ChatGPT Slow or Stuck Generating? How to Find the Cause

Verified against OpenAI support documentation · July 30, 2026

If ChatGPT is slow or stuck generating, do not begin by deleting the conversation or changing every setting at once. Wait 30–60 seconds, save a copy of your unfinished prompt, click Stop generating, and retry once. If it stalls again, send the same small prompt in a brand-new chat. That single comparison tells you whether the problem is probably limited to one conversation or affects ChatGPT more broadly.

Next, check the OpenAI Status page, test an incognito window with extensions disabled, and compare another network or device. The objective is not to cycle through random fixes. It is to determine which layer—service, conversation, model, browser, network, app, or account—changes the result.

Diagnose the scope before clearing anything

Test result	What it suggests	Best next action
Only one conversation stalls	The thread is the useful boundary to investigate	Start a new chat with a concise summary and the next task
One model stalls, another works	The issue may be model- or feature-specific	Record the model and retry later; check Status for a matching incident
Incognito works	Cached site state or an extension is likely involved	Disable extensions one at a time or clear ChatGPT site data
A different network works	VPN, proxy, secure DNS, firewall, or network policy may be interfering	Test those controls individually or ask the network administrator
Web works but the app does not	The app installation or local app session is the useful boundary	Restart and update the app, then sign in again if needed
Every chat fails on multiple devices and networks	A service or account-level problem is more plausible	Check Status, preserve timestamps, and prepare a support report

These results are evidence, not absolute proof. For example, a green status page reports aggregate availability; OpenAI notes that an individual customer’s experience can still vary by plan, model, and feature. Use two or three clean comparisons before deciding where the fault lies.

1. Use the fastest safe recovery sequence

OpenAI’s current troubleshooting sequence for an indefinitely spinning response begins with the least disruptive actions:

Wait 30–60 seconds in case the response completes.
Copy any prompt text that would be difficult to recreate.
Click Stop generating.
Use Regenerate once.
If it stalls again, open a new chat and send a smaller version of the request.
Refresh the page or restart the desktop or mobile app.

One retry is useful; repeated retries are not a diagnosis. If the same request fails three times in the same thread, moving to a clean chat produces more information than pressing Regenerate again.

If the interface shows a usage-limit or credit message rather than an endless spinner, treat that as a capacity issue, not a speed problem. Our guide to ChatGPT credits, usage, and limits covers that separate path.

2. Compare the old conversation with a clean chat

OpenAI specifically recommends starting a new chat when the affected conversation is long or contains many turns. Use a controlled prompt that does not need files, web search, connectors, or a custom GPT. For example, ask for a two-sentence summary of text pasted directly into the message.

Interpret the comparison carefully:

The clean chat works: keep the original conversation for reference, but continue in a new thread. Paste a short summary of the relevant decisions instead of copying the entire history.
Both chats stall: move to the model, browser, network, and status checks below.
Plain text works but a file-based request stalls: isolate the attachment. Check format, size, and upload completion using our ChatGPT file-upload troubleshooting guide.
Standard ChatGPT works but a custom GPT or connected tool stalls: record that specific GPT, app, connector, or tool. The general service may be healthy while that workflow is not.

Do not delete the original conversation as a troubleshooting step. A clean chat can test the boundary without destroying useful context.

3. Test whether the problem follows the model or feature

If your plan exposes another model, select it and repeat one small prompt in a new chat. Avoid comparing a simple text request with an image, deep research, coding, or tool-heavy request; that changes too many variables at once.

If only one model or feature fails, note its displayed name and the exact time. Continue with an available alternative when the work is urgent, then check the OpenAI Status page for a matching model or feature incident. Do not conclude that the account is broken merely because one route is degraded.

If every available model fails in a clean text chat, the browser, network, app, service, or account becomes a more useful place to investigate.

4. Read the error message instead of treating every failure alike

Message or symptom	First interpretation	Focused response
Thinking, Generating, Working, or endless spinner	The request has not completed	Wait briefly, stop, regenerate once, then test a new chat
There was an error generating a response	The response attempt failed	Regenerate once; if repeated, restart and isolate browser/network
Network error or WebSocket connection error	The live connection is unreliable or blocked	Disable VPN/proxy/security filtering temporarily and test another network
There was a problem preparing your chat	A browser extension or URL blocker may be interfering	Use incognito and disable script-, ad-, or privacy-blocking extensions
Blank, frozen, or endlessly loading page	The page or local session is not loading correctly	Hard refresh, sign in again, then clear site data if needed

The exact wording matters when you contact Support. Capture a screenshot before refreshing if it contains a request identifier or a distinctive message, but remove private conversation content and personal data.

5. Check OpenAI Status—but understand what green means

Open status.openai.com and look for an active incident affecting ChatGPT, conversations, a specific model, files, image generation, Voice, or another feature involved in your request.

If an incident matches your symptoms, local troubleshooting is unlikely to repair the service. Save your prompt, stop repeated attempts, and monitor the incident until OpenAI reports recovery.

