Tree testing vs card sorting: which to use, and when to run both
Tree testing vs card sorting
✓ one builds the structure, the other checks it
On this page
TL;DR
Card sorting and tree testing are both information architecture (IA) methods, and they answer opposite questions. Card sorting asks participants to group content, which shows how they think your site or app should be organized. Tree testing asks them to find things in a text-only version of your navigation, which shows whether a structure you already have works.
Card sorting comes first, while you are deciding on a structure. Tree testing comes after, to check it. Most IA projects benefit from running both in that order.
Tree testing vs card sorting at a glance
Card sorting | Tree testing | |
|---|---|---|
| Question it answers | Card sortingHow do people expect this content to be grouped and labeled? | Tree testingCan people find things in this structure? |
| Type of method | Card sortingGenerative: suggests a structure | Tree testingEvaluative: tests a structure |
| When to use it | Card sortingEarly, before or while you design the IA | Tree testingAfter you have a draft or existing IA |
| What participants do | Card sortingSort content cards into groups, and in open sorts name the groups | Tree testingClick through a text-only menu to complete tasks |
| What you get | Card sortingGroupings, agreement between participants, the labels people use | Tree testingSuccess rate, directness, time and the paths people took |
| Typical sample | Card sorting15 to 30 participants | Tree testingAround 50 for a directional quantitative result |
| Needs a design? | Card sortingNo, just a list of content | Tree testingNo, just the navigation hierarchy |
What is card sorting?
In a card sort, each card carries the name of a piece of content or a feature. Participants put the cards into groups that make sense to them. You look across everyone's results for the groupings most people agree on, and the words they use for each group.
There are three types. In an open sort, participants create and name their own groups, which is best for discovering how people think about your content. In a closed sort, you provide the categories and participants place the cards into them, which is best for checking a set of categories you already have in mind. A hybrid sort offers your categories but lets people add their own.
For the full method, including how to write cards and analyze results, see our card sorting guide.
What is tree testing?
In a tree test, participants see your navigation as a plain, clickable hierarchy (the tree) with no visual design. You give them tasks such as “Find where you would change your billing details,” and record where they click and where they end up.
The core metrics are success rate (the share who chose a correct destination), directness (the share who got there without backtracking), and time. The path data shows where people went wrong, which usually points at a specific label or branch.
Tree testing is sometimes called reverse card sorting, because it starts from a structure instead of building one. Our tree testing guide covers writing tasks and reading the results.
The key differences
They answer different questions
Card sorting tells you how people expect content to be organized. Tree testing tells you whether they can find content in the way you have organized it. A card sort cannot tell you if a navigation works, and a tree test cannot tell you what a better one would look like.
They happen at different points in a project
Card sorting is most useful before the structure is settled, when the results can still shape it. Tree testing needs a structure to test, so it comes once you have a draft, or when you want a baseline for the navigation you have today.
They produce different kinds of data
Card sort results are about agreement: which cards most people put together, and which ones nobody agrees on. Tree test results are about performance: per task, how many people succeeded, how directly, and where the rest went instead. Card sorts often lead to a conversation about labels; tree tests produce a ranked list of problems to fix.
They need different sample sizes
Card sorts look for patterns in how people group things, and 15 to 30 participants is usually enough to see them. Tree tests report percentages per task, so they need more people for those numbers to be stable; around 50 is a reasonable starting point for a directional quantitative study. Neither number is a rule, and both go up if you want to compare segments.
When to use card sorting, and when to use tree testing
Use card sorting when
- You are designing navigation for a new product, section or site.
- Your content has grown and nobody is sure the current groupings still make sense.
- You want to know the words your users use for things before you name them.
- Teams disagree about where content belongs and you need evidence from users.
Use tree testing when
- You have a draft IA and want to check it before design and build.
- You want a baseline for the navigation you have today, so you can measure a redesign against it.
- Analytics or support tickets suggest people cannot find something, and you want to know which part of the structure is responsible.
- You are choosing between two candidate structures and want to compare them on the same tasks.
How to run them together
The two methods work best as a sequence, with each one feeding the next:
- Run an open card sort with 15 to 30 participants to see how people group your content and what they call the groups.
- Draft an IA from the strongest groupings, using participants' labels where they agree.
- Optionally, run a closed card sort on the draft categories to check that cards land where you expect.
- Run a tree test on the draft with realistic tasks, especially for content that participants disagreed about in the card sort.
- Fix the labels and branches with low success or directness, then tree test again on the tasks that failed.
Use different participants for the card sort and the tree test where you can. People who have just sorted your content will recognize the labels in the tree, which makes the test look easier than it is. Once the structure holds up, a first click test on the visual design checks that the layout supports it.
Common mistakes
- Treating an open card sort as the final IA. It gives you ideas and language, not a tested structure.
- Writing tree test tasks that use the exact labels in the tree, which turns the test into a word-matching exercise.
- Putting 80 cards in a sort. Fatigue sets in well before that, and the later cards get sorted carelessly.
- Reading directness as success. Someone can go straight to the wrong answer; look at both numbers together.
- Testing only the parts of the tree you are worried about. Include a few tasks for areas you think work, to check the assumption.
Card sorting and tree testing in Great Question
Both are blocks in Great Question's unmoderated tests. Card sorting supports open, closed and hybrid sorts, with an agreement matrix and a tool for merging similar participant-created categories. Tree testing reports success rate, directness and duration, with a heatmap of the tree and each participant's path compared against the correct answers.
Because both draw on the same participant panel, you can recruit from your own customers and send the card sort to one group and the tree test to another. Card sorting and tree testing are available on the Enterprise plan.
To get started, use the card sorting template or the tree testing template.
Frequently asked questions
Should you do card sorting or tree testing first?
Card sorting first. It shows how people expect content to be grouped, which helps you draft a structure. Tree testing then checks whether people can find things in that structure.
Is tree testing the same as reverse card sorting?
Yes. Tree testing is sometimes called reverse card sorting because it works in the opposite direction: instead of participants building groups, they navigate a structure you have already built.
Can you run a card sort and a tree test on the same participants?
You can, but it is better not to. People who have just sorted your content will recognize the labels in the tree, which inflates success rates. Use separate groups if your panel allows it.
How many participants do you need for card sorting and tree testing?
Around 15 to 30 for a card sort, and around 50 for a tree test where you want stable success rates. Add more if you plan to compare segments, such as new and experienced users.
Is a closed card sort the same as a tree test?
No. A closed card sort asks people to place content into categories you provide, which checks whether the categories make sense. A tree test asks people to find content by navigating the full hierarchy, which checks whether the structure works for real tasks, including deeper levels a card sort does not cover.
