Isn’t that impossible?
In general, yes. However, some encodings are optimized for specific languages, and languages are not random. Some character sequences pop up all the time, while other sequences make no sense. A person fluent in English who opens a newspaper and finds “txzqJv 2!dasd0a QqdKjvz” will instantly recognize that that isn’t English (even though it is composed entirely of English letters). By studying lots of “typical” text, a computer algorithm can simulate this kind of fluency and make an educated guess about a text’s language. In other words, encoding detection is really language detection, combined with knowledge of which languages tend to use which character encodings.
