Counting characters with string length counts UTF-16 code units. An emoji can be two units, and a letter with a combining accent can be more than one. The counters here also show code points and grapheme clusters.
Words use Intl.Segmenter with word granularity, and only segments the browser marks as word-like. That is not the same as splitting on spaces. A sentence count from the same API is labeled as an approximation, because abbreviations and quotations do not always match how a person would split the text.
Chinese, Japanese and Korean text is not separated by spaces. The segmenter still returns a word count, and the page marks it as approximate. If the browser has no segmenter, those scripts get character counts and no word count, rather than a guess from a space split.
The counting rules are written down for operators in the project notes. They are not a ranking method.