When an OpenAI-style completion hangs and a Gemini-style reply also looks “stuck,” the instinct is to restart the laptop, flush DNS, and rewrite the prompt. That sequence mixes three failure domains: the model provider path, the product/workspace UI, and your local network. Separate them first.

I keep the first look in one place: getstatus.ai for website reachability plus connection/response trends across tools I actually ship against.

A four-step triage you can finish in under five minutes

  1. Pin the symptom clock. Note the minute the spinner or empty reply became abnormal—not “sometime after lunch.”
  2. Reproduce once. Private window or second device; if only one path fails, stop declaring a global outage.
  3. Read two independent status pages—not the same vendor badge twice.
  4. Name the feature. Chat vs embeddings, login vs generate, web UI vs API. Feature-level pain is not a dark homepage.

Google’s chapter on monitoring distributed systems still frames this well: watch symptoms, then causes. Cloudflare’s Radar docs on outages are a second habit for internet-scale corroboration when Slack screenshots disagree.

OpenAI provider path

Open OpenAI status and match your symptom start to the latest HTTP check and the 24h chart. Rising latency with successful TCP/TLS often means “shared blip / degraded,” not “rewrite the system prompt ten times.”

Gemini / workspace path

For reply stalls and generate pain in Gemini-shaped tools, use Gemini status the same way. A green marketing page with a bad generate path is still a feature incident—capture that distinction in the incident note before you open a “full outage” thread.

What not to do

  • Treat a green badge as “every API method works.”
  • Treat zero community reports as “nobody else is affected.”
  • Map 403 / 429 straight to outage (auth and rate limits are common).
  • Decide ETA from chat screenshots alone.

Two calm status reads beat twenty restarts. Ship the triage note; then decide whether to wait, switch model, or keep working offline.