A synthetic check can be one request, such as “does the home page return 200 and the expected text?”, or a journey of several steps, such as logging in, searching and checking out. Because it runs the same way every time from known locations, results are comparable, and a failure is found even at 3 a.m. with no traffic.
Its limit is coverage: it only tests the paths you script, from the places it runs. Real user monitoring complements it by measuring what actual visitors experience on their devices and networks, but only when they visit.
Use synthetic checks for availability and for the journeys that make money, and to catch problems before users do, for example right after a deploy.