Use exponential backoff with jitter: 3 retries (≈2s, 6s, 18s), cap at ~30s and stop after 3–5 attempts. In Zapier, wrap the ChatGPT API call in Paths: on timeout increment an attempt counter (Storage or Google Sheet), Delay by Zapier, then retry the Path. On final failure send a Slack message containing the prompt, error text, timestamp, attempt count and a link to the Zap run for manual review. Also log failures to a sheet/DB to avoid duplicate retries.

Compare Zapier and Make