· AI Seedling · Beginner Guides · 11 min read
ChatGPT Not Working? A Beginner's Troubleshooting Checklist
Work through what ChatGPT is actually doing: check the status page first, then match the error message to the fix. Includes the steps to collect before contacting support.
Most ChatGPT problems fall into three groups: the service is down, something on your network is blocking it, or something in your browser is interfering. The order you check them in matters, because the first one is not your fault and the other two are.
Start with the status page. It takes ten seconds and rules out an entire category.
Step one: check whether OpenAI is having a problem
Open status.openai.com before you clear a single cache. The page lists each ChatGPT component separately — Conversations, Login, Search, File uploads, Voice mode, and others — with a current state and a thirty-day uptime figure.
As of October 27, 2026, the ChatGPT group of sixteen components showed 99.70% uptime over the preceding quarter, with Login at 99.93% and Conversations at 99.95%. Those numbers are worth reading as context rather than reassurance: 99.7% still leaves room for several hours of trouble across a quarter, and you may be reading this during one of them.
If a component shows a degraded state or an active incident, stop. Nothing on your computer will fix it. Wait, or use a different tool for the task.
Step two: match the message to the cause
OpenAI’s troubleshooting page documents the specific errors, and each one points at a different cause. Matching the message saves you from trying fixes at random.
| What you see | What it usually means | First thing to try |
|---|---|---|
| “Something went wrong.” | Generic: server issue or local setup | Refresh, then check the status page |
| “A network error occurred.” | Your device cannot reach the servers reliably | Turn off VPN or proxy |
| “An error occurred while connecting to the websocket.” | Same family as above | Turn off VPN or proxy |
| “We detect suspicious activity.” | Traffic looks automated or shared | Disable VPN, change your password |
| “There was a problem preparing your chat.” | An extension or URL blocker is interfering | Disable extensions, try incognito |
| “There was an error generating a response.” | A single failed generation | Click Regenerate |
| Stuck on “Thinking…” with a spinner | Long conversation or a stuck request | Stop, then Regenerate |
| “Download failed” / “File Not Found” | The file expired, or exceeds 512 MB | Regenerate the file |
| Blank screen, no error at all | Page failed to load | Hard refresh |
The table is the fast path. The sections below explain the ones that are not obvious.
“Something went wrong” and other generic errors
OpenAI lists a specific sequence for this message:
Refresh the page or start a new chat. Check status.openai.com for known outages. Clear your browser cache and cookies. Use an incognito or private browsing window. Disable browser extensions, especially privacy or security tools. Turn off VPNs or secure DNS tools like Web Protect. Switch to a different network if possible.
Source: OpenAI, Troubleshooting ChatGPT Error Messages, checked October 27, 2026. The page had been updated five days before that check.
Notice the order. Status page first, then cache, then extensions, then network. Each step narrows the cause. Doing them in a different order can waste twenty minutes.
If the same message appears in incognito mode with extensions off and a different network, the problem is not on your side.
Network errors: the VPN is usually the culprit
Two messages point at the same layer:
“A network error occurred.” and “An error occurred while connecting to the websocket.”
Both mean your device cannot hold a reliable connection to OpenAI’s servers. OpenAI’s suggested fixes are unusually specific about one thing:
Disable VPNs or proxy connections. Turn off Web Protect or other security filters. Try a different browser or incognito mode. Switch to a different network or device.
Source: same page, checked October 27, 2026.
If you are on a work laptop, the proxy may be enforced by your employer and not something you can switch off. In that case, try a personal device on a mobile hotspot to confirm the diagnosis before you spend time on browser settings.
“We detect suspicious activity” is not an accusation
This one alarms people, because it sounds like an account problem. It is usually a network problem.
Disable VPNs or proxies that might route your traffic through flagged IPs. Restart your browser or device. Use a private browsing window to start a fresh session.
Source: same page, checked October 27, 2026.
A shared VPN exit node can carry traffic from thousands of users. If some of them were running automation, the IP gets flagged and everyone behind it inherits the suspicion. Turning the VPN off is often the whole fix.
OpenAI also suggests two account checks worth taking seriously:
- Make sure you are not sharing your account with other people.
- Change your password and confirm your email is secure.
If you are not running bots or automation and the message persists, OpenAI asks you to report it rather than keep retrying.
Which extensions actually break it
“There was a problem preparing your chat” is usually an extension, but not any extension. The ones that cause it fall into three groups:
| Extension type | Why it breaks the page |
|---|---|
| Ad blockers with aggressive rules | Treat the response stream as an ad request and block it |
| Privacy and anti-tracking tools | Strip headers the page needs to open a connection |
| Script blockers and URL filters | Stop the JavaScript that renders the interface |
The test is simple: open an incognito window, which disables extensions by default, and try again. If it works there, an extension is the cause. Turn them back on one at a time rather than all at once, or you will learn that one of them is guilty without learning which.
Security software deserves a separate mention. OpenAI names Web Protect and similar DNS-level filters explicitly, because they sit outside the browser and incognito mode does not switch them off. If incognito works but a fresh browser profile does not, look at the security software on the machine rather than at the browser.
Stuck on “Thinking…” forever
This is the most common complaint and the one with the longest documented sequence. OpenAI’s order is deliberate — it starts with waiting and ends with collecting logs.
- Wait 30–60 seconds to see if the response completes.
- Click Stop generating, then Regenerate.
- Start a new chat and resend your prompt.
- Hard refresh:
Cmd/Ctrl + Shift + R, or reload the desktop app. - Sign out and sign back in.
