A card sort shows how people group your content. You give them a set of cards, each one a topic, page or feature, and ask them to sort the cards into groups that make sense to them. The results tell you which things belong together in your navigation, and often what to call each section.
It's the natural partner to a tree test. Card sorting helps you design a structure. Tree testing checks whether people can find things in it.
Open, closed or hybrid?
| Type | What people do | Use it when |
|---|---|---|
| Open | Make their own groups and name them | You're starting from scratch, or want to learn how people think about your content |
| Closed | Sort cards into categories you provide | You have a planned structure and want to check where content belongs |
| Hybrid | Use your categories, and add their own when nothing fits | You have a draft structure but suspect gaps |
If you're unsure, start with an open sort. It tells you the most, and you can follow up with a closed sort or a tree test once you have a draft structure.
Choosing your cards
- Use 30 to 60 cards for a typical study. Fewer than 20 rarely shows much structure. More than 60 tires people out and they start sorting carelessly.
- Write short, plain labels. "Returns and refunds", not "Our hassle-free returns policy".
- Avoid giving the answer away. If three cards start with "Account", people will group them by the word, not the meaning. Reword so each card stands on its own.
- Pick a consistent level. Mix pages with whole sections and people won't know how to sort them.
- Leave out things everyone will find easily, such as the homepage or a search box. They add effort without adding insight.
How many people?
For an open sort, around 15 to 20 participants usually gives a stable picture. In a well-known analysis, Tullis and Wood found results from 15 people matched those from much larger groups with a correlation of 0.90, rising to 0.95 at 30 people.
Closed sorts measure how many people put each card in each category, which is a percentage, so they need more people, like a tree test. Aim for 30 or more. If you'll compare user groups, you need enough people in each one.
Reading the results
Agreement matrix
For every pair of cards, the matrix shows the share of people who put both in the same group. ClickTree orders the cards so related ones sit next to each other, which makes strong groups show up as dark blocks.
- Pairs above 70% are a clear signal: keep them together.
- Pairs around 40 to 60% are where people disagree. Look at these closely; a card might belong in two places, or its label may be unclear.
Dendrogram and suggested groups
The dendrogram shows the same information as a tree. Cards that join early were grouped together by most people. ClickTree cuts the tree at 50% agreement to suggest groups: sets of cards at least half of people put together.
Treat the suggested groups as a starting point. They show how people see your content, not necessarily the best navigation. Business needs, page volume and future plans still matter.
Group names
In open and hybrid sorts, the names people choose are a gift. A name chosen independently by many people is a strong candidate for a menu label. ClickTree counts near-identical names together, such as "Help & support" and "help and support".
Category placement
In closed and hybrid sorts, look at where each card ended up. A card split between two categories, with no category chosen by more than half of people, needs attention: a clearer label, a different home, or a link from both places.
After the sort
- Sketch a structure from the suggested groups and the names people used.
- Write it out as a tree and run a tree test to check people can find things in it.
- Fix what fails, and test again.
Card sorting and tree testing together take you from a pile of content to a navigation you know works, usually in a week or two.