Uptime alerts that mean something: what to check, from where, and who gets told

What a pass actually means

Uptime tools differ, but Google Cloud's documentation shows a typical design. A check passes if the response code meets your rule and, if you ask for it, the page contains the text you named. By default any 2xx response counts as success and redirects are followed, with the last response judged. The check does not load images or run scripts, and it is not signed in. So a check that only asks "does the home page answer?" passes when the page loads and says "database error", because that page still returns 200.

Add a content check and a second page

A keyword or content check closes that gap: it alerts if the page does not contain text that should always be there. Better Stack documents the same idea as a keyword monitor that alerts when the keyword is missing. Choose text that appears only when the page works properly, such as your business name in the footer or the word "Contact" on the contact page, and not text that appears on an error page too. Add one key page that matters to the business, such as booking or contact, because home pages are often cached and look healthy while the rest of the site is not.

• One check on the home page status.
• One check on the home page for agreed text.
• One check on a key page for its own text.
• Certificate expiry, and domain expiry where the provider offers it.

Check from outside, and avoid false alarms

Monitoring that runs on the same hosting account as the site can fail together with it. Outside checks, ideally from more than one region, avoid that. Google's default alert needs at least two regions to report failures for at least a minute, so a single blip does not wake anyone. Other providers use retries or confirmation periods for the same reason. Whatever the tool, set a confirmation step, and expect results to take a few minutes to appear when a check is first created.

• Run checks from outside the host that serves the site.
• Require more than one failure before alerting.
• Expect a short delay before the first results appear.

Tell two people, by two routes, and prove it

An alert is only useful if someone reads it. Send each alert to two named people and, where the provider allows it, through two routes such as email and a phone notification. Then prove it: create a check that is meant to fail, for example one that looks for text that is not on the page, and see who is told and when, and whether a recovery message follows. Write down the first three things to check when an alert arrives, so the person who reads it at an awkward time knows where to start.

What monitoring cannot see, and how the paid outcome is accepted

Outside checks cannot see anything behind a log-in, a failing payment, or a problem that affects some visitors only. Certificate monitoring tells you when expiry is near; it does not renew it. The fixed-price job for this is a published test price of £95, not yet tested with buyers, payable after sign-off. It sets up at most five checks at a provider you own, alerts two named people, runs a deliberate test failure and recovery, sets expiry warnings at least 14 days ahead and watches the checks for 48 hours. It includes no on-call cover, response time or repair of whatever the checks find.