Card sorting and tree testing

Two methods, two questions

The two methods answer different questions, and we suggest running them in sequence.

Card sorting

What it is

In a card sort, each participant receives a set of cards, each naming one piece of content or one topic. They sort the cards into groups that make sense to them. Digital NSW's Digital Service Toolkit defines it as participants "organising topics into categories that make sense to them". The US government's 18F method guide gives its purpose as gaining "insights from users about how to organize content in an intuitive way".

Open, closed and hybrid sorts

• Open sort. Participants create and name their own groups. Use it early, when you want to learn how people think about the content and which words they use. Digital NSW recommends an open sort "to learn how users group content and the terms or labels they give each category".
• Closed sort. You supply the group names and participants place cards into them. Use it to check whether an existing or proposed set of categories makes sense.
• Hybrid sort. Participants use your groups but may add their own. It is a useful middle ground when you have a draft structure but suspect it is missing something.

How many cards and participants

Digital NSW suggests a minimum of five to six participants and limiting sorts to 30 to 40 cards, especially for open sorts. More participants give more confidence in the patterns, and remote tools make larger groups affordable. The 18F guide estimates 15 to 30 minutes per participant.

Choosing the cards

Each card should name one topic, in plain words that do not give away the answer. Avoid copying your current menu labels onto the cards, because participants will simply rebuild your current menu. Avoid labels that share a distinctive word, such as "Parking permits" and "Building permits", because people group by matching words rather than meaning. A mix of popular content and content you are unsure about gives useful results.

Running the sort

Sorts can run in person with paper cards or remotely with an online tool. In person, you can ask participants to think aloud, which explains their choices. Remotely, you reach more people for less effort. 18F's steps for an open sort are a good outline: give people the cards, ask them to group them, ask them to name each group, then ask why they grouped and labelled things as they did.

Reading the results

Look for cards that most participants placed together, cards that ended up in many different groups, and the words participants used for group names. Cards that scatter widely are the ones your structure needs to handle carefully, perhaps by placing them in one group and linking to them from others.

Tree testing

What it is

A tree test gives participants a text-only version of your proposed structure, such as the menu labels expanded level by level, and a series of tasks: "Where would you go to find out how to apply for a parking permit?" They click through the tree until they choose the place they think holds the answer. The European Commission's web guide describes tree testing as applying "tasks and scenarios taken from user research" to evaluate the structure, so that it matches users' logic.

Because there is no visual design, a tree test isolates the structure and labels. If people fail a task, the cause is the grouping or the wording, not a colour or an image.

Writing tasks

• Base tasks on real visitor needs from your task list, not on what the organisation wants to promote.
• Do not use the words that appear in the target label. "Find the page about parking permits" tests reading, not structure.
• Cover the important tasks and the ones you are least sure about. We plan ten to fifteen tasks per participant so that attention holds to the end. That is our working range, not a published rule.

What to measure

• Success. Did the participant choose the right place?
• Directness. Did they go straight there, or wander and backtrack?
• First click. Where did they go first? The rest of the path depends on that choice, so a wrong first click is worth studying even when the participant recovers.
• Time. Slow, successful tasks still point to a label that needs thought.

Acting on the results

When a task fails, look at where people went instead. If most went to one wrong place, the content may belong there, or the label of the right place may need changing. Revise the tree and test again with new participants. A second round shows whether your changes fixed the problem or moved it somewhere else.

Running both on a small budget

You do not need a research agency to learn something useful.

1. Recruit five to ten people from your real audience: customers, members, students or residents. Your newsletter or a short notice on the site can find them.
2. Run an open card sort with 30 to 40 cards, using a free or low-cost online tool or paper cards.
3. Draft a structure from the results.
4. Tree test it with a fresh group and ten or so tasks.
5. Revise and, if time allows, test again.

Offer a small thank-you, such as a gift card, and keep sessions short.

Privacy when recruiting participants

Recruiting participants means collecting names and contact details, so the Australian Privacy Principles may apply. The OAIC explains that the Act reaches only some small businesses, meaning those with turnover of $3 million or less, and that exceptions such as health service providers are covered whatever their size. Even if the Act does not apply to you, it is good practice to tell participants what you collect, why, and how long you keep it, in line with APP 5 on notification. Use testing tools that let you avoid collecting more than you need.

How these methods fit into a project

Card sorting and tree testing are not the whole of information architecture. They sit alongside a content inventory, task analysis and staff interviews. They earn their cost where the structure involves real choices, such as how to split services between sections or what to call a group of content that has no obvious name. Our navigation design and enterprise services include optional testing, quoted separately because it involves participants and tools.

If you want a quick check on your existing labels before recruiting anyone, the taxonomy planner will flag duplicates and gaps.

Sources checked 4 November 2026 for card sorting and tree testing

1. Structure testing, European Commissionsecondary
2. APP quick reference, OAIC
3. Card sorting, Digital NSW
4. Card sorting method, 18Fsecondary
5. Small business, OAIC