ChatGPT Not Working? Check Access and Save Your Progress

If ChatGPT is not working, first check service status, then test whether you can sign in and send a short message in a new chat. Save any visible work before refreshing, signing out, or clearing browser data. Those checks separate a general interruption from a problem limited to your account, device, or conversation.

fenbs keeps a project checkpoint outside the affected chat, so you can resume useful work while you investigate. It cannot restore a failed ChatGPT session, but it can keep an interruption from becoming a lost afternoon of explanations.

What should you try first?

1. Copy an unsent prompt and any visible result you need into a safe working document. Note whether the last requested action actually finished.
2. Open OpenAI Status↗ (opens in a new tab) and look for an incident affecting ChatGPT or the feature you were using. Read the incident details rather than assuming every symptom has the same cause.
3. If you can use ChatGPT, open a new chat and send a harmless, short message. Avoid uploads and connected tools for this first test.
4. If the new chat works, retry a small version of the original request. If it fails too, compare a private browser window or another device.
5. Write down each test and result. Change one thing at a time so a successful retry tells you something useful.

For example, “new chat works; original conversation stays blank” is a much better starting point than “ChatGPT is broken.” Keep the original conversation available while testing elsewhere. Do not delete it as a troubleshooting shortcut.

Is the whole service affected or just your session?

Compare the failure across a few simple checks. A working status page does not prove your particular account is working: OpenAI reports availability across groups of users and features. Conversely, one failed request does not establish a service-wide outage.

These are clues, not a diagnosis. Record exactly what changes between tests, including whether you switched accounts or networks. A phone test with a different account cannot isolate the laptop browser as the cause.

Are you signing in to the right account?

Check the account identity while a working session is still open. OpenAI’s login troubleshooting guide↗ (opens in a new tab) says to use the account email or the same Google, Microsoft, or Apple provider account used for sign-in. A billing email can differ from the account email.

If the account opens but expected work is absent, verify the selected workspace before recreating anything. For a managed work account, ask your administrator to check your invitation and access. Creating another account can make the situation harder to explain because you now have another identity to distinguish.

If you see an explicit suspension or verification message, follow that message’s recovery route. Repeatedly refreshing cannot establish account ownership. Keep a record of the error without including passwords or verification codes.

Does a clean browser session work?

OpenAI’s error-message guide↗ (opens in a new tab) recommends testing a private window, another browser, and another network for loading and connection problems. It also identifies extensions and network filtering as possible causes. Use those tests to narrow the problem before clearing stored site data.

Use the same account and the same short test message in each browser. Check which extensions are allowed in private windows; do not assume every extension is disabled there. If a clean session works, return to the original browser and test optional extensions individually.

On an employer-managed device, give your IT team the comparison results before changing company network controls. If you clear site data, expect to sign in again. First copy anything unsent and make sure you know which account you need.

What if ordinary chat works but one feature fails?

Stop treating the whole app as unavailable once a plain chat succeeds. Write down the exact failing action: opening a project, using a saved instruction, starting research, or waiting for a scheduled run. Give that feature its own test instead of repeatedly resetting a working browser.

Use the focused guides for memory, projects, deep research, scheduled tasks, or custom instructions when that is the only affected feature.

For an outside service that will not connect or return information, start with ChatGPT connection troubleshooting. The useful distinction is whether the chat failed, the connection failed, or the requested action was refused.

How can you preserve the work you can still see?

Save a checkpoint in your own words if ChatGPT cannot produce one. Include the goal, the last verified result, what remains uncertain, and the next action. A partial response is not proof that a file was saved, a message was sent, or a change was tested.

• Goal: finish the revised proposal for internal review.
• Verified: the introduction was revised and copied into the working document.
• Uncertain: the comparison table stopped halfway through; its figures have not been checked.
• Next: check the source figures, then finish the table before requesting approval.

Keeping this checkpoint in fenbs makes the AI handoff available to the next connected assistant without relying on the interrupted conversation.

How to do this with fenbs

1. Create a free account. One person can start free, with no credit card.
2. Add the project and a task describing the work interrupted by the failure. Keep the troubleshooting symptom separate from the deliverable you were preparing.
3. Write Where we left off with what finished, what was checked, and the next step. Add useful source links as Links and listings rather than copying an entire conversation.
4. Connect a supported assistant through browser sign-in when it is available. Ask it to read the project context before continuing and save progress when it stops.

fenbs does not import ChatGPT chats or repair account access. If you only need to recover a login and have no ongoing project to hand off, OpenAI Support is the better place to focus.

What should you send to support?

Use OpenAI’s support instructions↗ (opens in a new tab) to report a persistent failure. Prepare the exact error, time and time zone, device, browser, and a short reproduction. Provide account details through the support channel, and remove sensitive material from screenshots or diagnostic files.

If you continue in a coding assistant, Claude Code troubleshooting covers that separate environment. Our AI at work hub has more workflows for keeping the work understandable when a tool changes.