When ChatGPT stops answering, you first need to know whether the problem is yours to fix or OpenAI’s. It’s worth settling that before you clear data or switch off your VPN. Some of those fixes sign you out or lower your privacy, and none of them helps during a real outage. OpenAI’s troubleshooting guide for ChatGPT errors says the failures can come from network configuration, browser extensions or temporary server-side problems.
That gives three places to look when you find ChatGPT not working: the service itself, your connection, or your browser session. Three quick checks that don’t change anything can point you to the right one: the status page, a private window and a second network. They won’t prove the cause, but they tell you which section below is worth your time.
The 3-Minute Fault Isolation Test (Decision Tree)
The “three minutes” is a target, not a promise. If you need to sign in again, turn on mobile data or ask someone for a second phone, it’ll take longer. Don’t let that push you into skipping steps. Each step below leaves your chats, settings and security tools exactly as they are. The splitting into service, connection and session is this site’s working model, put together from OpenAI’s help pages. OpenAI doesn’t name it.
Step 1: Check the status page for the part that’s failing. Open OpenAI Status and look at the ChatGPT group of components, not just the headline banner. If there’s an active incident that matches what you’re seeing, stop here. Wait, and watch the incident for updates. Changing things on your own device won’t fix a problem on OpenAI’s side. If the page is green, go to Step 2. A green page doesn’t clear OpenAI, for reasons explained in the next section.
Step 2: Open a private window and start a new, short chat. OpenAI’s error guide recommends using an incognito or private browsing window. Its guide to slow responses adds that incognito mode disables most extensions by default. Go straight to chatgpt.com, not a link from somewhere else. Sign in with the same method you used when you signed up, whether that’s Google, Microsoft, Apple or an email and password. Then send one plain sentence.
- If it works in the private window, the fault most likely sits in your normal browser: an extension, cookies or stored site data. Go to the “Something Went Wrong” section.
- If it fails there too, go to Step 3.
Step 3: Switch to a different network. If you’re on Wi-Fi, turn it off and use mobile data. If you’re already on mobile data, try a Wi-Fi network you trust. OpenAI’s login guidance gives switching from corporate Wi-Fi to mobile data as a test, and its error guide suggests a different network or device. One short test prompt uses very little data. Expect some slowness on mobile data, though: OpenAI notes that it may have lower bandwidth, so slower doesn’t mean broken.
- If it works on the other network, the problem is most likely in the path your first network takes: the router, a VPN, a web filter or a firewall. Go to the “Network Error” section.
- If it fails on both networks, and also on a second phone or laptop, the service, your account or one feature becomes the more likely cause. Go to the next section, and then to Support.
One OpenAI Community poster described this exact split. They kept getting “Something went wrong” and wrote, “I have verified that it only happens when I connect to Wi-Fi but with a data plan it works normally.” That’s one person’s case, not a pattern. Still, it shows how clearly this test can separate a network problem from everything else.
Step 4: Narrow it to one chat, one model or one feature. If a new chat works but an old, long one doesn’t, the problem is in that conversation. OpenAI’s error guide recommends starting a new chat when a conversation is long or has many turns. OpenAI also suggests switching to a different model in a brand-new chat to see whether the problem affects one model or your whole account. If ordinary text chats work but file uploads, voice or image generation don’t, check the status page for that specific feature.
How to Tell If ChatGPT Is Actually Down (Beyond the Status Page)
If you’re wondering “is ChatGPT down today”, the status page is the right place to start, but its headline only tells part of the story. At the bottom of OpenAI Status, OpenAI says its availability metrics are “reported at an aggregate level across all tiers, models and error types.” It adds that “individual customer availability may vary depending on their subscription tier as well as the specific model and API features in use.” In other words, a green banner means things look healthy overall. It doesn’t guarantee that your plan, your region or the feature you’re using is working.
