Slow chats: diagnose before changing your workspace

First identify what is slow

A slow page, a delayed first response and a long-running tool task are different problems. Browser rendering may be affected by a large conversation. Response generation can be affected by request size, model/tool behavior, network conditions or provider demand. A single slow reply does not tell you which cause applies.

Do not repeatedly resubmit an action that could change a record or send a message. Check whether it completed first. Save unsaved work and the last known task state before refreshing or navigating away.

Try a safe diagnostic sequence

Start with low-risk checks that preserve your work. Use a harmless short prompt for comparison, not a real customer action. Compare results without assuming that a fresh conversation has all the original context.

• Check your connection and the provider's official status page.
• See whether the page itself is unresponsive or only the AI response is delayed.
• Try a small, non-sensitive request to see whether the issue is general.
• Break a large task into one document, one question or one output at a time.
• Use only the files necessary for that step; confirm they were actually processed.
• If an error persists, record the message and use the provider's current support guidance without attaching sensitive content.

Reduce the job, not the safeguards

Ask for an outline before a full report, one table before a multi-document analysis, or a short extraction before a polished draft. This makes the work easier to inspect even if it does not resolve a provider-side delay. Keep source and privacy constraints in the smaller request.

Avoid treating a faster answer as proof of correctness. A short reply can still omit the exception that matters.

Housekeeping is not an inference-speed fix

Deleting or archiving old chats is a housekeeping and retention decision—not a reliable way to speed up model inference. It may remove useful records, and it does not solve provider demand, a complex request or a poor connection.

If a new conversation is appropriate, use a reviewed handoff and verify the required references. For Ginny, follow the product's Work Session and recovery instructions. Do not reinstall the product, create a second workspace or delete history simply because a response is slow.