Don't stop here
Hand-picked guides our readers explore right after this one.
Master ChatGPT with advanced prompting techniques, mega-prompts, and proven frameworks
Read the guideExpert guide to Claude prompts with XML tags, artifacts, and complex reasoning
Read the guideAI-powered code editor prompts for Composer, Chat, and inline editing
Read the guide'Something went wrong' is ChatGPT's catch-all error, and it shows up in two places that mean different things. 'Hmm...something went wrong' on load usually means a stale session or a browser/network problem. 'Something went wrong while generating the response' mid-chat usually means the streaming connection dropped, often on a very long conversation or after ~15 to 20 minutes of an idle open tab. In roughly 60% of cases the root cause is a temporary hiccup on OpenAI's side, so step one is always to check status.openai.com. If the status page is green, the fix is almost always local: hard refresh, clear cookies for chatgpt.com, disable a VPN/ad blocker, or start a fresh chat. This guide covers both variants and the order to try fixes in.
'Hmm...something went wrong' appears when the page loads
'Something went wrong while generating the response' cuts off a reply mid-stream
The error hits on long conversations but new chats work fine
A regenerate sometimes works, sometimes throws the same error
Works in incognito but not your normal browser
Error appears after leaving a tab open and idle for a while
The most common cause (~60% of reports). ChatGPT streams responses over a WebSocket connection that is more sensitive to interruptions than a normal page load, so a brief server or edge hiccup surfaces as this generic error. These clear within 30 minutes to a few hours.
A tab left open and idle for 15 to 20 minutes can let the auth session go stale, so the next action fails with 'something went wrong'. A refresh or re-login re-establishes it.
Corrupted cached data for chatgpt.com, or extensions like aggressive ad blockers, Grammarly, and VPN extensions, frequently break the interface after OpenAI ships an update.
The longer a thread gets, the more data streams on every turn, which makes the mid-response variant more likely. Starting a new chat resets it.
When to try: First, before touching anything local
Open status.openai.com (it now covers ChatGPT, the API, and Codex) and downdetector.com. If there is an active incident, no local fix will help, wait it out. Server-side issues typically resolve within 30 minutes to a few hours.
When to try: Immediately, if status is green
Press Ctrl+Shift+R (Windows/Linux) or Cmd+Shift+R (Mac) to force a refresh that bypasses cache. This resets the browser-to-OpenAI connection and clears the stale-session and post-update variants, which is a large share of cases.
When to try: When the error is tied to one long conversation
If the error only hits one long conversation, open a fresh chat. Long threads stream more data per turn and trip the mid-response variant. Copy any key context into the new chat, or ask ChatGPT to summarize the thread in 300 words first and paste that across.
When to try: If a hard refresh did not help
In your browser, clear cookies and cached data specifically for chatgpt.com (the app moved from chat.openai.com). On Chrome: Settings, Privacy, Cookies and site data, See all site data, search 'openai' and 'chatgpt', delete. Fully close and reopen the browser, then sign back in.
When to try: To isolate an extension conflict
Open a private/incognito window and log in. If ChatGPT works there but not in your normal browser, an extension is the culprit. Disable extensions one at a time (start with ad blockers, Grammarly, and VPN extensions) to find which one.
When to try: If you use a VPN or a restrictive network
Turn off any VPN or proxy and retry. OpenAI's own guidance repeatedly points to VPNs and proxies as a cause, since some exit nodes are rate-limited or blocked. Test on mobile data to rule out your network.
When to try: To confirm whether it is device-specific
If it fails on your laptop but works on your phone (or vice versa), the problem is device-specific. ChatGPT runs best on Chromium browsers (Chrome, Edge). Trying a different one isolates whether the issue is account-wide or local.
When to try: When every local fix fails
If nothing local works and the error persists across browsers and devices, it is almost certainly server-side. Monitor status.openai.com and try again in 30 to 60 minutes. There is no local fix for an outage.
Do not leave ChatGPT tabs open and idle for long stretches, refresh before a long session
Keep long projects in shorter chats and carry a summary across, rather than one endless thread
Keep your browser updated and your ad blocker allow-listed for chatgpt.com
Bookmark status.openai.com so your first check takes five seconds
Contact OpenAI via help.openai.com only if the error persists for hours across multiple browsers and devices while status.openai.com shows everything operational, and especially if it coincides with a billing or account notice. For a one-off 'something went wrong', self-help resolves it far faster than support, whose response times are measured in days.
The first appears on page load and usually points to a stale session or a browser/network problem, fix it with a refresh, cookie clear, or VPN off. The second cuts off a reply mid-stream and usually means the streaming connection dropped, often on a very long conversation, fix it by regenerating or starting a new chat.
No, but it is the single most common cause (~60% of reports). Always check status.openai.com first. If it is green, the problem is local: cache, cookies, an extension, a VPN, or a stale session.
A successful regenerate means the earlier failure was a transient streaming or server hiccup. If regenerating keeps failing on the same thread, the conversation has likely grown too long, start a new chat and carry a summary across.
That is the signature of a browser extension conflict or corrupted cached data. Incognito loads without extensions and with a clean cache. Clear cookies for chatgpt.com and disable extensions one by one, starting with ad blockers and Grammarly.
Yes, it makes the mid-response variant more likely. Each turn in a long thread streams more data, so the connection has more chances to drop. Splitting work into shorter chats with a carried-over summary avoids it.