OpenAI’s own incident records show how narrow a real outage can be. All of the incidents below are resolved and listed only as past examples:
- On 04/09/2026, OpenAI reported that users in the APAC region may face increased errors in ChatGPT, Work, image generation, file upload, Voice and Codex Cloud. That was one region, while other regions weren’t listed as affected.
- At the end of August 2026, an incident covered elevated errors in ChatGPT conversations for Free and Go plans only. It was dated 01/09/2026 on the page as displayed.
- On 23/09/2026, the affected group was the reverse: elevated error rates for ChatGPT across Plus and Pro plans.
- On 20/08/2026, the problem was one feature. OpenAI investigated errors when using Thinking mode and Image Generation.
So a friend on a different plan telling you “mine works” doesn’t prove the problem is on your side. When you open an incident, read which components it names, which plans or regions it mentions, and its current stage. The incident pages above moved through labels such as Investigating, Identified, Monitoring and Resolved. “Monitoring” means a fix has been applied, but things may still be recovering.
Sometimes the page is green and ChatGPT still isn’t working for you. On 09/05/2024, one OpenAI Community member wrote, “I can’t get a response out of it currently, but there’s nothing about it being down on the Status page.” Other people in that thread reported the same failure at around the same time. The thread shows the mismatch can happen, but it doesn’t explain why, and it doesn’t prove the status page runs late in general.
Crowd sites can back up what you’re seeing, but they don’t replace the official page. Downdetector keeps a Nigeria page for OpenAI built from problem reports that users submit. A spike there tells you other people are reporting trouble. It isn’t a confirmed incident, and a quiet page doesn’t clear your own setup either. Because of that, the private-window and second-network tests are still your best way to narrow things down when the status page gives you no answer.
Fixing the ChatGPT “Network Error” Loop (VPNs and WebSocket Drops)
A ChatGPT network error doesn’t always mean your internet is down. The page can load fine while a reply dies halfway. OpenAI’s error guide says messages like “A network error occurred” and “An error occurred while connecting to the websocket” mean your device is unable to establish a reliable connection to its servers. The key word there is reliable.
This is why an ordinary page load can succeed while a reply fails. According to OpenAI’s network recommendations, some ChatGPT features use a secure WebSocket (a connection that stays open while the reply streams in) in addition to ordinary web requests. For ChatGPT, that connection runs to wss://ws.chatgpt.com over TCP port 443. OpenAI warns that if WebSocket traffic is blocked, features “may stall, disconnect, or fail to stream updates correctly.” Web filters, proxies and TLS inspection can also block the connection or close it early, and so can idle timeouts, which cut a connection that looks inactive.
The table below matches when the error shows up to what it most likely points to. Treat each row as a lead to check, not a diagnosis.
| When the error shows up | What it most likely points to | What to try first |
|---|---|---|
| Right after you come back to a tab you’d left in the background on desktop Chrome | The tab was put to sleep while the reply was streaming | Add ChatGPT to Chrome’s “Always keep these sites active” list |
| Only while a VPN, Cloudflare WARP or a security app with web filtering is on | The tunnel or filter is interfering with the connection | Test once with it off, then turn it back on |
| On one Wi-Fi network, but not on mobile data | That network’s router, filter or firewall | Ask whoever manages the network, or test on another network you trust |
| On an office or school network | The organisation’s web filters, proxy or firewall | Ask IT to check WebSocket access to wss://ws.chatgpt.com |
| On every browser, device and network | A service, account or model issue | Check the status page, then contact Support |
In short, the error on its own tells you little. When it happens and what’s switched on at the time tell you much more.
When switching tabs kills the reply
To save memory, desktop Chrome deactivates tabs you aren’t using through its Memory Saver feature. When you go back to a deactivated tab, it reloads. Supercharge Browser, a company that sells a tab-management extension, says a mid-reply error after you switch tabs comes from Chrome discarding the tab and killing the WebSocket. That’s the company’s own explanation, not a finding from OpenAI. The fix doesn’t need anyone’s extension, though. Google’s instructions are: in Chrome, select More, then Settings, then Performance, and next to “Always keep these sites active”, select Add and enter chatgpt.com. Google also notes that you can’t change these performance settings on Chrome for Android or iOS, so this fix only applies on a computer.
