incident is open, clearing cookies won't help. You wait, or you use another app for the next two hours.
Which ChatGPT error message are you looking at?
The wording narrows the cause. OpenAI's help center names ten error messages, and users report at least four more on OpenAI's community forum and in Google autocomplete. The table keeps them apart, with the source for each string, as of October 2026.
| Message on screen | Listed by | What it usually means | First fix |
|---|---|---|---|
| "Something went wrong." | OpenAI help center | The general error: OpenAI's servers or your own setup | Steps 2 to 4 above |
| "Something went wrong. If this issue persists please contact us through our help center at help.openai.com." | OpenAI forum thread, April 2023 | The same general error in older wording | Steps 2 to 4 above |
| "Hmm...something seems to have gone wrong." | OpenAI forum threads | The same family; forum reports tie it to one account rather than one browser | Sign out and in, then try a second account or support |
| "Something went wrong while generating the response." | OpenAI forum, autocomplete | The reply failed partway | Regenerate once, then a new chat |
| "There was an error generating a response." | OpenAI help center | A failed reply, often a one off | Regenerate, then VPN and network checks |
| "Error in message stream" | OpenAI forum thread, September 2026 | Forum posters describe a reply cut off partway; in one 2026 thread it hit every chat inside a single Project | Regenerate; if it repeats in one Project, try a chat outside it, then another network |
| "A network error occurred." | OpenAI help center | Your device cannot hold a connection to OpenAI | VPN, proxy and Web Protect off, then another network |
| "An error occurred while connecting to the websocket." | OpenAI help center | Same as the network error | Same as the network error |
| "There was a problem preparing your chat." | OpenAI help center | An extension or URL blocker interfered with the page | Extensions off, or a private window |
| "We detect suspicious activity." or "Unusual Activity Detected" | OpenAI help center | Traffic that looks automated, often a flagged VPN address or a shared account | VPN off; change the password if others use the account |
| "Something went wrong while generating your image. Sorry about that." | Autocomplete | Image generation failed; text chat is a separate component | Check Image Generation on the status page and keep writing |
| "Something went wrong while canceling your subscription" | Autocomplete | The billing flow failed, not the chat | Cancel where you subscribed: how to cancel ChatGPT Plus |
| "Download failed" or "File Not Found" | OpenAI help center | The generated file expired, or is over the 512 MB limit | Ask ChatGPT to make the file again |
Two rows deserve a note. The "Hmm" version is the one that follows an account around: in the OpenAI community thread opened in December 2023 and 107 posts long, users report it surviving private windows, VPNs, wifi and phones, while one poster's backup account worked. When steps 5 and 6 change nothing, that's the pattern. It points at support rather than your browser. Image failures are their own problem. The Image Generation component logged 18 incidents of its own in the same 90 days, so a broken picture says nothing about text chat.
Is ChatGPT down, or is it just me?
Read the incident title as well as the banner. In the 90 days to 2 October 2026, 18 of the 31 incidents on ChatGPT Conversations named a plan, a region, an app or a feature. So an open incident may not include you, and a friend's working ChatGPT doesn't clear OpenAI.
| Scope named in the incident title | Incidents | Example title, with date |
|---|---|---|
| One or more plans (Free, Go, Plus, Pro or "paid plans") | 10 | "Increased error rate for Plus and Pro users." 22 September 2026 |
| One region | 2 | "Elevated errors for ChatGPT users in Europe" 13 September 2026 |
| One app or device type | 2 | "Elevated conversation and login errors on ChatGPT for iOS and macOS" 12 July 2026 |
| One feature or mode | 4 | "Elevated errors in ChatGPT conversations with files" 5 August 2026 |
| No scope named | 13 | "Elevated errors affecting ChatGPT conversations" 25 July 2026 |
So use the status page in two passes. First, is anything open in the ChatGPT group, and does its title include your plan, region or app? Second, if it's all green, take the phone test: ChatGPT on your phone with wifi off. Working there and failing on the laptop means your browser or network, and steps 5 to 7 are for you. Failing on both, with a green page, points at your account or an incident not yet posted.
