Card sorting: understanding someone else's structure without suggesting your own
When should you use open or closed card sorting? Prepare topics, conduct a study and turn disputed groups into a hypothesis for site structure.
Article language: English

In this article
The category “Other” can hold anything, especially things its creator could not name. If participants put half the topics there in a study with predefined categories, you do not yet have a structure for the site. You have learned that the boxes you prepared were not enough. That is a useful result, provided you do not present it as a minor departure from an approved menu.
Card sorting helps you understand how people connect content. A participant receives topics on separate cards, arranges them into groups and, depending on the format, either names the groups or uses categories supplied in advance. The task is not to guess the company's structure. Nor is it a test of how neatly someone can move rectangles around.
Imagine a fictional library of guides for people organising small exhibitions. It includes material on preparing a room, transporting exhibits, signs and helping visitors. The team wants to understand relationships between topics before designing navigation. Below is a proposed way to work, not a report of invented answers from a completed study.
Should participants create categories or test yours?
Digital.gov distinguishes open sorting, in which participants group and name categories themselves, from closed sorting, which supplies categories in advance. It suggests asking about the reasons for assignments in both. The difference is therefore not just in how a tool looks, but in what you can learn.
If we do not know how the audience understands the library as a whole, supplying headings such as “Planning”, “Delivery” and “Development” imposes part of the answer. Participants can still reveal disagreements over where cards belong, but they cannot freely provide their own division. That is a useful method for a different question, not a failed version of the same exercise.
In our example, the initial question is: which materials do organisers consider related, and what names do they give those connections? We choose an open sort. Later, when a draft set of categories exists, a separate closed sort can be prepared. Do not combine the two sets of results without showing that participants worked under different conditions.
If the same people return for both stages, record their earlier participation. In the second study they may remember their own groups or names heard during discussion. They will not start from the same position as someone encountering the categories for the first time. You can study the same group again, but frame the question around change or refinement rather than presenting the results as entirely independent confirmation.
Website information architecture (in Polish) also covers the jobs of individual pages and the links between them. Sorting supplies a particular kind of evidence for that work. It does not automatically determine the menu hierarchy, business priorities or the number of levels visitors will be willing to navigate.
In a co-design workshop (in Polish), users can create and change proposed solutions. Sorting alone has a narrower purpose. Do not quietly expand it into a vote on the site's appearance: you would lose track of which answers concerned grouping and which expressed an aesthetic preference.
Preparing the cards is part of the research
First define the part of the library covered by the decision. If you want to understand its overall structure, a set consisting only of guides about signs will not show their relationship to room preparation. At the other extreme, adding the entire company archive may make participants manage their fatigue instead of the content.
For our study, we select topics representing different areas and known points of uncertainty. That is a decision for this proposed study, not a universal number of cards. Alongside a clear card such as “Preparing labels for exhibits”, there might be a borderline topic: “Planning how visitors move through the room”. We want to hear why someone connects it with the space or with visitor assistance.
Card names should make their contents recognisable. “See more possibilities” is not a topic. “Organising flow” may need clarification: does it concern people, materials or project approval? Do not put the name of the category where you think the card belongs in its description.
Give every card a stable identifier and the set a version. If a preparatory test leads you to improve an unclear name, record the change. A card first called “Support” and later “Helping visitors during the exhibition” may be grouped differently. Do not mechanically combine the results just because the card has the same number in the system. What participants actually received determines the conditions for comparison.
In a 2018 account of its studies, the GDS team reported that titles of existing pages made card contents difficult for participants to understand. This is a historical observation from a particular project. It suggests checking the clarity of your own material rather than assuming a page title is ready to become a card.
Each card should have one recognisable meaning. If it combines preparing exhibit labels with handling visitor enquiries, participants may assign it according to either half. An analysis will then show disagreement, but the material's construction may be the cause. Look at the card itself before calling that disagreement a difference in understanding the topic.
UX research recruitment (in Polish) should follow the research question. Someone organising their first exhibition and someone who regularly prepares exhibitions may understand the same topic differently. Do not replace them solely with employees who know the company's names for its guides. You could easily end up testing memory of the structure you are trying to change.
