How to read an AI status page

Almost every AI provider has an official status page. The labels on it are precise, and the difference between them is easy to miss when something is not working.

It is a list of parts, not one light

A status page does not show a single “up” or “down”. It lists components, and each has its own state. A provider’s page might list its chat website, its developer API, a coding tool and a developer console as separate lines. One can be struggling while the others work normally, so the first thing to read is which component is affected. Our ChatGPT and Claude pages show how the same idea applies to two real services.

What the component labels mean

Many providers use the same status-page software, which uses a small set of labels for each component:

Operational. Nothing is reported.
Degraded performance. It works but is slower or less reliable than usual.
Partial outage. Some users, regions or features are affected.
Major outage. The component is largely unavailable.
Under maintenance. The provider has planned the downtime.

How an incident moves along

Providers commonly describe an incident in stages, such as investigating, identified, monitoring and resolved. “Identified” means the cause is known and a fix is under way. “Monitoring” means a fix has gone in and the provider is watching to see whether it holds, so errors can briefly come back. Read the latest update text, not just the headline. It often says what is and is not affected, for example that a developer console is slow while the chat product is unaffected.

What a status page cannot tell you

It can lag. A page is updated by people, so the first minutes of a real fault may show green.
It describes the provider, not you. It cannot see your account, your plan, your usage limits, your region or your network.
It rarely measures slowness. A service can be technically up and still painfully slow.
Some providers publish nothing readable. For those, an independent check is the only signal available.

How we use them

We read each provider’s own page where a script can, and combine it with a test of our own. Our method guide explains which components decide “down” for each service, and each service page names any component we deliberately do not count.

If the page says everything is fine and you still cannot get in, read why a service can be up but broken for you.

Sources

Read on 8 October 2026. Providers change their documentation, so check it for anything important.