The fastest check is OpenAI’s own status page: open status.openai.com, read the headline at the top, and then find the ChatGPT group in the component list — if ChatGPT or the specific part you use (conversations, login, file uploads, voice) shows anything other than Operational, or an incident is open above the list, the problem is on OpenAI’s side and the only fix is to wait for its updates. If the page reports everything operational, ChatGPT is almost certainly not down, and the failure is on your side — your browser, network, VPN or app — which the checks further down this page will isolate. This is a guide to checking, not a verdict: any page that announces “ChatGPT is down right now” is describing the minute it was written, and ChatGPT’s state can flip several times in a day.

The 30-second route:

  1. Open status.openai.com.
  2. Read the overall headline, then the ChatGPT components underneath it.
  3. If an incident is open, click into it and read the newest update and its time.
  4. If everything is Operational, skip to the own-side checks below instead of retrying blindly.

What OpenAI’s status page shows

status.openai.com is OpenAI’s official record of its own service health — the page OpenAI itself updates when something breaks, and the only source that can actually confirm an outage. It covers more than ChatGPT: the same page tracks the OpenAI API and other OpenAI services as separate groups, each broken into components. For ChatGPT, the components distinguish the parts of the product — among them conversations, login, file uploads and image and voice features — and the API is listed separately with its own components, such as chat completions.

That separation answers the most common confusion during incidents: ChatGPT and the API are different surfaces and can fail independently. A ChatGPT outage does not automatically mean apps built on the API are down, and an API incident can break third-party tools that embed ChatGPT’s models while chatgpt.com works normally in your browser. So when something stops working, match the symptom to the surface first: the website or app you chat in, or a separate tool that uses OpenAI behind the scenes. The component list then tells you whether OpenAI agrees that surface is impaired.

The page also keeps the past visible. Under the current status sits an incident history and a 90-day uptime view per component, so you can see recent incidents, how long they ran, and how a component has performed over time. That history matters for two judgements: whether today’s symptom matches a pattern OpenAI has already acknowledged, and whether a component you depend on is genuinely unreliable or has simply had one bad hour that looms large in memory. If you rely on ChatGPT’s file features, note that uploads have their own component and their own limits in normal times too — see ChatGPT file upload limits.

How to read an incident

When OpenAI opens an incident, it appears above the component list with a title describing the symptom — typically phrased around what users would notice, such as elevated errors on a named feature — and a timeline of updates. Reading that timeline well is a small skill:

  • Investigating — OpenAI has confirmed symptoms and is looking for the cause. Early updates are short because the cause is not yet known; treat predictions made at this stage, anywhere, as guesses.
  • Identified — the cause has been found and a fix is being applied or prepared.
  • Monitoring — the fix is in place and OpenAI is watching the service to confirm it holds. The service usually works during monitoring, with occasional flaps possible.
  • Resolved — the incident is closed and the affected components return to Operational.

Components carry their own states during an incident, along the familiar scale: Operational, Degraded Performance (working, but slow or with elevated errors), Partial Outage (a part of the service unavailable), Major Outage (broadly down), and Under Maintenance for planned work. The page’s overall headline summarises the worst of it, which is why the headline alone can mislead: a headline about degraded performance may refer to one small component — a connector, an export feature — while conversations run perfectly. Always read which components the incident names before deciding it explains your problem. And read the newest update, with its timestamp: incidents evolve, and a screenshot of the first “investigating” post will circulate long after the incident has been resolved.

One more habit helps: note the incident’s start time in your own timezone. An incident that began before your problem started is a plausible cause; one that began after is not, however similar the title sounds.

Downdetector: the second signal

If the official page is green but ChatGPT is failing for you and, apparently, for half the internet, check Downdetector’s ChatGPT page next. Downdetector does not probe OpenAI’s servers; it aggregates problem reports from users and flags a possible incident when the volume is significantly above the normal level. In Downdetector’s own methodology page, it explains that it builds a baseline of typical report volume for each service and time of day from the previous six months, and only reports an incident when reports are significantly higher than that baseline, re-evaluating every few minutes.

Used that way, Downdetector has one real strength: speed. A sharp report spike often forms minutes before an official incident is posted, because thousands of users hit the failure before OpenAI has confirmed and written it up. Its weaknesses are the mirror image. Reports are unverified, so a spike can be manufactured by a regional internet outage, a viral post, or users confusing ChatGPT with another service. It cannot say which component is affected, or distinguish an outage from a popular feature launch overwhelming a single feature. And its absence proves nothing — an API-only failure may generate almost no consumer reports. The sound rule: official page green + a large rising Downdetector spike = probable real problem, not yet confirmed, recheck both in ten minutes; official page green + ordinary report levels = the problem is very likely yours, and the checks below are the way forward.

What ChatGPT looks like during a real outage

Outages have a recognisable shape, and it differs from a local fault in ways you can test. In a confirmed OpenAI-side incident affecting ChatGPT, typical symptoms include: conversations failing to load or new messages returning errors for many users at once; sign-in failing at the same step for everyone, rather than looping in your browser specifically; and the generic in-product errors — “Something went wrong” and “There was an error generating a response” — appearing on every prompt, in every chat, rather than on one long conversation. OpenAI’s own troubleshooting guide for ChatGPT error messages treats exactly these messages as potentially server-side and puts checking the status page among the first steps for them.

