Content Strategy and IA
What card sorting actually tells you about information architecture
Card sorting results are often read too literally. This post explains how to interpret sorting data alongside tree testing to build a navigation structure people can actually use.
Overview
A commonly misread method
Card sorting is one of the most common methods in an information architecture project, and one of the most frequently misread. Teams run a study, look at the resulting similarity data, and treat the most popular grouping as the answer. The data is more useful, and more limited, than that. Card sorting tells you how people conceptually group content. It does not tell you how they will search for that content, what labels will make sense in context, or whether the groups will hold up once real navigation constraints like screen space and click depth are applied.
Scope
What the data can and cannot answer
An open card sort asks participants to group a set of items and label the groups themselves. A closed sort asks them to place items into categories the team has already defined. Both produce useful signal about mental models, but neither produces a finished navigation structure. Participants group content based on the information in front of them at that moment, not the mix of tasks, urgency, and prior knowledge they bring to a live product. A grouping that looks clean in a sorting exercise can fail once it competes with a real interface for a user's attention.
This is why card sorting results should be read as hypotheses, not conclusions. The similarity data tells a team where to test a proposed structure, not where to stop testing.
Validation
Reading a card sort alongside tree testing
The most reliable pattern in this research area pairs card sorting with tree testing. Card sorting generates a candidate structure grounded in how participants actually group the content. Tree testing then places that structure in front of a new set of participants and asks them to find specific items within it, without any visual design or styling to help them. If people cannot find what they are looking for inside the proposed tree, the structure needs revision regardless of how clean the original sort looked.
Running these two methods in sequence, rather than relying on either alone, catches a common failure mode: a structure that makes sense conceptually but fails under the pressure of an actual search task. It also gives a team two independent data sets pointing at the same problem, which is a stronger basis for a structural decision than either study alone.
A card sort tells you how people group content. It does not tell you whether they can find it again once real navigation is in front of them.
Language
Labels matter as much as groups
A grouping can be right and still fail if the label attached to it does not match how people describe the content elsewhere. Card sort data is frequently reported without close attention to the labels participants generated for their own groups, even though those labels are often the most direct evidence of the language a navigation system should use. When a team imposes its own labels onto a structure derived from a card sort, it discards some of the most valuable data the method produced.
Takeaways
What to carry forward
Treat card sorting results as hypotheses about structure, not a finished navigation system.
Pair card sorting with tree testing to confirm a proposed structure holds up under real search tasks.
Pay close attention to the labels participants generate during an open sort, not just the groupings.
A structure that looks clean in similarity data can still fail once it competes with a real interface.
References
Further reading
Tankala, Samhita, and Sherwin, Katie. Card Sorting: Uncover Users' Mental Models for Better Information Architecture. Nielsen Norman Group, 2024.
Laubheimer, Page. Tree Testing: Fast, Iterative Evaluation of Menu Labels and Categories. Nielsen Norman Group, 2023.
Nielsen Norman Group. Card Sorting vs. Tree Testing.