We really should use a HTML parser but I like the monitoring probes to be as light on dependencies as possible.