UX Foundations · Step 4 · 7 min read
Card sorting and tree testing
Two simple methods to build a structure people understand. Card sorting shows how people group your content; tree testing checks whether they can find things in your proposed structure. Open, closed and hybrid sorts, how many participants NN/g recommends, and how to read the results.
You can't guess how other people group things; you have to ask them. Your team knows the product too well to see it like a newcomer. Card sorting hands people a pile of cards and lets them group them their way. Tree testing hands them your proposed menu, with no design at all, and asks them to find things in it. One method gives you ideas; the other checks them. Both are cheap, and both can start with paper.
What you'll be able to do after reading:
- Explain the difference between card sorting and tree testing
- Choose an open, closed or hybrid card sort
- Plan how many participants you need, and know why the numbers differ
- Read the results without over-trusting them
1. Two methods, two questions
- 1Card sort
People group your content their way.
Generates ideas - 2Draft a tree
You turn the patterns into a structure.
Groups and labels - 3Tree test
People look for things in that structure.
Evaluates it - 4Fix and design
Change labels and places, then design screens.
Then usability tests
Test the tree again after big changes.
- Card sorting asks: how do people think this content fits together? NN/g defines it as "a research method in which study participants place individually labeled cards into groups according to criteria that make the most sense to them" (NN/g, Tankala and Sherwin, Card Sorting: Uncover Users' Mental Models, 2024). It's a generative method: it produces ideas for a structure.
- Tree testing asks: can people find things in this structure? It is "an evaluation of a hierarchical category structure, or tree, by having users find the locations in the tree where specific resources or features can be found" (NN/g, Laubheimer, Tree Testing, 2023). It's an evaluative method: it checks a structure you already drafted.
NN/g's common recipe: test the current product, run a card sort to generate options, then tree-test one or more of them before designing screens.
2. Card sorting: open, closed or hybrid
People make and name their own groups.
Best for learning how they thinkPeople place cards into your groups.
A tree test usually does this betterPeople use yours or add new ones.
Only when you are sure of one or twoThe cards can be pages, products, features or topics. NN/g recommends 30 to 50 cards to avoid fatigue, and warns against repeating words across cards: if three cards start with "Toyota", people will group them by brand without thinking further.
| Type | What participants do | NN/g's advice |
|---|---|---|
| Open | Create and name their own groups | "The most common type"; recommended because you see people's mental models |
| Closed | Put cards into groups you give them | "Sometimes used for validation", but tree testing is usually better |
| Hybrid | Use your groups or add their own | Only if you're confident about one or two groups; your groups can bias the rest |
Tell people that groups can be any size, that there's no right number of groups, that they can change their mind, and that cards they don't understand can go to one side. Ask them to name the groups after sorting, and to think aloud while they work.
3. How many people?
The numbers differ because the methods do different jobs.
| Study | NN/g's recommendation | Why |
|---|---|---|
| Usability test (qualitative) | 5 per round | You're looking for problems; see Usability testing with five people |
| Card sort, qualitative | At least 15 | People's mental models vary a lot |
| Card sort, quantitative | 30 to 50 | To trust the statistics |
| Tree test, qualitative | "Only a few participants" | You're looking for confusing labels |
| Tree test, quantitative | "A larger sample size" (NN/g's general advice for quantitative studies is about 40) | To measure and compare success rates |
The card-sort numbers come from Tankala and Sherwin (2024) and from Nielsen's 2004 analysis of a study by Tullis and Wood: with 5 participants, results correlated only 0.75 with the final picture; 15 reached 0.90, and 30 reached 0.95 (NN/g, Nielsen, Card Sorting: How Many Users to Test, 2004). Tree-testing guidance is from Laubheimer (2023); the 40 is from NN/g, Budiu and Moran, How Many Participants for Quantitative Usability Studies, 2021.
A sort with three to five people (like the Run a card sort lab) is a good way to learn the method and spot obvious problems, not a basis for a final structure.
4. Reading card-sort results
Look for these, in this order:
- Pairs that always go together. If almost everyone puts "Pots" with "Soil", that's a strong group. Tools show this as a similarity matrix; on paper, count it.
