The worst way to learn your site is down is from a customer. The second worst is from a Google Analytics graph three days later. Uptime monitoring is a robot that loads your site every few minutes from somewhere else on the internet and hollers when it can’t — twenty minutes to set up, no code, and it changes downtime from “how long was that?” to “four minutes, we know.” (We run our own robot across every site we maintain — that’s a whole story — but one site doesn’t need a fleet tool.)

Pick a monitoring service

For a first monitor, pick something free and boring. UptimeRobot’s free tier gives you 50 monitors checked every 5 minutes — plenty — though its terms limit the free plan to non-commercial use, so a business site should budget the few dollars a month for a paid seat there or at an alternative like Better Stack or StatusCake. Paid tiers also drop the check interval to one minute, which is the difference between a four-minute outage you catch and one you mostly miss. Sign up, verify your email, done.

Monitor a page that matters, not just the homepage

Everyone monitors the homepage, and the homepage is the most cached, most static, hardest-to-break page you own. It can serve happily from cache while your database is facedown in the snow. Add your homepage monitor, sure — then add a second one for a page that has to do real work: a product page, the checkout, a search results page, anything that touches the database on every load. Create it as an HTTP(S) monitor with the full URL, and treat that one as the site’s real pulse.

Add a keyword check so a broken page counts as down

Here’s the sneaky failure: a page that returns 200 OK and says “Database connection error” in the middle of it. To a basic HTTP check, that’s a healthy site. Use a keyword monitor instead — it fetches the page and checks that a specific phrase is present. Pick a string that only appears when the page truly rendered: your footer tagline, a product price, the “Add to cart” button text. Set the monitor to alert when the keyword does not exist. Now “up” means “actually working,” which is the only definition worth paying attention to.

Route alerts somewhere you’ll actually see them

An alert that lands in an inbox you check weekly is a diary, not an alarm. Email is the floor; add at least one channel that interrupts you — SMS, a push notification from the service’s mobile app, or a Slack channel your team actually watches. Two rules from running this at fleet scale: send alerts to two people if two people exist, and turn on the “back up” recovery notice too, so a 3 a.m. blip that fixed itself reads as one line of history instead of a mystery.

Test the alert path on purpose

Never trust an alarm you haven’t heard ring. Create one extra monitor pointed at a page that doesn’t exist:

https://yoursite.com/this-page-is-a-smoke-test

That returns a 404, which the monitor treats as down. Within one check interval — five minutes on a free plan — your phone should buzz. That buzz is the verification: the whole path from robot to pocket works. Delete the test monitor, and if the site ever goes down for real, you’ll already know exactly what to do next.