Describe the symptom
Is the page failing to load, the first answer taking longer, or the response stopping midway? These symptoms need different evidence. Write down the local time, product, model selection and approximate task size. Avoid private prompts and API keys in shared screenshots. A clear symptom helps more than saying AI is slow.
Reduce the experiment
Try one short request in the same product. This does not prove service health, but it helps distinguish one complex task from a broader symptom. Compare the same workflow again later instead of switching several settings at once. Document whether files, tools or long conversation history were involved.
Read errors carefully
Authentication failures concern credentials or access, while quota responses may concern an individual account. Neither establishes a global outage. Distinguish an HTTP failure from an error that arrives after text starts. Retrying until a request succeeds should not erase the original failure from your notes.
Use monitoring as context
Inspect the timestamp, probe region and sample coverage before using a value. If the site says unavailable or stale, it cannot confirm the cause of your slowdown. Check the official status page and support guidance. Do not change account or network configuration based solely on one observation from a different location.