- Open an incognito window with extensions disabled.
- Disable VPNs, proxies, or secure DNS.
- Try a different browser, network, or device.
- Check the status page again.
If the issue persists across browsers, devices, and networks, collect a HAR file and browser console errors with timestamps, note the model used and the conversation URL or ID, then contact Support.
Source: same page, checked October 27, 2026.
Two things about that sequence are worth knowing.
The long conversation is often the cause. A chat with hundreds of turns carries all of that context. Starting a new chat is not a workaround for a bug; it is the fix for a design limit.
Step nine is the real diagnosis. If the problem follows you across browsers, devices, and networks, it is not your setup. That is the point at which contacting support is the efficient move rather than a last resort.
What to collect before contacting support
Support requests that arrive without evidence take longer to resolve. OpenAI asks for specific items:
| Item | Why it matters |
|---|---|
| A HAR file | Records every network request the browser made during the failure |
| Browser console errors | Shows what the page itself reported |
| Timestamps | Lets support match your failure to a server event |
| The model you were using | Different models have different capacity |
| The conversation URL or ID | Points them at the exact session |
A HAR file sounds technical but takes about a minute to produce: open developer tools, go to the Network tab, reproduce the error, then export. Do it while the problem is happening — a HAR file recorded afterwards contains nothing useful.
“Download failed” has two ordinary causes
Ensure that the file is recently generated, files generated by ChatGPT expire quickly. Ensure the file downloaded is less than the file size limit of 512 MB.
Source: same page, checked October 27, 2026.
Both causes are easy to fix and neither is a bug. Ask for the file again in the same conversation, or ask for a smaller output split across several files. The 512 MB ceiling is generous for documents and easy to hit with datasets or archives.
Blank screen, no error message
A page that loads nothing is usually a caching or extension problem rather than a server one:
- Hard refresh with
Cmd/Ctrl + Shift + R - Sign out and back in
- Clear site data for the ChatGPT domain
- Disable content blockers
- Try incognito, a fresh browser profile, or another browser
- Turn off VPNs and secure DNS
If it still fails, OpenAI’s advice is the same as above: console logs plus a HAR file with timestamps.
If it only breaks on mobile
Phone problems are usually app or store problems rather than browser ones, so the fixes differ.
On iOS: force-quit the app, then reopen it. If it still fails, check the App Store for a pending update, and open the same account in a browser to see whether the problem follows you. If the web works and the app does not, the issue is local to the app.
On Android: the same sequence applies, and OpenAI keeps a separate reinstall guide for cases where tools freeze or return errors. Uninstalling and reinstalling is a reasonable step once a force-quit has failed, because it clears local state you cannot reach any other way.
“Phone number required” on desktop is its own case. The Windows and macOS apps sometimes ask for a mobile number before continuing. That is not an error — it needs a number from a supported country, and the on-screen steps handle the verification.
One rule applies to every device: if the problem is identical on your phone and your laptop, on two different networks, it is not the device.
A short diagnostic order that saves time
When something breaks and you do not want to think about it, run this in order:
- Status page. Ten seconds, rules out the biggest category.
- Hard refresh. Fixes stale page state more often than people expect.
- New chat. Fixes long-conversation failures.
- Incognito with extensions off. Rules out your browser.
- VPN off. Rules out your network path.
- Another device or network. Confirms it is not you.
- Collect HAR and contact support. Now you have evidence.
Most problems end at step 2 or 3. The ones that reach step 7 are the ones worth escalating, and arriving with a HAR file turns a vague complaint into a report someone can act on.
A walkthrough, start to finish
Here is how the checklist plays out in a case common enough to be worth walking through: ChatGPT stops responding in the middle of a work session.
Minute 1. The status page shows every component operational. Not a server issue.
Minute 2. Hard refresh. The page loads, but the conversation still hangs on “Thinking…”.
Minute 3. Start a new chat with the same prompt. It works. The problem was the length of the old conversation, not the service.
Minute 4. The original chat is still needed. Open it and ask a shorter follow-up. It responds. The conversation is intact; it simply could not carry another long turn.
Outcome: no support ticket, no cache clearing, four minutes spent. The expensive paths — clearing all cookies, reinstalling the browser, contacting support — were never needed.
That is the normal shape of a ChatGPT problem. It is usually one specific constraint rather than a broken service, and the checklist exists to find which constraint without guessing.
Questions beginners often ask
Is ChatGPT down right now?
Check status.openai.com. It shows each component separately, so “ChatGPT is down” is usually more specific than that — Search may be degraded while Conversations is fine.
Why does it work on my phone but not my laptop?
That points at the laptop’s browser, extensions, or network, not at your account. Compare the two devices on the same network before concluding anything.
Should I delete my account and start over?
Almost never. Account deletion is irreversible and does not fix browser or network problems. Work through the checklist first.
Does clearing cookies log me out of everything?
It logs you out of ChatGPT. If you use the same browser for other services, clear site data for the ChatGPT domain specifically rather than everything.
How long do outages usually last?
OpenAI posts updates on the status page as incidents progress. The page is more current than any article, including this one.
When to ask a person
Ask your IT department if the problem only happens on a work device — a proxy or security tool may be enforced and outside your control. Ask OpenAI Support when the checklist above is exhausted and you have a HAR file. Ask before changing account security settings if you share a workspace with other people, because some changes affect everyone.
If you are working through a task and the tool is down, the honest advice is to switch to something else rather than keep refreshing. A working method beats a broken tool, and you can finish the task later. The AI Level Test works without any account if you want to keep practising the judgement part while the service recovers.
