How to make a word cloud that actually tells you something
A word cloud is a picture of word frequency: the more often a word appears in a text, the bigger it is drawn. To make one worth looking at, start with clean text, remove filler words, decide how to treat different forms of the same word, and then read the result as a quick impression rather than as evidence. The word cloud generator does the counting and layout in your browser; this guide covers the choices that make the difference between a pretty picture and a useful one.
What a word cloud shows
A word cloud shows exactly one thing: how often each word occurs. Size encodes count, and nothing else does. Position and colour are decoration, so a word in the middle is not more important and two words side by side are not related. The format became familiar on websites in the 2000s as the tag cloud, where links were sized by how often a tag was used, and it later became a popular way to summarise speeches, survey answers and books.
My generator lowercases the text, counts every word of two or more characters, drops common function words, and takes the 80 most frequent. Font size scales in a straight line from the least to the most frequent word among those 80. Words are placed one by one along a spiral from the centre, each in the first spot where it does not overlap anything already placed; a word that cannot find room is left out, and the status line tells you how many were placed. The result is a 1200 by 700 pixel image you can download as a PNG.
Preparing your text
The cloud can only be as good as the text you feed it. A few minutes of cleaning pays off:
- Remove boilerplate: email signatures, page headers, question labels in survey exports and speaker names in transcripts will otherwise dominate.
- Watch out for one long answer: in survey data, a single very long response can outweigh dozens of short ones.
- Strip formatting debris: stray symbols, URLs and leftover markup from copying out of documents; the text cleaner handles most of it.
- Remove the obvious subject: in hotel reviews, hotel and room will be biggest and tell you nothing. Deleting them lets the more interesting words grow.
Stop words, word forms and phrases
Stop words are the grammatical glue of a language: the, and, of, to, you, was. Left in, they would fill every cloud, so the generator removes a built-in list of common English function words. That list includes not and no, which matters: "not good" will appear simply as good.
The generator does not merge word forms, so run, runs and running are counted separately, as are plurals and singulars. If you want them combined, standardise them first with find and replace. Phrases are split into single words too, so "customer service" becomes customer and service. Because the counter keeps hyphenated words together, joining a phrase with a hyphen (customer-service) before you build the cloud makes it count as one item.
Reading a cloud honestly
Clouds are good at one thing: showing at a glance what a body of text is mostly about, or which word you lean on too heavily in your own writing. They are poor at everything else. They lose context, so a word used in praise and in complaint looks the same. They lose negation. People also judge relative font sizes poorly, especially for long words, which look bigger than short words of the same size. Treat a cloud as a starting point for questions, and check anything you plan to report against the actual counts.
When a table beats a cloud
When the numbers matter, use a table. The word frequency counter gives exact counts and percentages for every word, which is what you need to compare two documents or quote a figure. The keyword density checker goes a step further and lists repeated two- and three-word phrases, which a single-word cloud cannot show. A simple bar chart of the top 15 words is often clearer than any cloud, a point How to choose a chart type โ and the two ways charts lie makes about chart choice in general. For web copy, Keyword density explained: how it is calculated and whether it matters covers how to use these counts without writing for a percentage.
Sources and further reading
The claims in this guide rest on these references, which were checked when the guide was last updated. Spotted an error? The contact page says how to report it.