Card Sorting and Tree Testing: How to Validate Site Navigation
Practice4 minInnovialabs editorial team
In short
Card sorting shows how users group your content on their own, while tree testing checks whether they can find things in the resulting structure; sort first, then test the tree, then fix and retest.
The two methods in brief
Card sorting: participants arrange cards with page or feature names into groups. It answers the question: how do users think about your content?
Tree testing: participants see only a text-based navigation tree, with no visual design, and get tasks such as "Where would you look for the return policy?". It answers the question: can people find things in the proposed structure?
The right order: sorting helps you build the structure, tree testing helps you validate it.
Three types of card sorting
| Type | How it works | When to use |
|---|---|---|
| Open | Participants create and name their own groups | A new site or full redesign, when you need the mental model |
| Closed | Categories are predefined, participants place cards into them | Validating existing sections or fitting new content into a structure |
| Hybrid | Categories are given, but participants can add their own | You have a draft structure but are not sure it is complete |
Preparing a card sort
- List the cards. Usually real pages or features. Too many cards tire participants, so keep the most important ones at a consistent level of detail.
- Write cards in plain language. Avoid internal jargon and repeated words: if three cards contain "payment", people group them by the word, not by meaning.
- Add descriptions where needed. A short note helps when a name alone is ambiguous.
- Pick a format. Remote and unmoderated through an online tool is fast and scalable. Moderated, in person or by video, means fewer participants but you hear their reasoning.
- Run a pilot with one or two colleagues to catch confusing cards.
Recruiting and sample size
- Participants should come from your target audience, not your own staff. If the site serves several segments, such as buyers and partners, recruit each.
- For open card sorting, practitioners commonly aim for roughly 15-20 participants per segment, after which new patterns rarely appear.
- Tree testing is quantitative, so it needs more participants, usually several dozen, for success rates to be stable.
- For multilingual sites, test in the language people will read the navigation in.
Tools
Online studies typically use dedicated services: OptimalSort and Treejack from Optimal Workshop, UXtweak, Maze and similar. For moderated sorts, any online whiteboard like Miro works, or physical cards on a table.
Reading card sort results
A similarity matrix shows what percentage of participants put two cards in the same group. Dark clusters along the diagonal are groups users agree on.
- Strong pair: the cards almost always go together. Keep them in one section.
- Card without a clear partner: its place is not obvious. It may need links from several sections or a new name.
- A dendrogram shows how cards merge into clusters at different agreement levels and helps you choose the number of sections.
In open sorts, also look at group names: the words participants use are good candidates for menu labels.
Running a tree test
- Build the navigation tree as a text hierarchy.
- Write tasks based on real user goals. Do not reuse words from the labels in the task wording, or the test becomes word matching.
- Mark the correct answer, or several, for each task.
- Analyze the metrics:
- Success rate: the share of participants who found the correct item.
- Directness: the share who got there without backtracking.
- Time to complete the task.
- First click: where people went first. A wrong first click often points to a problem with a top-level label.
Tasks with low success are the main candidates for changes. Revise the tree and test again.
Common mistakes
- Testing with colleagues who know the company's internal structure.
- Hints in the wording of tree test tasks.
- Copying sort results into the menu literally, without expert refinement.
- Testing the tree together with visual design: then you cannot tell whether the problem is the structure or the interface.