Tab sleep isn’t the only possible cause, and there may be none you can find. In December 2023, one Plus subscriber on the OpenAI Community wrote, “I’m getting network error every other post,” and asked whether it was a known issue. The thread doesn’t give a cause. If the error keeps coming back even with the tab in front of you, go through the rest of the table.
VPNs, WARP and security apps
OpenAI’s first suggestions for this error include disabling VPNs or proxy connections and turning off Web Protect or other security filters. Do this as a short test with the tool off, then switch it back on. Don’t treat it as a permanent fix. Also, don’t do it on public Wi-Fi if you rely on the VPN for privacy there. Cloudflare’s WARP app, for example, has modes in which all device traffic is routed through WARP, so it becomes part of every connection ChatGPT makes.
One Cloudflare Community user reported that with WARP+ on, about one in three requests failed. They fixed their own case by excluding OpenAI domains from the tunnel. That’s one person’s result on one Mac. Cloudflare’s own documentation notes that traffic you exclude this way bypasses the client’s filtering, so check the current steps for your own device before changing any split-tunnel settings.
A VPN can cause a second kind of trouble. OpenAI links its “suspicious activity” warning to VPNs or proxies that route traffic through flagged IPs. Nigeria is on OpenAI’s list of countries where ChatGPT is supported, and that same page warns that using the service from outside listed countries “may result in your account being blocked or suspended.” If your VPN routes your traffic through an unsupported country, fix that, don’t work around it.
On an office, school or other managed network, don’t try to get around the organisation’s controls. OpenAI wrote its network guidance for IT teams, so the useful move is to send it to your IT person. It covers allowing WebSocket traffic on port 443, the standard “Upgrade: websocket” handshake, and TLS inspection and idle-timeout settings.
Why ChatGPT Says “Something Went Wrong” (And How to Clear It)
“Something went wrong” is OpenAI’s catch-all message. The error guide says it may indicate a temporary server issue or a problem with your local setup. On its own, the message doesn’t tell you which one. Your results from the isolation test do. Before you start, copy any prompt you’ve typed but not sent into a notes app, because reloading the page can lose unsent text.
Work through the fixes from least to most disruptive:
- Refresh the page or start a new chat. If the error only happened once, this is often enough, and OpenAI lists it first.
- Check OpenAI Status for the ChatGPT component you’re using.
- Try a private window if you didn’t in the isolation test. If it works there, your normal browser setup is the problem.
- Turn off extensions one at a time. OpenAI singles out privacy and security tools in particular.
- Clear ChatGPT’s cookies and site data. This will sign you out of ChatGPT in that browser. Make sure you know which sign-in method you used before you do it. OpenAI’s slow-response guide specifically says to clear the site data, not just the cache.
- Turn off your VPN or secure DNS tool for a test, or switch networks, as covered in the section above.
- Try a smaller request. Use a new chat with plain text and no attachment. If that works, the size of your first request or a particular tool may be involved, but that’s a clue, not proof.
If the error appears while you’re signing in, follow OpenAI’s login-specific steps. For generic sign-in errors, they say to wait at least 60 seconds before trying again. On Android, OpenAI says a sign-in “Something went wrong” error can happen when your Google Play Store version is out of date. To check, open the Play Store, tap your profile icon, then Settings > About > Play Store version.
Clearing site data won’t delete your chats
Clearing site data doesn’t touch your chat history. OpenAI’s chat-management page says regular and archived chats remain saved to your account until you delete them, or until a workspace retention policy removes them. Clearing cookies in your browser is a different action from deleting a chat in ChatGPT, so your conversations should be there when you sign back in.
