ChatGPT stuck on Thinking? Fix the spinner without losing the work.

When ChatGPT stays on Thinking..., Generating..., Working..., one last tweak, or returns thinking failed, the fix is not to keep hammering retry. Preserve the coding state first, then check whether the problem is the conversation, browser, network, or a live service incident.

Quick answer: wait 30 to 60 seconds, use Stop generating, then try Regenerate once. Before refreshing or leaving the conversation, capture the current goal, git status, git diff, changed files, exact error, failed attempt, and last verified check. If only that chat is stuck, restart it with a compact working packet. If every chat is stuck, test the browser, network, and service status.

Match the failure before retrying.

Only one chat is stuck

The conversation may be unusually long or overloaded. Preserve the working state, try one bounded regenerate, then continue in a new chat without pasting the entire stale transcript.

Every chat is stuck

Check OpenAI Status, then test a clean browser session, another network, or another device. A broad failure points away from one conversation.

The page loads but never finishes

Extensions, VPNs, proxies, secure DNS, cached site data, or an unstable connection can leave ChatGPT stuck on Thinking or Generating.

There is a live incident

Do not clear working state or stack repeated prompts. Preserve the task, monitor the official status page, and resume after recovery.

Search phrases that point to the same problem.

Search Console is showing the same failure in several words: chatgpt stuck on thinking, chatgpt stuck thinking, chatgpt stuck in thinking mode, chatgpt stuck on one last tweak, and chatgpt thinking failed. The labels differ, but the recovery path is the same: preserve the work first, then decide whether the stuck state is one chat, the browser, or the service.

Stuck on one last tweak

Wait briefly, then stop once. If the answer never lands, restart from a saved packet rather than asking the frozen turn to finish itself.

Thinking failed

Capture the prompt, goal, and last useful state. Retry once after checking status; repeated retries can bury the real working context.

Stuck generating

If the text stream stops mid-answer, save the partial answer, ask for only the missing section, or continue in a new chat with the verified state.

Just spinning

Test a clean browser profile, another network, and the status page before clearing data or abandoning the original task.

Safe recovery path.

Wait 30 to 60 seconds. Use Stop generating once, then Regenerate once. Avoid stacking duplicate requests.
Record the goal, changed files, current diff, failed attempt, decisions, and last verification result outside the chat.
Test a new chat. If the new chat works, continue there with the compact working packet.
If every chat is affected, check status.openai.com, hard refresh or reload the app, then sign out and back in.
Try an incognito window; disable extensions, VPNs, proxies, and secure DNS; then test another browser, network, or device.
If it still fails, capture timestamps, model, conversation URL or ID, browser console errors, and a sanitized HAR file for OpenAI Support.

When to retry, refresh, or start a new chat.

Use Regenerate when the page is responsive and only the last answer failed. Use a new chat when one long conversation keeps freezing but other chats work. Use a browser or network reset only when every chat is affected. Use the official status page when failures appear across devices or accounts.

Mobile and browser checks.

If ChatGPT is stuck on iPhone, Android, Chrome, Edge, or Safari, test one clean path before changing the task: reload the app, switch networks, disable extensions or content blockers, and try a private window. If the same prompt works elsewhere, the issue is likely local to the browser session rather than the coding task.

Before you hard refresh.

A hard refresh can fix the interface, but it cannot reconstruct uncommitted reasoning or an unfinished coding plan. Save disk-backed evidence first. Keep the handoff short enough for the replacement chat to reason from it instead of re-reading a transcript.

Goal:
Changed files:
Verified facts:
Exact failure:
Attempts already made:
Last passing check:
Next action:

What the next AI should receive.

Use a forward brief, not a transcript dump: the current goal, files that matter, verified changes, decisions and constraints, the exact failure, attempts that should not be repeated, test status, risk, and one next action. Any claim that is not backed by disk or command output should remain labeled as inferred.

How ShardStitch works with ChatGPT.

ShardStitch does not need the stuck ChatGPT turn to summarize itself. It prepares a clipboard-ready packet from the project state, labels disk-backed facts and unverified claims, and lets the work continue in ChatGPT or another supported target.

Source trail.

Use the documented ChatGPT controls and the reported interruption symptoms below to distinguish a stalled response from a service or session problem before retrying.

FAQ.

What should I preserve before restarting ChatGPT?

Preserve the goal, Git diff, changed files, exact error, failed attempts, decisions, verification status, and next smallest check.

Why is ChatGPT stuck on Thinking in only one chat?

If other chats work, the affected conversation may be unusually long or overloaded. Preserve the project state, try Stop generating and Regenerate once, then continue in a new chat with a compact working packet.

What does one last tweak mean when ChatGPT is stuck?

It usually means the response is near the end but the interface or model run has stalled. Preserve the work, wait briefly, stop once, then regenerate or continue in a new chat with a compact handoff.

How do I fix ChatGPT thinking failed?

Treat thinking failed like a recoverable interruption: save the task state, try one regenerate, check status and browser conditions, then restart with a short packet instead of a full transcript.

What if every ChatGPT chat is stuck?

Check OpenAI Status, hard refresh or reload the app, sign out and back in, try an incognito window, disable extensions and VPNs, and test another browser, network, or device.

What should I send OpenAI Support?

For a persistent failure, include timestamps with time zone, the model, conversation URL or ID, browser console errors, and a sanitized HAR file captured while reproducing the problem.

Scope note: ChatGPT is a web handoff target. ShardStitch prepares the packet locally for paste; it does not claim private access to ChatGPT servers or a blocked browser session.