Card sorting: open, closed, and what to validate next

Card sorting is the method where participants organize labeled cards into groups that make sense to them. Open sorting lets people create the groups and the names. Closed sorting hands them ready-made categories. Hybrid mixes both. Use open sorting to learn the mental model. To validate the navigation you built, run a tree test, not a closed sort. NN/g sees the two methods as partners, each with its own job.

What the method answers

The Nielsen Norman Group card-sorting guide defines the study as participants placing labeled cards into groups by criteria that make sense to each person. Cards stand for pages, links, images, or descriptions of what someone would find in the product.

Use the method when you need to organize content, services, or offers so people can find them without guessing the team’s internal vocabulary. It doesn’t hand you the final navigation tree. It hands you grouping and naming ideas that still need judgment and another test.

Open, closed, and hybrid

The same guide describes three variations:

Open. No predefined categories. People create groups and label them afterward. NN/g recommends this in most cases, because it shows the mental model without steering the answer.
Closed. Every category arrives named. In theory it validates an existing architecture. In practice, NN/g prefers tree testing for that validation.
Hybrid. Some categories are ready, and people may still create others. The guide advises against it in most cases, because given categories bias the rest. It only fits when you trust one or two categories and doubt the others.

Ask for the group label after the group exists. If people name groups before sorting, they lock themselves into the categories they invented first.

How many cards and how many participants

The guide suggests about 30 to 50 cards. With fewer, the study covers too little. With more, people get tired and an “other” pile fills up with cards they just pushed aside (NN/g).

For sample size:

Qualitative, with think-aloud and follow-up questions: at least about 15 participants (NN/g).
Quantitative, to see pairing frequency: about 30 to 50 participants. If the pattern doesn’t stabilize, recruit more (NN/g).

Even in a remote quantitative sort, run a few moderated sessions. Without them you get the number but not the reason a card was hard to place.

Labels that don’t become a word game

The article on terminology matches warns that participants tend to join cards that share a word, instead of thinking about the concept. Matching a word is easier than thinking about what it means.

Three practical moves:

Use a synonym where the same word would repeat.
Break parallel wording so people can’t match on wording alone.
Add a short description of the content, not only the page name.

Explain at the start that the goal is the concept, ask for think-aloud, and allow rearranging after follow-up questions. A vague label like “employee resources” hides real relationships between items.

How it played on the PicPay home

On the PicPay homepage redesign, the core problem was grouping. The home has to work for someone opening an account for the first time and for someone arriving with a specific financial problem. On the lending and investments side, options had to be grouped so a person could recognize their own situation.

The work set a clear hierarchy among personal loans, payroll-deductible loans, advances on FGTS (the Brazilian severance fund), and vehicle-backed credit, and kept the investing area in plain language. Those grouping and naming calls are what an open sort helps you find before a screen ships. Without the method, the team risks grouping by internal product logic while the person is searching by their own situation.

After the sort: the tree test

The article that compares the two methods keeps the split clear. Card sorting generates organization ideas. Tree testing evaluates a hierarchy you already proposed.

In a tree test, people navigate a category tree as accordions, with no layout and no imagery, until they reach where they think the asked resource lives. The study measures whether the structure and labels support findability. NN/g recommends a tree test after the sort, because a sort almost never returns one “correct” tree.

A common recipe: usability test on the current product, open sort to generate options, tree test to choose and refine the hierarchy before any screen design.

Where the sources diverge

Closed for validation. The definition guide still lists closed sorting as a validation option. The comparison article and the tree-testing article say tree testing is the right method to evaluate navigation.
Hybrid. The guide allows hybrid when one or two categories look solid. In the same text, the general advice is to avoid it, because given categories bias the rest.
What the sort delivers. The definition guide talks about improving findability and discoverability. The comparison article stresses that you mostly leave with grouping ideas, not the final structure.

The common mistake

The mistake is taking sort results and using them as a finished menu. Participants invent a generic category for leftovers, and the same people would avoid a vague label in a real menu. Another mistake is validating the tree with a closed sort. Build the hierarchy, write findability tasks, and run the tree test. The sort shows how people group things. The tree test shows whether they can find them.