The counting rule, step by step
When you press Generate, the tool follows the same few steps every time. We ran this rule on the examples below in Node.js, so the numbers on this page are what it produces.
- The whole text is turned to lower case.
- It looks for runs made of letters (from any alphabet), digits and straight apostrophes. Everything else, including spaces, full stops, hyphens, underscores, @ signs and curly apostrophes, ends a run.
- Apostrophes at the very start or end of a run are trimmed.
- A run shorter than two characters is dropped.
- If the filler-word filter is on, a run found on its list of 180 English words is dropped. So are any words you removed with the chips.
- What is left is counted, sorted from most to least frequent, and cut to the Max words setting.
The “letters from any alphabet” part is a JavaScript feature called a Unicode character class escape (MDN, Unicode character class escape, read 21 October 2026).
Worked example: one awkward paragraph
Well-known writers don’t always agree. Some say it’s the Story; others say the story's THE point. In 2026, 3 editors read 3 stories—each story won. Café, cafe and CAFÉ.
The rule cuts this into 33 pieces. Twelve are dropped: the single characters t, s, 3 and 3, and the filler words it, some, the (three times), in, each and and. That leaves 21 counted words in 18 different forms:
| Word | Count |
|---|---|
| say | 2 |
| story | 2 |
| café | 2 |
Every other word appears once: 2026, well, known, writers, don, always, agree, others, story’s, point, editors, read, stories, won and cafe. The sections below explain each surprise.
Punctuation and joined words
Hyphens split words, so well-known became well and known. So do underscores, full stops inside numbers (3.14 becomes 3 and 14, and the 3 is then dropped for being a single character), and email addresses, which fall apart into pieces. The same thing happens when a PDF breaks a word at the end of a line: inter- and national are counted as two words. Join those back together before you paste.
Numbers are words as far as the tool is concerned. 2026 was counted; the single digit 3 was not.
Capital letters and accents
Capitals are ignored: Story, story and STORY pool into one entry. Accents are not ignored. Café was counted twice (once capitalised, once not) and the unaccented cafe was a separate entry with a count of one. The same goes for German Straße and STRASSE, which are different strings to the tool. If you want them pooled, make the spelling consistent first.
The curly apostrophe trap
Word processors turn typed apostrophes into curly ones. The tool recognises only the straight apostrophe as part of a word. So a straight don't is a single word on the filler list and is dropped, but a curly don’t splits into don and t. The t is discarded and don survives as a bogus word. That is where don in the table came from. The same applies to it’s, which breaks into it and s.
The fix is a find-and-replace. We changed the curly apostrophes in the paragraph to straight ones and ran it again: 31 pieces, 17 different words and 20 counted, and don had gone.
Words, not phrases
The tool counts single words only. New York arrives as new and york. If a phrase matters, join it into one word without a space, such as newyork. We tested that: it stays as one entry. It will look like a typo in the picture, so tell your audience, or keep the phrase out of the cloud and discuss it in the caption.
Languages
The filler-word list is English only, so switch the filter off for other languages and remove unwanted words by hand. Beyond that, results depend on the writing system. We tested a few strings:
- Accented Latin text such as El niño come mañana is split correctly into four words.
- Chinese has no spaces between words, so a whole run of characters, 我喜欢苹果, was counted as one “word”. The tool does not segment it.
- Hindi in Devanagari, नमस्ते, was chopped into fragments, because the vowel signs and joining marks are not letters as far as the rule is concerned.
- Turkish capital İ loses its dot when lowercased and splits İstanbul into i and stanbul.
If your text is in one of those, this tool is the wrong choice today; for most space-separated alphabets it works.
Where your text comes from
Text copied from a web page may carry menu items and cookie notices. Text copied from a PDF may carry hyphenated line ends and page numbers. Paste into a plain-text editor first, read it through and then paste into the tool. Anything repeated will be big, whether or not it matters.
Copyright reminder
GOV.UK says copyright protects original written work, including web content, automatically (gov.uk/copyright, read 21 October 2026). Use text you wrote or have permission to use if you plan to publish the result.
Related: the free generator page lists the size and export limits.
- Why is a word like don in my cloud?
- Your text probably uses curly apostrophes, as word processors produce. The tool treats only the straight apostrophe as part of a word, so don’t becomes don and t, and the single letter is dropped. Replace curly apostrophes with straight ones before you paste.
- Can I keep a phrase together?
- Not as a phrase: the tool counts single words. Joining the words with no space, such as newyork, keeps them together, but the cloud will show the joined form.
- Is capitalisation counted?
- No. All text is lowercased first, so Story, story and STORY are one word. Accents are not ignored, though: cafe and café are two words.
- Does it work for languages other than English?
- For languages that separate words with spaces and use letters the tool recognises, such as French, Spanish or Russian, yes, with the filler-word filter switched off. Chinese is not split into words, and scripts such as Devanagari are broken into fragments.
- Does word order matter?
- Only for ties. Words with the same count keep the order in which they first appeared in your text. Nothing else about order affects the cloud.