The discriminating tests are quick. An outage follows you: it fails the same way in the app and the browser, on Wi-Fi and on mobile data, and (if you can ask) for other people in other places. A local fault does not follow you — it stays with one browser, one device or one network. Two minutes spent switching surfaces tells you which kind of problem you have, and therefore whether to wait or to fix. If the failure is the response error specifically and everything else works, our dedicated guide to the ChatGPT “error generating a response” message works through that case step by step.

If the status page is green: the own-side checks

When OpenAI reports everything operational, work through these in order — they are ordered by how often each one is the actual cause, and they follow the same sequence as OpenAI’s own troubleshooting guidance:

  1. Reload, then start a new chat. A single stuck conversation — especially a very long one — can fail while the service is fine. A new chat is the clean test.
  2. Try an incognito or private window. This loads ChatGPT without your extensions and stored cookies. If it works there, an extension (privacy and security tools are the usual suspects) or stale site data in your normal profile is the cause.
  3. Turn off the VPN, proxy or security filter. OpenAI’s troubleshooting page names VPNs, proxies and Web Protect–style filters repeatedly as causes of connection and response errors. Disconnect fully and retry.
  4. Clear cache and cookies for the ChatGPT site if the problem is a login loop or a page that will not load properly — stored site data goes stale in ways a reload does not fix.
  5. Switch networks or devices. Phone on mobile data versus laptop on Wi-Fi separates a network fault from an account or service fault in one step. If it works on the second network, your first network (or its DNS or filtering) is the problem.
  6. Check the app separately from the browser. An outdated or stalled app can fail while the web version works, and vice versa. Update the app before drawing conclusions from it.

If ChatGPT fails identically across browsers, devices and networks while the status page stays green for hours, you have the rare case that deserves a support report rather than more retries: OpenAI’s help centre at help.openai.com is the route, and the exact error text, times, and what you already tried are what make the report useful. Note also what is not an outage: hitting a plan’s usage or message limits shows limit wording with reset information, and a single conversation that has grown too long failing to continue is a property of that conversation — starting a new chat and summarising what came before is the standard workaround, in the same way that a model inventing a fact — what an AI hallucination is — is a quality problem, not downtime.

What to do while an outage runs

Once the status page confirms an incident on the component you use, the useful actions are few:

  • Stop fixing your own setup. Confirmed provider incidents are not shortened by clearing caches or reinstalling apps, and doing so mid-incident only muddies your own troubleshooting afterwards.
  • Follow the incident on status.openai.com — that timeline is where resolution is posted, and anything you read elsewhere is a copy of it, older than the original.
  • Mind your mid-flight work. Your saved conversations live on OpenAI’s side and normally survive an outage untouched; what is lost is only the response that failed to generate. If you were drafting a long prompt in the message box, keep a copy elsewhere until service is stable.
  • Wait for Resolved before trusting it fully. A service in Monitoring can flap. Give important or long tasks a few minutes after resolution, or run a small test prompt first.
  • Switch tools only if the work cannot wait. Another assistant can cover an urgent task, but your ChatGPT history, memory and files do not transfer — plan for a fresh start in whichever tool you borrow.

Frequently asked questions

Where is ChatGPT’s official status page? status.openai.com. It shows the current state of ChatGPT and the OpenAI API component by component, any open incidents with a running timeline of updates, and a 90-day uptime history.

ChatGPT is down for me but the status page says everything is operational. What now? Most likely the fault is local: run the own-side checks — new chat, incognito window, VPN off, different network or device. A failure that stays with one browser or one network is not an outage, however total it feels from inside it.

Can ChatGPT be down while the OpenAI API still works, or the other way round? Yes. They are separate groups of components on the same status page, and incidents frequently affect one without the other. That is why a third-party app using the API can break while chatgpt.com works, and why the component list, not the headline, is the part to read.

What does “Degraded Performance” mean — should I wait or keep trying? The service is up but impaired: slower responses and some failed requests. Retrying is reasonable for light use; for anything long or important, waiting for the incident to resolve is the better use of the time.

ChatGPT is slow for everyone, but not down. Is that an outage? Not in the status page’s terms. Slow-but-answering is Degraded Performance when OpenAI confirms it, and ordinary load when it does not — either way the test is the same: if it is slow in the app and the browser, on two networks, it is not your setup. Third-party speed monitors can show a slowdown before OpenAI posts anything, but only the status page confirms it.

How long will an outage last? No honest general figure exists, and this article will not manufacture one. The incident’s own timeline — how far it has progressed from Investigating toward Resolved — is the best available estimate, and it is on the status page.

This article is a checking guide, checked in November 2026 against OpenAI’s official status page and Help Center linked above. It deliberately makes no claim about whether ChatGPT is up or down on the day you read it — take the live answer from status.openai.com, and use this page for how to read it.