GPTMap

ChatGPT Not Working? Troubleshoot in This Order (2026)

When ChatGPT stops working, the cause sits in one of three layers: a platform incident, your local environment, or your account. Check status.openai.com first — most 'won't load' cases turn out to be local.

TL;DR
When ChatGPT stops working, isolate the layer first: check status.openai.com (status.json + components.json APIs, 34 components) for a platform incident. If green, bisect locally: incognito rules out extensions, then clear site data, another browser, another network (drop VPN/proxy); last, login state and quota. Most 'won't load' cases are local.
ChatGPT not working covers failures to load chatgpt.com or its apps, errors mid-conversation, and suddenly missing features; the first troubleshooting principle is layering — determine whether you're looking at an OpenAI platform incident, a local (browser/network) problem, or an account-state issue before acting.

"ChatGPT isn't working" shows up in many forms — blank pages, a spinning login, messages that won't send, features that vanished — but the root cause lives in one of three layers: a platform incident, your local environment, or your account state. The first principle is to isolate the layer before touching anything; otherwise you end up tweaking browser settings during an outage, or waiting on OpenAI while your own extension is the culprit. Here's a path where every step cuts the suspect list in half.

1. Always start at the official status page

Open status.openai.com. It's OpenAI's official status page, and it ships machine-readable APIs — handy for monitoring scripts or a quick check:

  • status.openai.com/api/v2/status.json — overall status, returning indicator and description (the healthy values observed 2026-09-24: none plus "All Systems Operational"; incidents are reported at their severity level)
  • status.openai.com/api/v2/components.json — per-component status for 34 components: Conversations, Login, Voice mode, Connectors/Apps, ChatGPT Atlas, ChatGPT Work, Codex Web, Codex in ChatGPT Desktop, CLI, VS Code extension, plus the API endpoints (Responses, Chat Completions, Embeddings, Realtime, and more)

Quick command-line check (or drop it into a monitoring script):

curl -s https://status.openai.com/api/v2/status.json
# healthy: {"status":{"indicator":"none","description":"All Systems Operational"}, ...}

curl -s https://status.openai.com/api/v2/components.json | grep -o '"name":"[^"]*","status":"[^"]*"'
# per-component view — check Conversations and Login first

The key read: an incident on the status page means local changes can't help — wait it out; a fully green page means skip to section 2 immediately instead of waiting.

2. Local environment: bisect with an incognito window

If the platform is fine but you're not, suspects narrow to your browser, network, or account. The highest-value single step is opening chatgpt.com in an incognito/private window — extensions are disabled and existing cache is ignored, testing two culprit classes at once:

Incognito resultConclusionNext step
Works fineExtension or cache/cookie problemClear site data for chatgpt.com; disable extensions one by one (ad blockers and privacy add-ons are the usual suspects)
Still brokenNetwork or account problemSwitch networks (drop VPN/proxy, try a phone hotspot); then another browser or device

Each step halves the suspect list — far faster than the "have you tried restarting" guessing game.

3. Network egress: VPNs, proxies, and work networks

ChatGPT is sensitive to where your traffic exits:

  • VPN/proxy egress: a rate-limited node or unsupported region shows up as load failures or errors. Dropping the proxy is the fastest test; if you need the VPN, switch nodes
  • Work/school networks: plenty of org policies block AI sites. The test is simple — if mobile data works, it's the company egress; file an IT ticket instead of fighting your browser
  • DNS or TLS interception: shows up as certificate errors or resolution failures; verify by switching DNS (or networks)

4. Account side: login state and quota

With the network cleared, what remains is the account:

  • Login fails or spins: check the Login component on the status page — sign-in and conversations are reported as separate components, so login can fail alone. If it's green but you can't get in, corrupted cookies/session is the usual cause: clear site data for chatgpt.com and sign in again
  • Works but features shrank / 'try later' prompts: the free tier carries quota and peak-hour limits (exact quotas as shown in the product), and those never appear on the status page — green status plus an in-product notice usually means a quota, so wait for the window to reset
  • Workspace/org accounts: Workspace members can be limited by admin policy — it looks like 'features missing'; confirm with your admin

5. Cross-check the clients

Web, desktop app, and mobile app are three ends: only one failing = a problem on that end (app update, local cache), not your account or the platform. If the web works but the app won't open, update or reinstall the app; if it's the reverse, go back to section 2 for browser checks. Only when all three fail on a green status page does account-side risk control enter the picture.

6. When to wait vs when to report

  • Incident posted → wait; the status page offers update subscriptions
  • Full layered sweep done, still broken for hours → record the symptoms (error text, timestamp, platform) and go through official support; the entry point is as shown in the chatgpt.com interface

7. Next steps

Key points

  • Check status.openai.com first: component-level status is machine-readable (status.json summary + components.json with 34 components)
  • Green status page + still broken = local problem — skip ahead to browser/network checks instead of waiting
  • An incognito window is the highest-value single step: works in incognito = extension or cache; still broken = network or account
  • VPNs, proxies, and corporate egress are frequent culprits — switching networks is the fastest bisection
  • Cross-check web vs desktop/mobile app: only one side failing means the problem is on that end, not your account or the platform

Frequently asked questions

Check status.openai.com. It also exposes machine-readable APIs: /api/v2/status.json returns the overall indicator and description (when healthy: none plus All Systems Operational), and /api/v2/components.json returns per-component status for 34 entries (Conversations, Login, Voice mode, Codex Web, the API endpoints, and more). If an incident is posted, waiting is the only fix — local fiddling can't solve a platform outage.

Official references

Related articles

Subscribe to GPTMap Weekly

One email every Monday: curated OpenAI updates, deep dives, and best practices. No ads, unsubscribe anytime.

Submitting opens Buttondown in a new tab to confirm your subscription.

GPTMap EditorialPublished 2026-09-24 5 min read
Test environment (EEAT)
Last tested: 2026-09-24
Model used: gpt-5.6