- The names people give groups. These are candidate labels in your users' own words.
- Cards left aside or moved around. They usually have unclear names, or belong in more than one place.
- "Miscellaneous" piles. A sign of tired participants, too many cards, or content that doesn't fit.
- Real disagreements. Two different but consistent ways of grouping may mean two kinds of users.
A worked example from a fictional plant shop:
| Card | Most common group | Placed there by | What it tells you |
|---|---|---|---|
| Snake plant | Indoor plants | 4 of 5 | Strong; keep |
| Watering can | Tools | 5 of 5 | Strong; keep |
| Plant food | Tools (2), Soil and care (3) | Split | Needs a clearer label, or a place in both |
| Gift cards | Left aside by 3 | Unclear | Not a plant; probably belongs in utility navigation |
Don't follow the numbers blindly. Nielsen: "I don't recommend designing an information architecture based purely on a card sort's numeric similarity scores"; the comments people make while sorting matter as much. And people's group names are ideas, not final labels.
5. Tree testing in practice
You need two things: the tree (your proposed menu as a plain list, every level) and tasks. Write tasks as goals, without the words in your labels. NN/g's example: "Find information about starting a business" gives the answer away; "You are considering opening a lawn-care service. See if there are any resources on this site that can help you begin the process" doesn't.
For each task, NN/g looks at success (found the right place), directness (went there without backtracking) and time, plus the first click (NN/g, Laubheimer, Tree Testing Part 2: Interpreting the Results, 2024). In a review of 98 tree tests cited there, Albert and Tullis found a median task success of 62%. Because tree tests strip away design, search and content, scores are usually lower than in the finished product. Compare tasks with each other and with earlier versions, not with 100%.
6. Common mistakes
- Using a card sort to validate a structure; that's a tree test's job.
- Card labels that give away the group ("Garden tools: spade").
- Copying participants' group names straight into the menu.
- Tree-test tasks that use the exact label words.
- Testing only with colleagues, who know the current structure.
- Deciding from five card sorts as if they were fifty.
7. Vocabulary
| Term | Meaning |
|---|---|
| Card sorting | People group labelled cards in the way that makes sense to them |
| Open / closed / hybrid sort | People make the groups / you give the groups / a mix |
| Tree testing | People find items in a text-only version of your menu |
| Tree | The hierarchy written as a plain list of levels |
| Similarity matrix | A table of how often each pair of cards was grouped together |
| Directness | Share of people who went to the right place without backtracking |
| Generative / evaluative | A method that produces ideas / one that checks a design |
8. Try it (no tools needed)
Do the Run a card sort lab: write 20 to 30 paper cards for a product you know, ask three to five people to sort them, and look for patterns.
Reflection questions: Which cards did everyone agree on? Which group name surprised you? What would you test next with a tree test?
9. If you already practise
- Run a few moderated sessions even when you run an unmoderated online sort; the reasons behind the groups are where the insight is.
- Tree-test the current structure before a redesign, so you have a baseline to compare against.
- Write tree-test tasks for your top tasks first, then for the labels your team argues about.
Checklist
Learn more
- NN/g: Card Sorting: Uncover Users' Mental Models for Better Information Architecture (2024) · Card Sorting: How Many Users to Test (2004) · Tree Testing: Evaluate Menu Labels and Categories (2023) · Tree Testing Part 2: Interpreting the Results (2024) · How Many Participants for Quantitative Usability Studies (2021)
Books, courses and videos to go deeper
- Card Sorting: Designing Usable CategoriesDonna SpencerBook · Rosenfeld Media, 2009 (external site)
- Information Architecture: For the Web and BeyondLouis Rosenfeld, Peter Morville and Jorge ArangoBook · 4th edition, O'Reilly, 2015 (Amazon)
Book links go to Amazon. As an Amazon Associate I earn from qualifying purchases, at no extra cost to you.
Related
Run a card sort · Information architecture · Navigation patterns · Usability testing with five people · Affinity mapping