What is Synthetic Monitoring?
Also called synthetic checks, active monitoring, probes.
Synthetic monitoring runs scripted transactions against a system on a schedule, from outside it, to verify that critical paths still work. Because the checks run continuously whether or not real users are active, they detect breakage during quiet periods and cover flows that real traffic exercises too rarely to reveal a problem promptly.
The complement to synthetic monitoring is passive observation of real traffic. Real traffic reflects genuine experience but only covers what users happen to do, and it goes silent overnight or in a low-traffic region, which is exactly when a deploy might break something. Synthetic checks provide a constant baseline and can exercise a signup, a checkout, or a login on a fixed cadence regardless of demand.
Checks range in depth. A shallow check confirms an endpoint answers with the expected status. A deeper check drives a full browser through a multi-step flow and asserts on rendered content. Depth buys coverage and costs maintenance, since scripted flows break when the interface changes for legitimate reasons. Excessive false alarms erode responder trust faster than any other monitoring defect.
Test data handling determines whether the practice is safe. Synthetic runs create accounts, submit forms, and sometimes trigger downstream effects, so they need clearly marked identities, exclusion from analytics and billing, and a cleanup path. Unmarked synthetic activity contaminates product metrics and can generate real notifications to real people, which is a self-inflicted incident.
For AI systems, a synthetic check can go beyond availability to assert on output quality. A probe that submits a fixed prompt and verifies that the response contains an expected element detects silent degradation that a status code cannot: a retrieval index that returned empty, a tool that stopped being offered, a prompt change that broke a required output format. Each probe run consumes metered capacity, so cadence is a cost decision.
Key points
- Scripted transactions run on a schedule from outside the system
- Covers quiet periods and rarely exercised critical flows
- Deeper browser flows cost maintenance and can produce false alarms
- Synthetic identities must be excluded from analytics and billing
- Quality assertions catch silent AI degradation that status codes miss
In practice
Every ten minutes a probe signs in as a dedicated monitoring account, opens a chat, sends a fixed question whose answer requires a retrieval lookup, and asserts that the reply cites a known document. One morning the check fails while every service reports healthy, because a reindex job left the vector store empty. The alert arrives about forty minutes before the first user report.