Card Sorting

Card sorting is a research method in which participants organize a set of items into groups that make sense to them, and often name those groups themselves. It is used primarily to inform information architecture: how content should be categorized, what those categories should be called, and where individual items belong. Each item is written on a card, physical or digital, and the participant's groupings reveal their mental model of the domain.

Three variants serve different purposes. In an open sort, participants create and name their own categories, which is appropriate when the structure is undecided and the goal is to discover how people naturally organize the material. In a closed sort, categories are fixed and participants place items into them, which validates a proposed structure and identifies items whose placement is ambiguous. A hybrid sort provides categories while allowing participants to add their own, which tests a draft structure while leaving room for gaps to surface.

The method addresses a specific and expensive failure: sites organized around internal structure rather than customer understanding. Organizations categorize by business unit, product family, or the way their systems store data, and customers arrive with entirely different groupings based on their goals. The mismatch produces navigation nobody uses, site searches that return nothing, and support contacts about information that is present but unfindable. Card sorting exposes the gap before the structure is built rather than after.

Analysis looks for agreement rather than for a single correct answer. Similarity matrices show which items were repeatedly grouped together, dendrograms visualize the clustering, and the category names participants supply reveal the vocabulary that should appear in navigation. Items that different participants placed in different groups are the most informative results, because they identify content whose location will need to be duplicated, cross-linked, or made findable by search rather than by browsing.

The method has real limits worth stating. Card sorting reveals how people group items when presented with them all at once and with no task in mind, which is not how anyone uses a website. It says nothing about whether people will find an item when looking for it in a live structure, which is what tree testing measures. It is also sensitive to how the cards are worded, since the labels supplied become the primary information participants have. For this reason card sorting is usually paired with tree testing: the first generates a candidate structure, the second validates whether it works for real tasks.

Interpreting the results requires resisting the temptation to design by majority. The output describes how a particular set of participants grouped a particular set of labels, and directly implementing the most common grouping frequently produces a structure that satisfies nobody, particularly where distinct audience segments organize the domain differently. Analyzing results by segment usually reveals whether one structure can serve everyone or whether multiple routes to the same content are required. Where segments genuinely differ, the resolution is normally cross-linking, multiple entry points, and strong search rather than choosing a winner, since forcing one group to navigate according to another group's model is a reliable way to lose them.

Sample sizes are modest, with most practitioners finding that agreement patterns stabilize after roughly fifteen to thirty participants for a given audience segment, and unmoderated online tools make it inexpensive to run at that scale. In practice the method sits early in a product design engagement, before navigation and taxonomy are settled, and is normally run as part of the wider user research programme, since the vocabulary it produces is useful well beyond navigation, informing page headings, search synonyms, and the terminology used across marketing content.