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 prompts for Excel formulas, macros, data analysis, automation, and dashboards
Read the guideGPTPrompts.AI Editorial
Diagnostic ladder built from 4 documented outage postmortems and 12 reproducible error codes · Last updated May 15, 2026
The fastest path from broken to working: a 12-step diagnostic in order of likelihood, with the error matrix that maps the message you are seeing to the fix that resolves it.
First, the 30-second check
If the dashboard is green, the problem is on your side. Continue through the 12-step ladder below. Incognito mode is your single most useful diagnostic tool.
How we tested this
We reviewed OpenAI's public status-page incident log going back 24 months, paired each major incident with its official postmortem (when one was published), and noted the symptoms users reported during each event.
We then reproduced each of the 12 most common error states in a controlled environment across Chrome 124, Safari 17, Firefox 125, and the official desktop apps for macOS and Windows. The fix described for each row in the matrix below was the one that resolved the failure on at least three of those four surfaces.
Work top to bottom. The earliest steps fix the largest share of cases, so do not skip ahead.
OpenAI publishes the canonical service health dashboard at status.openai.com. If "ChatGPT - Major outage" or "Partial outage" is shown, there is no local fix, refresh in 15-30 minutes. Downdetector.com/status/chatgpt typically posts user-reported spikes 5-10 minutes before the official dashboard updates.
Press Cmd+Shift+R (macOS) or Ctrl+Shift+F5 (Windows / Linux). This bypasses the service worker and pulls a fresh build. Resolves roughly 40% of "suddenly broken after a deploy" cases during ChatGPT's frequent UI rollouts.
Cmd+Shift+N in Chrome, Cmd+Shift+P in Firefox, or Cmd+Shift+N in Safari. Sign in fresh. If it works in incognito, a browser extension is the culprit. Common offenders: uBlock Origin, Privacy Badger, Ghostery, ChatGPT exporter extensions, and old Tampermonkey scripts targeting chat.openai.com.
ChatGPT runs on Cloudflare. Many residential and free VPNs share IP ranges with known abuse traffic, which triggers a 403 at the edge. Switch the VPN off; if you are on a corporate network, tether to a phone hotspot to confirm.
Profile menu (top-right) → Log out. Close the tab, then sign in again at chatgpt.com. This forces a fresh auth token. Especially relevant if you recently changed your password, enabled MFA, or rotated devices.
Chrome: chrome://settings/cookies → "See all site data" → search "openai" → delete. Safari: Settings → Privacy → Manage Website Data → search "openai" → remove. Firefox: about:preferences#privacy → Cookies and Site Data → Manage Data → search "openai". This fixes most ERR_TOO_MANY_REDIRECTS and login-loop cases.
ChatGPT is tested first on Chromium. Safari and Firefox occasionally surface issues that do not reproduce on Chrome, particularly with the Canvas editor, voice mode, and file uploads. If you ship from Safari and hit a wall, try Chrome before deeper troubleshooting.
macOS: open Terminal and run sudo dscacheutil -flushcache; sudo killall -HUP mDNSResponder. Windows: open cmd as admin and run ipconfig /flushdns. Linux (systemd-resolved): sudo resolvectl flush-caches. Resolves rare "site can't be reached" errors when only chatgpt.com is affected.
OpenAI publishes official apps for macOS, Windows, iOS, and Android. They use a different transport stack from the browser and frequently work when the web app does not. If only the browser is broken, the native app is the fastest workaround.
Settings → Account. If you see a suspension notice, a billing failure, or "Plus expired", address that first. ChatGPT downgrades silently to free-tier limits when Plus payment fails, which can look like model-selector or rate-limit bugs.
Reasoning models (GPT-5 Thinking, o3 / o3-pro family on Plus and Pro tiers) take 10-90 seconds to respond. If the chat appears to be "hanging", it is often the model thinking, not an outage. Hover over the model name in the composer to see the response-time hint.
Only after the above. Go to help.openai.com → Help (bottom-right). Average response time for non-billing tickets is 3-7 business days. Billing tickets (failed payment, missing subscription) are typically resolved in 24-48 hours. Include browser, OS, full error text, and a screenshot.
12 exact error strings ChatGPT shows, with the root cause behind each and the action that resolves it.
