Detect tower · floor

Word count: why two tools disagree

A word count looks like the least controversial number in writing. It is not. Two reputable tools can look at the same paragraph and return figures that differ by half, and both of them are right.

That matters because word counts are almost never idle curiosity. There is a limit attached: an assignment, a submission, a job application, a field on a form that refuses anything longer.

The words people search reflect two different needs. A word counter or a word count is somebody checking against a limit. A character counter, a letter counter or count characters is almost always a form with a hard field length — a bio, a listing, a subject line. And words to minutes is a different question altogether: not how long the text is, but how long it takes to say.

The lesson is not that counters are unreliable. It is that a count is the answer to a question — what counts as a word — and different tools answer it differently. Knowing which question your limit is asking is the whole of this floor.

What this tower will not do

It will not rank the counting websites. They are all doing arithmetic, most of them do it correctly, and the differences between them are the rule differences described above rather than quality.

It will not tell you a universal answer to what is excluded. Titles, references and footnotes are counted by some institutions and not others, and any page that states a rule as if it were universal is guessing on your behalf.

And it will keep pointing at the least technical answer on the floor: ask. One message to whoever set the limit removes every ambiguity above, and it is faster than any tool here.

A note on the test at the top, because it is the method this whole site runs on. We did not read that counters disagree and repeat it. We wrote a sentence containing every awkward case at once — a four-part hyphenation, a currency figure with separators, a decimal, an em dash, a compound adjective, a URL, a contraction and a trailing number — and counted it under six explicit rules. The sample is printed in full so that anybody can paste it into any counter and check whether our figures hold. That is the only kind of claim worth publishing about a subject this easy to test. What holds instead is simple: the counting test above is published in full so anybody can reproduce it.