If the page says systems are operational, continue testing. OpenAI’s status page explains that its metrics are aggregated across tiers, models, and error types. A regional, account-specific, feature-specific, or newly emerging problem may not appear as a broad outage.

6. Run a clean browser test

Open an incognito or private window, sign in to the same ChatGPT account, and send one small prompt in a new chat. Most extensions are disabled in a private session, so this test separates the account and service from much of the normal browser environment.

If private browsing works:

Return to the normal browser window.
Disable ad blockers, privacy tools, script blockers, grammar extensions, and page-modifying extensions.
Re-enable them one at a time until the problem returns.
Allow ChatGPT in the responsible extension rather than permanently disabling every protection.

If private browsing also fails, try another browser before erasing data. A second browser is a cleaner test than clearing all cookies immediately.

7. Refresh the session and clear only the necessary site data

Use a hard refresh—Cmd + Shift + R on macOS or Ctrl + Shift + R on Windows and Linux. If the page remains frozen, sign out and sign back in.

OpenAI recommends clearing the site data or cookies for ChatGPT if the problem continues. Save unfinished text first. Clearing site data resets local ChatGPT browser state and signs you out, so confirm that you know the correct account and sign-in method before proceeding. It is more targeted than deleting all browser history and cookies.

If signing back in becomes the new problem, stop the performance troubleshooting and follow our guide to ChatGPT login errors and authentication methods.

8. Isolate VPN, proxy, secure DNS, and network policy

Network and WebSocket errors often point to a connection that is blocked or unstable. Temporarily test without a VPN, proxy, secure DNS filter, “Web Protect” feature, or other traffic inspection tool. This is a diagnostic test, not advice to leave security controls disabled permanently.

Then switch one network variable:

home Wi-Fi to mobile data;
mobile data to a trusted Wi-Fi connection;
corporate network to an administrator-approved alternative;
one device to another device on the same account.

If ChatGPT works on another network, restore your security tools and change them one at a time. On a managed company or school network, give the result to IT instead of attempting to bypass organizational controls.

9. Separate an app problem from an account problem

If the iOS, Android, Windows, or macOS app is slow, force-close and restart it, then install the latest available update. Test the same account on chatgpt.com with a simple new chat.

Web works and the app fails: the local app session or installation is the useful boundary.
App works and one browser fails: focus on that browser’s site data and extensions.
Both fail on the same network but work elsewhere: focus on the network path.
All routes fail for the same account: record the evidence for Support.

Do not create a second paid account simply to test performance. If your expected subscription or workspace is missing after signing in again, use our subscription-not-showing checklist before purchasing anything.

10. Prepare a support report that can be investigated

If the failure persists across a clean chat, browsers, devices, and networks while no matching incident is reported, contact OpenAI Support through the chat bubble at help.openai.com. Include:

the exact symptom or error message;
the date, time, and time zone;
the model and feature involved;
the conversation URL or ID when safe to share;
browser and version, operating system, device, and app version;
whether the network is home, mobile, school, or corporate;
the clean-chat, incognito, alternate-network, and alternate-device results;
a screenshot with private content removed;
browser console errors and a sanitized HAR file if requested.

A HAR file can contain request data and other sensitive information. Review and sanitize it before sharing. Never include your password, one-time code, session token, API key, or unrelated private conversation content. OpenAI notes that the ChatGPT desktop app does not currently provide browser developer tools for HAR capture, so reproduce the issue on the web when Support requests a HAR.

Frequently asked questions

How long should I wait before stopping a stuck response?

OpenAI recommends waiting 30–60 seconds. If it still hangs, stop generation, regenerate once, and then test a new chat.

Why does a new chat work when the old conversation is stuck?

The result shows that the conversation is a useful boundary: the service and account can complete a clean request, while something about the original thread or its workflow differs. Continue with a concise summary in the new chat rather than deleting the old one.

Does a green OpenAI Status page prove the problem is on my device?

No. Status metrics are aggregated, and individual availability can vary. A green page makes a broad known outage less likely, but it does not rule out a model-, feature-, region-, account-, or newly emerging issue.

Should I clear all browser data?

Not first. Test incognito and another browser, then clear ChatGPT site data if those comparisons point to the normal browser session. Save unfinished text and confirm your login method before signing out.

Can a long conversation make ChatGPT slow?

OpenAI recommends starting a new chat when a conversation is long or has many turns. If the same small prompt works in a clean thread, move the necessary context as a short summary.

Is a VPN always the cause?

No. Temporarily disabling it is a comparison test. If nothing changes, restore it and continue to the browser, device, and service checks.

Where can I find more ChatGPT troubleshooting guides?

Browse the HiseHub ChatGPT help center for account, subscription, upload, memory, credits, and plan guidance.

Official sources

HiseHub is independent and is not affiliated with OpenAI. Interface labels, models, troubleshooting instructions, and service behavior can change; the current OpenAI Help Center and Status page control.