An essay has a 2,000 word limit. Word says 1,994. Google Docs says 2,011. The submission portal rejects it for being over. Nothing is malfunctioning; the three tools disagree about what a word is, and the portal's opinion is the only one that counts.
What everyone agrees on
Almost every word counter splits text on runs of whitespace and counts the pieces. That single rule produces most of the agreement between tools, and all of the interesting disagreements come from what happens at the edges.
Where they diverge
Hyphenated compounds. Is state-of-the-art one word or four? Whitespace splitting says one. Some academic style checkers say four. A four-word difference is nothing; a document full of hyphenated technical terms can differ by several percent.
Numbers and dates. Most tools count 2026 and 27/07/2026 as one word each. Some skip pure numbers entirely on the grounds that they are not words. A data-heavy report can differ noticeably.
What is inside the document. This is the big one for word processors. Word counts text in headers, footers, footnotes, endnotes and text boxes — or does not, depending on which checkbox is set and whether you have text selected. Google Docs excludes headers, footers and footnotes from its main count entirely. Paste the body text alone into a third tool and you get a third number, and that number is often the one that matches what a marker will read.
Em dashes and punctuation. word—word with no spaces is one token. The same phrase typed as word — word is three. Identical on the page, different in the count.
Characters are worse
Word counts differ by a percent or two. Character counts differ in ways that reject your submission.
Spaces. Every tool offers both “with spaces” and “without spaces”, and limits routinely fail to say which they mean. Assume with spaces unless told otherwise. That is what nearly every automated validator measures.
What counts as one character. This is where it stops being a convention and becomes a technical fact. A café emoji is one character to a human. In JavaScript, "👋".length is 2, because the emoji needs two units to store. A flag emoji can be 4. A family emoji with skin tones can be 11.
So a form with a 100 character limit implemented in JavaScript will reject a message that looks like 80 characters if it contains a dozen emoji. The user sees a bug. The developer sees the limit working exactly as written.
Twitter, now X, counts differently again. Its limit was never simply characters: Latin script counts as one unit per character, while Chinese, Japanese, Korean and some other scripts count as two, and URLs are replaced by a fixed length regardless of how long they actually are. A tweet with a 200-character link costs the same as one with a 20-character link.
Reading time is an estimate, not a measurement
The “5 min read” badge is words divided by an assumed speed. The number most tools use, and the one our counter uses, is 238 words per minute — the figure a large 2019 meta-analysis of reading studies converged on for adult silent reading of non-fiction.
Some sites use 200 to be generous, some 250. Medium famously used 265. The differences are large: 2,000 words is 7.5 minutes at 265 wpm and 10 minutes at 200. Neither is wrong, and neither reflects how long you take.
Speaking is much slower. A comfortable presentation pace is 130 words per minute, so a ten-minute talk is around 1,300 words — not the 2,380 that reading speed would suggest. Overrunning a conference slot is usually a word counter set to reading speed.
Which number should you trust?
The one belonging to whoever enforces the limit. That sounds glib; it is the entire practical answer.
For a university essay, the marker reads the body text, so count the body text — not the bibliography and footnotes your word processor is quietly including. For a submission portal, the portal's validator is definitive; paste into it and watch what it says before deciding you are under. For a meta description, the character count including spaces is what Google truncates on. For a tweet, only the composer knows.
Where the limit is tight, leave 2% of margin and stop arguing with the tools. The difference between 1,994 and 2,011 words is not worth a rejected submission.
A word about privacy
Worth stating, because it is a genuine reason to care which counter you use: most online word counters send what you paste to a server.
The text people paste into a word counter is a dissertation before submission, a cover letter, a clause from a contract, a press release under embargo. None of it needs to travel. Counting words is a string operation a browser completes in under a millisecond — the only reason to upload it is that the site was built as a web application rather than as a page.
The counter here updates as you type, which is only possible because nothing is being sent. A round trip per keystroke would be unusable. The privacy and the instant feedback are the same decision.
The short version
Word counts differ by a percent or two because tools disagree about hyphens, numbers and which parts of a document to include. Character counts differ far more, because emoji are not one character to a computer and spaces are optional to a limit. Reading time is an assumption about you dressed as a measurement.
Count what the person enforcing the limit will count, leave a small margin, and paste confidential drafts into something that does not upload them.