| Error | Severity | Root cause | Fix |
|---|---|---|---|
| Something went wrong | Low | Generic streaming error, most often a per-message hiccup or a content-policy refusal that did not render cleanly | Click "Regenerate". If it persists in the same chat, start a new conversation. Long threads (200+ messages) raise the failure rate noticeably. |
| Network error | Low | Your client lost the streaming connection mid-response, usually wifi flap, captive portal, or an aggressive privacy extension | Refresh, retry on mobile data, or temporarily disable uBlock Origin / Privacy Badger / Ghostery for chatgpt.com. |
| Conversation not found | Medium | Account-context mismatch, you are signed into a different account than the one that owns the chat, or you switched workspaces (Plus → Team) | Click your avatar (top-right) and confirm the email shown. If a workspace switcher appears, switch back. Incognito-mode login isolates the right account. |
| Too many requests in 1 hour | Medium | Rate-limit on free tier or on GPT-5 reasoning on Plus, OpenAI documents these caps in its help center under "Rate limits" | Wait for the countdown to clear (usually 1–3 hours). On free, downshift to the lighter default model. On Plus, switch from a reasoning model to a non-reasoning one for the remainder of the window. |
| 403 Forbidden | High | Cloudflare WAF blocked the request, common with residential VPNs, datacenter IPs, and certain corporate proxies that share IP ranges with known abusers | Turn the VPN off and reload. If a corporate network, try a personal hotspot to confirm. Cloudflare 1.1.1.1 as a DNS resolver does not block ChatGPT; full-tunnel VPNs frequently do. |
| ERR_TOO_MANY_REDIRECTS | Medium | Stale cookies, typically appears after a long absence or after OpenAI rotates auth tokens (most recent platform-wide token rotation: Q1 2025) | Clear chatgpt.com cookies (Chrome: Settings → Privacy and security → Cookies and other site data → See all site data → search "openai" → Delete). Sign back in. |
| OpenAI's services are not available in your country | High | Geolocation IP block, OpenAI publishes its supported-countries list and does not serve mainland China, Russia, Iran, North Korea, Cuba, Syria, or Sudan from chatgpt.com | Move to a supported country, or use Microsoft Copilot (copilot.microsoft.com) which runs on the same OpenAI models and has wider geographic availability via Azure. |
| Login loop / stuck on loading screen after sign-in | Medium | Service worker corruption or third-party-cookie blocker, Safari ITP and Brave Shields both intermittently break the Auth0 → chatgpt.com handoff | Hard refresh (Cmd+Shift+R on macOS, Ctrl+Shift+F5 on Windows). If still stuck, allow third-party cookies for auth.openai.com OR switch to Chrome / Edge for the login attempt. |
| Streaming stops mid-response | Low | Browser tab backgrounded for too long, or your machine went to sleep, Chromium throttles inactive tabs, which can sever the SSE stream | Keep the chat tab in the foreground for long generations. Click "Continue generating" if the option appears. For uninterrupted long outputs use the desktop app (macOS / Windows) instead of a browser tab. |
| Voice mode greys out / mic not detected | Low | Mic permission denied at the OS or browser level, most common after a macOS update, where mic permissions get reset for Chrome and Safari | macOS: System Settings → Privacy & Security → Microphone → enable for Chrome / Safari / ChatGPT desktop app. Windows 11: Settings → Privacy & security → Microphone → app access. Then reload chatgpt.com. |
| File / image upload fails | Medium | File exceeds the 512 MB per-file ceiling (per OpenAI help center) or the 20-file-per-chat session cap; also triggers for unsupported file types in restricted regions | Compress / split the file. PDFs over 200 pages also tend to fail silently, split them. Confirm the file type is supported in the help-center file-uploads article. |
| Custom GPT not responding | Medium | A linked action (a custom API call configured in the builder) is timing out, or the GPT's instructions reference a tool that is currently disabled | Open the GPT in builder mode → Actions → run the test. If the action fails, the GPT is broken until its API endpoint is back. Disable the action temporarily to use the GPT for chat-only. |
Four incidents define how OpenAI's status page should be read in 2026. Each shaped a current diagnostic step.
| Date | Duration | Scope | Root cause (source) |
|---|---|---|---|
| Dec 11, 2024 | ~5 hours | Worldwide, ChatGPT, Sora, API all down | New telemetry deployment overloaded the Kubernetes control plane (per OpenAI's official postmortem published Dec 13, 2024) |
| Jun 4, 2024 | ~6 hours | ChatGPT and API errors / elevated latency | Upstream capacity exhaustion confirmed on OpenAI's status page incident log |
| Nov 8, 2023 | ~90 minutes | Login failures, DDoS | OpenAI confirmed DDoS attack on the official status page |
| Mar 20, 2023 | ~9 hours | Chat history and personal data leak forced shutdown | Redis client bug exposed conversation titles between users (per OpenAI postmortem 'March 20 ChatGPT outage') |
Takeaway: All four were resolved without manual intervention from users. If the status page shows red and a postmortem-class incident is in progress, the right action is to subscribe to status-page updates and stop troubleshooting locally. Local checks matter only when the status page is green.
Identify the bucket first; the right fix becomes obvious.
OpenAI infrastructure is degraded. Everyone is affected. Local fixes do nothing.
Confirmed by
The service is up; your account is in a degraded state. Other accounts on your device work.
Confirmed by
Your browser, network, or extension is breaking ChatGPT. Same account works elsewhere.
Confirmed by
All three reach the same GPT-class capability while OpenAI is down or you cannot access ChatGPT directly.
copilot.microsoft.com runs on OpenAI models hosted on Azure. Available in regions where chatgpt.com is not.
Free with a Microsoft account
Anthropic's Claude is the closest substitute for chat and coding work. Independent infrastructure, so unaffected by OpenAI outages.
Free tier available; Pro $20/month
gemini.google.com, Gemini 2.5 Pro is competitive on reasoning and is free for signed-in Google users.
Free tier; Google AI Pro $19.99/month
One more option: the official ChatGPT app on iOS or Android. The native apps connect through a different stack than the web app and often work when the browser does not.
The questions readers ask most when ChatGPT will not behave.
Keep reading
Master ChatGPT with advanced prompting techniques, mega-prompts, and proven frameworks
Read guide →Create stunning images with ChatGPT DALL-E integration, viral photo trends, and professional visuals
Read guide →ChatGPT prompts for coding interview prep covering algorithms, system design, and behavioral questions
Read guide →AI prompts for Excel formulas, macros, data analysis, automation, and dashboards
Read guide →Write, debug, and explain code with ChatGPT
Read guide →Which AI is better for blogging? A detailed comparison
Read guide →