Why Word Count Estimates Vary Between Tools
Paste the same essay into a submission portal and it says 512 words. The word processor it was drafted in says 498. The cutoff is 500. Worrying that one of the tools is simply wrong, or that the piece is about to get bounced for being over the limit, is a completely reasonable reaction to that gap. A word count feels like it should be an exact, objective number. Run the same text through two different tools and it often isn't. Because "word" turns out to be a surprisingly negotiable definition.
Where the disagreement actually comes from
The simplest word-counting method splits text on whitespace and counts the pieces, which is what most tools do by default. But it immediately runs into judgment calls that different tools resolve differently. Is a hyphenated compound like "well-known" one word or two? Is a contraction like "don't" one word or does the apostrophe count as a boundary? Do standalone numbers, em-dashes, or bullet markers count as words? None of these has a single universally agreed answer, so two tools processing identical text through slightly different tokenization rules can legitimately land on different totals, both "correct" by their own definition.
A gap of a dozen or two words on a full essay isn't a sign either tool is broken. It's usually exactly the size you'd expect from a handful of hyphenated words, contractions, and stray numbers being tokenized differently, the kind of edge cases above. The tools aren't disagreeing about the essay. They're disagreeing about where one word ends and the next one starts.
Character counts have the same problem, from a different angle
Character counts look less ambiguous but carry their own version of the issue: with or without spaces, and how multi-byte or combined Unicode characters get counted, varies by tool. It mattered enough in practice that early Twitter counted double-byte characters. The CJK languages, essentially. As two characters against its 140-character limit, on the reasoning that a single Chinese or Japanese character typically conveys as much as a full Latin word; that rule was dropped in a 2017 update to a simpler, more uniform counting scheme after years of complaints that it effectively gave non-Latin-script users a shorter limit.
"Page count" was never a real unit to begin with
The common "250 words per page" conversion that word counters use to estimate page length doesn't come from any physical property of a page. It comes from standard manuscript format, the double-spaced, 12-point, one-inch-margin layout the publishing and screenwriting industries settled on as a submission convention. Change the font, spacing, or margins. As any real document does. And the actual number of words that fit on a physical page moves substantially, which is why page-count estimates are explicitly rougher than the character or word counts they're derived from.
Reading time is an estimate stacked on an estimate
Reading time estimates divide word count by an assumed reading speed, and that assumed speed is itself just a population average with wide individual variation. Commonly cited figures put average adult silent reading speed somewhere around 200-250 words per minute, but that number moves with text difficulty, familiarity with the subject, and the individual reader, easily by a factor of two in either direction. Two tools using different assumed reading speeds will produce different time estimates from the identical, correctly-counted word total. The discrepancy has nothing to do with counting accuracy at all by that point.
What to actually do about it
If the goal is clearing a hard cap for a submission, count with whatever tool will actually judge the submission, then leave a small buffer, five or ten words, rather than trimming right down to the limit; that buffer absorbs the kind of tokenization differences described above without a stricter counter putting the piece over on someone else's count. If the goal is estimating how many printed pages a manuscript will run, treat the 250-words-per-page conversion as a ballpark for standard manuscript format specifically, not your actual formatted document, and check the real page count in whatever format you're submitting instead. If the goal is just gauging how long a piece takes to read, treat the number as illustrative rather than exact: it's an average reading speed applied to your text, not a measurement of your specific reader.
Count words, characters, and get a reading-time estimate for your own text with the word counter.