Word clouds are popular because they look like analysis. They are not analysis; they are a frequency count with typography applied. Understanding exactly what they show makes them useful and stops them being misleading.
The picture tells you which words appear often. It cannot tell you why, in what tone, or in what context — a word appearing constantly because customers are complaining about it looks identical to one appearing because they love it.
The stop list does most of the work
Without one, every word cloud in English reads THE, AND, OF, TO in the largest type. Those are the most common words in the language by an enormous margin and they carry no information about any particular text.
Filtering them is what makes the cloud about the subject rather than about English. The list is conventionally a few hundred function words — articles, prepositions, pronouns, auxiliary verbs.
It is worth knowing that stop lists are a judgement call. "Not" is a stop word in most lists and is occasionally the most important word in a set of survey responses. If a result looks wrong, the filter is the first thing to check.
Why the biggest word is not proportionally biggest
Word frequencies follow a power law — Zipf’s law, in its classic form: the most common word appears about twice as often as the second, three times as often as the third, and so on.
Scale font size linearly with that and the top word is enormous while everything below the fifth is unreadable. The distribution is simply too skewed.
| Rank | Typical frequency | Linear size | Square-root size |
|---|---|---|---|
| 1 | 100 | 100% | 100% |
| 2 | 50 | 50% | 71% |
| 5 | 20 | 20% | 45% |
| 10 | 10 | 10% | 32% |
Taking the square root compresses that range into something the eye can compare. It is a deliberate distortion, and it is the standard one because the alternative is unreadable.
How the layout works
Words are placed largest first, each starting at the centre and walking outward along an Archimedean spiral until it finds a position where its box overlaps nothing already placed.
That is why clouds look organic rather than gridded, and why the shape is roughly elliptical: the spiral fills from the middle out.
Some words will not fit. A good implementation reports how many it dropped rather than silently discarding them, because a cloud that has quietly lost a third of the vocabulary is misrepresenting the text.
When is a word cloud the wrong tool?
Whenever the answer matters. For anything where you need to know how often, in what proportion, or how it changed — a bar chart is better in every respect, because a reader can compare lengths accurately and cannot compare areas of text at all.
Word clouds are good at one thing: giving a rough impression of a body of text at a glance, in a context where an impression is what is wanted. A conference slide, a report cover, a poster.
They are actively bad for survey analysis, which is unfortunately where they are used most. Frequency without sentiment can invert the finding.
Questions people ask
Why is a word missing? Either it was filtered as too short or too common, or it did not fit and was reported as skipped.
Can I keep common words? Yes, and occasionally you should — a cloud of function words is how you spot an unusual writing style.
Does colour mean anything? Not in a standard word cloud. It is decoration, which is worth saying because readers assume it encodes something.
Why do some words turn sideways? To fill vertical space. Purely a packing decision.
The word cloud generator uses a stop list and a square-root scale and reports anything it could not fit, and the word counter and character frequency counter give you the numbers behind the picture.