What you really need to watch for is which account you sign back into. OpenAI warns that different sign-in methods can create separate accounts and that it doesn’t merge them automatically. If your sidebar looks empty after signing back in, check that you used the same Google, Apple, Microsoft or email login as before. And never use Delete all chats as a fix. OpenAI states that this action cannot be undone.
What to Do When ChatGPT Is Not Responding or Stuck Generating
A reply that’s slow but still moving is different from one that has stopped. If ChatGPT is not responding and the dots or spinner keep going with no new text, OpenAI gives a specific order of steps in its error guide. One thing comes first: if part of the answer is already on screen and you need it, copy it now. During one incident on 04/08/2026, OpenAI warned that some conversation turns may not be saved. So don’t assume that half-written text will still be there after a reload.
- Wait 30–60 seconds to see whether the reply finishes.
- Select “Stop generating”, then “Regenerate” to retry the same request.
- Start a new chat and send your prompt again.
- Hard refresh the page with Ctrl + Shift + R, or Cmd + Shift + R on a Mac. In the desktop app, reload the app instead.
- Sign out and back in, using the same sign-in method as before.
- Open a private window with extensions disabled, then try turning off your VPN, proxy or secure DNS tool.
- Try another browser, device or network, and check the status page again.
Pressing Stop is harmless, and sometimes it’s the whole fix. On 30/08/2026, one OpenAI Community member described a bug in a specific model. The long reply visibly finished while the page still acted as though it was generating. They also said that clicking Stop “immediately finalizes the turn.” In their case, the problem stayed in incognito mode and after clearing site data. That shows not every freeze can be fixed on your side.
If ChatGPT is not loading at all and you get a blank screen or endless loading with no error message, OpenAI’s steps are similar. Hard refresh, sign out and back in, clear site data for the site, disable content blockers, then try a fresh profile or another browser.
Taking it to OpenAI Support
If the problem happens on every browser, device and network you try, it’s time to contact Support. OpenAI asks for a HAR file (a recording of the browser’s network activity), any browser console errors with timestamps, the model you used, and the conversation link or ID. You reach Support through the chat bubble at help.openai.com, and the same page explains how to record the HAR file in your browser’s developer tools. Include your time zone with the timestamps. For Nigeria, that’s West Africa Time (WAT).
Be careful with that file. OpenAI warns that HAR files may include personal or other sensitive data, and says some browsers, such as Chrome, automatically clean the file when you save it. Review what’s in it before you send it, never post it publicly, and never share your password or one-time codes. OpenAI also notes that the ChatGPT desktop app has no developer tools for recording a HAR file, so reproduce the problem in a browser if you need one.
What to Check First
Start with the status page, and read the component your problem involves, not just the headline. Then try a private window and a second network. Together, those tell you whether to wait for OpenAI, fix your connection or clean up your browser. Leave clearing data and turning off your VPN until the tests point you there, and put your privacy tools back on once you’ve finished testing.
Frequently Asked Questions
Will clearing my browser data delete my ChatGPT chats?
No. OpenAI says chats remain saved to your account until you delete them. Clearing site data only signs you out of that browser. Sign back in with the same method you used before, because a different method can land you in a separate account.
Why does ChatGPT work on mobile data but fail on my Wi-Fi?
That points to something in your Wi-Fi’s path, such as the router, a web filter, a VPN or a firewall. OpenAI notes that proxies and filters can block or close the connection that ChatGPT uses to stream replies. Ask whoever manages that network to check it.
Should I leave my VPN off for ChatGPT?
Only turn it off briefly, as a test. If ChatGPT works with the VPN off, you’ve found a variable worth looking into, but put your protection back on afterwards. Also check that your VPN isn’t routing you through a country missing from OpenAI’s supported countries list.
Is it safe to send a HAR file to OpenAI Support?
It can be, if you’re careful. OpenAI warns that HAR files may include personal or other sensitive data, and notes that some browsers, such as Chrome, clean the file automatically when you save it. Send it only through the official Support chat, and never share passwords or one-time codes.
