Random keys use many different characters evenly, so their entropy is high; ordinary words score lower. Scanners use it to flag key-like strings, at the cost of false positives such as checksums.