Word cloud questions

A word cloud question asks for a word or a short phrase: "One word for how this quarter felt." Every answer is encrypted in the respondent's browser like any other answer. The results panel shows the words people used, sized by how many people used them.

#What a respondent sees

A short answer box capped at 50 characters by default, with a privacy tip above it that asks them to keep the answer short and to avoid details that would identify them; their team name, their role, a specific project.

The cap is the point. There's no room in fifty characters for the account of a specific incident, which is exactly the kind of answer that becomes identifying no matter how well it's encrypted. If you want that account, ask for it with an open text question and accept the different risk that comes with it.

#Setting one up

Add a question in the poll wizard and pick Word cloud. Two fields matter:

• The prompt, up to 200 characters. Ask for one thing. "One word for how this quarter felt" returns usable words; "How do you feel about the quarter?" returns sentences that get cut off.
• Max answer length (characters), anywhere from 1 to 50, defaulting to 50. Lowering it to 20 or so is a reasonable way to force single words if that's what you want to chart.

#How the counts work

Each answer is split into words. Punctuation at the edges is trimmed and everything is lower-cased, so Tired. and tired are the same word.

Within one answer a repeated word counts once. Someone who writes tired, tired, tired contributes a single tired. The number beside a word is how many different people used it, not how many times it appeared; which is why a word's count can never exceed the number of people who answered.

#Reading the results

Nothing's readable until the poll's minimum group size has answered, and that's never below three. While the poll's open, the first results appear three answers after that, then new answers reach you in random groups of three to five, with a few always held in reserve until the poll closes. At close, everything shows. How batched release works.

Once results are unlocked, the panel shows three things:

• The cloud itself: the top 60 words, sized by count. Words two or more people used are highlighted.
• Every word, by how many people used it: listed underneath. The first ten show; Show all opens the rest.
• An Export menu: two downloads: Word counts (CSV) and Raw answers, shuffled (CSV).

The word counts file has one row per distinct word with its count and its share of respondents. The raw file has one row per answer, shuffled, so an answer's position in the file says nothing about when it was submitted.

#An on-device summary of the words

Once at least five people have answered, the panel offers a short AI draft describing the answers: not the words the chart already shows, but what they add up to. It's written by a small model running inside your own browser tab, on answers your browser has already decrypted, and nothing about it reaches our servers or any AI provider. The full account of what's downloaded and what never leaves the tab is in on-device AI summaries.

Two things are worth knowing before you use one on a word cloud specifically. A fifty-character answer is short enough that a distinctive one can be paraphrased closely, so read a draft as a pointer back to the answers rather than as a neutral aggregate. And the draft is a draft: it's labelled as one, it isn't stored, and it never appears on a public results link or in a shared snapshot.

#A word only one person used still appears to you

This is the honest difference between a word cloud and an open text question, and it's worth knowing before you write the prompt.

Open text only shows a recurring word once at least two different respondents have used it, because one person's vocabulary isn't a theme. Your own word cloud doesn't filter: after results unlock, the full list includes words with a count of one. That's deliberate; dropping the long tail would give you a chart that quietly disagrees with the answers behind it.

It does mean an unusual word is visible to you as an unusual word. Nothing attaches it to a person, and nothing pairs two words back to the same answer, but a distinctive turn of phrase in a team of nine is still a distinctive turn of phrase. Set the expectation in the prompt and keep word clouds out of questions where someone would want to explain themselves. Nothing's readable until the poll's Minimum group size has answered (three at the lowest); for a very small group, raise it or leave the word cloud out.

#What a public results link includes

If you publish results to a public results link, the word cloud on that page shows only words used by two or more people (and not very short or common words). A word one person used stays in your signed-in view. The public page has no downloads: neither the word counts nor the raw answers leave the workspace that way.

The same filter applies to a shared slice for a manager, and it's applied before the slice is written, so single-use words never end up in the file at all.

#Word cloud vs. open text

See also Viewing results. Each page covers the adjacent settings and workflow steps.

#Related

• Open text questions
• Batched release and the exit pool
• On-device AI summaries: what runs where
• Aggregate results and anti-inference controls