Why counts disagree
A word count depends on the definition of a word boundary. Hyphenated compounds, contractions, numbers and URLs are treated differently by different tools.
"State-of-the-art" is one word to some counters and four to others. A URL may be one token or several. Over a long document these differences add up to a few per cent.
Characters with and without spaces
Publishing and translation work usually specifies characters including spaces. Some platform limits count characters excluding spaces. The difference is roughly one sixth of the total for English prose.
Social platforms count differently again: some count certain characters as two, and shortened links as a fixed length regardless of the actual URL.
Practical rules
When a limit matters, find out which count is meant before writing, not after. Rewriting to fit the wrong metric is wasted work.
Leave a margin of about five per cent under any hard limit to absorb counting differences between your tool and the destination.
For estimating reading time, 200 to 250 words per minute is a reasonable range for general prose; technical material runs slower.
Other counts worth having
Sentence and paragraph counts reveal rhythm problems. A document averaging 40 words per sentence will read heavily regardless of vocabulary.
Unique word count against total word count gives a rough sense of repetition, which is useful when editing your own drafts.