An invoice might sit under Account because that’s where a product team manages subscriptions, while the customer looking for it expects a section called Billing. Neither arrangement is particularly strange, but the difference is enough to turn a routine download into a search through several menus. For the team, the location may seem so obvious that it never comes up for discussion until customers start asking for help.
Card sorting offers a way to investigate these differences while there is still room to change the product’s information architecture. By asking people to group cards representing content or features, researchers can explore which things seem to belong together and how people describe those relationships. Nielsen Norman Group’s introduction to card sorting describes the method and its role in developing a structure. The results help a team reconsider familiar categories, although turning them into useful navigation requires an understanding of why participants made their choices.
How card sorting works: open, closed, and hybrid
Consider a fictional software company reorganising its help centre, with articles covering invoices, payment details, invitations, permissions, and reports. Those articles might be maintained by different departments, each with its own terminology and responsibilities. Organising the help centre around those departments would make maintenance straightforward, but it would also ask customers to understand something about the company before they could find an answer.
An open card sort lets participants suggest an arrangement of their own. Someone might gather invoices and payment details under “Money”, while putting invitations and permissions under “People”. Even if those names are too broad for the finished site, they give the team something to work with: a view of the content organised around a customer’s concerns, which can be compared with the structure the company already uses.
The choice of format determines how much freedom participants have to propose a different arrangement. A closed sort gives them categories to work within, making it useful for exploring where they expect particular items to go among a set of options. A hybrid sort allows them to add categories as well as use the supplied ones, though those suggestions may still influence the result. If the team provides “Account management”, for example, its appearance in the completed sort cannot be taken as evidence that customers would have chosen that category themselves.
Interpreting card sorting results
Some of the most useful questions arise when an item lands in several different places. In the help-centre example, “Export a report” might be grouped with invoices by one participant and with team permissions by another. Before treating that as a disagreement about navigation, the researcher needs to find out what each person thought the report contained. A financial statement and a list of employees would quite reasonably suggest different homes.
If the wording caused the confusion, moving the article into the most popular category would leave that problem unresolved. The team could instead try a more specific label, such as “Export a monthly spending report”, and investigate whether people now understand the content well enough to place it. Following up on an uncertain choice therefore helps distinguish a naming problem from a grouping problem, which matters because each calls for a different design response.
This is also why the explanations deserve to be kept alongside the results. A summary showing how often two cards were grouped together can help identify a pattern, but it may leave out the circumstances that make the pattern useful. An administrator handling subscriptions and an employee claiming expenses could approach the same invoice differently; if both use the help centre, their needs should remain visible when the team discusses how to organise it.
Preparing the cards and choosing participants
These interpretive problems begin with the material the researcher chooses. A set of cards labelled “Billing settings”, “Billing history”, and “Billing contacts” gives participants a strong verbal cue to group them together. That may be how they would organise the content anyway, but the repeated wording makes it harder to tell. The task is to describe each item clearly enough to be understood without quietly supplying the categories the study is meant to explore.
A pilot with people from the intended audience helps reveal where that balance has gone wrong. Ask what unfamiliar labels mean to them, notice which items need an explanation, and revise the cards before collecting more responses. Include content that is difficult to place as well as the obvious groups, since a study built entirely around straightforward items may tell the team little about the parts of the help centre that need attention.
As the study develops, keep a record of changes to the cards and instructions so that differences between sessions can be interpreted properly. Where participants have different responsibilities or levels of experience, compare their responses before combining them. The aim is to understand the arrangements the site needs to accommodate, including the cases where a single category may not serve everyone equally well.
Testing the proposed navigation
Once the results suggest a structure, there is another question to answer: can people find what they need within it? A card sort presents the content and asks someone to organise it, whereas using a help centre begins with a task and a set of navigation labels. Even a person who grouped invoices and payment details together might overlook the name a designer eventually gives that section.
Tree testing lets the team investigate this by asking participants to find particular items in a text version of the proposed navigation. Our fictional company could ask someone to locate last month’s invoice without naming the category it expects them to choose, then examine where they go and where they hesitate. Testing the interface afterwards brings in the effects of layout, search, and visual emphasis, which the grouping exercise cannot establish on its own.
The findings may support the proposed categories while also showing that invoices need a direct link from the subscription page. In that case, the card sort has helped the team understand the relationship between the content, while the follow-up has shown how to make it accessible at the moment someone needs it. Together, those decisions give the customer a better chance of finding the invoice without having to learn how the company has filed it.
Running a card sort in Ballpark
If you’re ready to put a study together, the Ballpark card-sorting feature page covers the available formats. For a closer look at one analysis feature, the product update on AI summaries for card-sort data explains what the summaries cover. Treat a summary as a starting point for reviewing the results, checking its account against participants’ groupings and explanations before making design decisions.