Two more numbers from the same page, read on 3 October 2026. OpenAI shows 99.65 percent uptime for the ChatGPT group over 90 days and 99.96 percent for its APIs. Of the 31 conversation incidents, 27 listed no API component at all. That gap is why apps built on OpenAI's API can keep answering while chatgpt.com errors. And the uptime number hides the clock: an incident on Conversations was open for 79 of the 2,160 hours in the window, even though the component itself shows 99.96 percent. Only 3 of the 31 were posted as a partial outage and none as a full outage. The other 28 carried the mildest label, degraded performance.
Why does ChatGPT keep saying something went wrong in one chat?
When new chats work and one doesn't, suspect the thread or its model first. OpenAI's help page says to "Start a new chat if the conversation is long or has many turns", and a new chat costs nothing to try.
Model scoped incidents are the other case. On 15 September 2026 the status page posted "Elevated errors from GPT-5.6 and GPT-5.6 Instant on paid plans" for 42 minutes, and on 20 August "Elevated Errors for Thinking mode in ChatGPT" ran 176 minutes. Both titles name a model or a mode, not ChatGPT as a whole. That's why step 3 says switch models before you clear anything.
The reverse pattern, every chat failing in the browser while the phone app works, is a browser or network problem. The forum thread on "Something went wrong while generating the response", opened in September 2024, has users with cleared cookies and no VPN whose accounts only worked in the iPhone app, and others where a second account was fine. Steps 5 to 7 split them.
One lookalike isn't an error at all. "You've reached the maximum length for this conversation" is a context limit, and a summary pasted into a new chat fixes it, which the ChatGPT conversation too long fix walks through. If nothing errors but the answers turned short and forgetful, ChatGPT probably switched you to a weaker model.
What if it is an outage and the work cannot wait?
Use another assistant for the next two hours, because nothing on your side shortens an OpenAI incident. What to use when ChatGPT is down lists the alternatives with their own status pages, and how to carry the context over in one paragraph.
Whizi, which publishes this page, is one of those options. GPT and Claude models sit in one picker on every paid plan, with Gemini joining from Pro at $29.99 a month, so an incident on one provider means picking another row. You can also import your ChatGPT conversations. Where it loses: Whizi has a general error of its own, "Something went wrong. Please retry." (every Whizi send error, quoted). The import brings conversations and not memory or custom GPTs. And the $15.99 Starter plan carries only ChatGPT Luna, Claude Haiku and Whizi AI. Gemini and the larger Claude models start on Pro.
- Copy the unsent text out of the box before any refresh, since saved chats survive and drafts do not.
- Regenerate once, then start a new chat or switch models: OpenAI's help page says to start a new chat when a conversation is long or has many turns.
- Open status.openai.com and read the incident title for your plan, region or app as well as the banner.
- If the page is green, test ChatGPT on your phone over mobile data to split a network problem from an account problem.
- Turn off extensions, VPNs, proxies and secure DNS before clearing cookies, in that order.
- If an incident is open, plan on the median 108 minutes and move urgent work to another assistant.
Frequently asked questions
Does clearing the cache fix ChatGPT's something went wrong error?
Sometimes, but it belongs near the end of the list, not the start. OpenAI's help page puts clearing cache and cookies third for "Something went wrong.", after refreshing or starting a new chat and checking the status page, and it does nothing during an OpenAI incident. Try a new chat, the status page and a private window first.
Can a VPN cause ChatGPT errors?
Yes, and it's the most named cause in OpenAI's own guide: 7 of the 9 entries on its troubleshooting page, as of October 2026, tell you to turn off VPNs or proxies. A VPN can also trigger "We detect suspicious activity." when its address is shared with automated traffic.
Does paying for ChatGPT Plus stop the errors?
No. Of the 10 conversation incidents in the 90 days to 2 October 2026 whose titles named a plan, 6 named Plus, Pro or "paid plans", and Free or Go were named in the other 4. Paid accounts see the same general error and the same outages.
Will I lose my conversation when ChatGPT says something went wrong?
No. Saved conversations stay on OpenAI's side and come back when the error clears. The unsent text in the message box is the one thing a refresh destroys. Copy it out first. A reply that failed mid generation may be missing, and Regenerate asks for it again.