Before the main round, test both the material and the way people use it. A participant should understand the instructions, be able to move or assign a card and name a group of their own. Distinguish difficulty with the tool from hesitation about the content. Offer a way to participate that does not make expressing one's own structure depend on being able to drag items.
Hear the reasoning behind the arrangement
A neutral instruction might be: “Put together the topics that you think belong together. Make as many groups as you need and name them in your own words.” In the discussion that follows, ask about selected reasons and unclear cards. This is a proposal for our study and needs testing with the prepared material.

Do not add, “Shouldn't that be under room preparation?” The question already contains the expected assignment. Ask instead how the participant understands the topic and why it went into the chosen group. Let them change their mind, while noting why. The final arrangement will not show that on its own.
Allow people to flag a topic they do not understand or cannot place. In a closed sort, forcing a choice among the categories can turn hesitation into apparent agreement. If the study aims to understand those boundaries, provide a way to record them. Asking “Where would you look for this if you had to choose?” explores something different from a spontaneous “This doesn't fit anywhere for me”. Keep that distinction in the report.
Record group names exactly as given, without immediately correcting them into company language. A company glossary (in Polish) helps establish shared internal meanings. Here we are also studying how the audience speaks. A name does not have to be ready for use as a page heading to be valuable evidence.
If you use automatic transcription, keep each statement connected to the speaker and the specific card. When checking who is speaking in an AI transcript (in Polish), we distinguish the speaker label from the words themselves. Assigning an explanation to the researcher instead of the participant, or vice versa, changes the meaning of the note even if the words are clear.
Do not merge meanings along with spelling mistakes
After the study, compare both group names and their contents. Two people may write “Preparation”, while one means materials before transport and the other means work already done in the room. Identical wording does not prove the categories mean the same thing. Conversely, “For guests” and “Visitors” may cover similar sets of topics.
Optimal's documentation on standardising categories recommends examining the meaning of labels and the cards in groups being merged. It describes analysis in a particular tool. Similar words alone are not enough to merge responses as agreements.
Keep the original arrangements and record your own standardisation separately. When combining two names, add the reason and references to the cards and explanations. That lets you revisit the decision if further analysis shows that one shared label concealed two different ways of thinking.
Separate how often cards are grouped together from similarity of group names. Two cards may regularly appear together in groups with different labels. That signals a relationship between the content, but does not yet choose a heading. A frequently used name is of little help, on the other hand, if the contents of groups bearing it differ each time. In our example, “On site” may be a useful linguistic lead, but it requires looking at what each person put inside.
For “Planning visitor movement”, do not look for a winner at any cost. Disagreement may suggest a need to link to the topic from several places. It may also mean the name is too broad. The decision should follow from the material and another test, rather than a rule that every card must have one permanent home.
In a numerical summary, show how many people actually saw each card and how many completed the relevant part of the study. Do not compare raw assignment counts from sets with different coverage as though they shared a denominator. Record why an interrupted session was omitted. Rejecting answers merely because a participant created an unexpected division would be fitting the evidence to a ready-made menu. An unusual but considered arrangement remains an observation to understand.
The result is a hypothesis to use
The Department for Education described using a sort to prepare a draft structure and planned a separate tree test. A tree test checks whether people can find content in a category tree without the site's full visual design. That stage answers a different question from grouping topics. It is also worth turning your own grouping into a specific question about use.
For our library, prepare a draft with proposed names and a task: find the information needed to plan entry to the exhibition. A first-click test (in Polish) can later check whether the first screen points people towards the right starting path. It replaces neither sorting nor completing the whole task.
Preserve the path from observation to proposal. For the room plan, you might record: participants connect the topic both with the space and with preparing a visit; we propose a page linked from both places; we will then test whether people can find it using two different tasks. This is an example for a future report, not a finding about actual participants. It shows how a grouping result leads to a decision that can later be challenged or supported.
In work on content structure with Aeneia Design, you can discuss the content scope and navigation design. If research with participants is needed, agree before work starts who will prepare and conduct it, the scope of their responsibilities and the terms.
Aeneia Design
Discuss the question about content structure and how navigation might be tested. Keep participants' original groups separate from design decisions; agree the scope of any research beforehand.
Discuss preparing a content-grouping study
Close the analysis with a short decision for each disputed area: what remains a hypothesis, what needs renaming and what we still do not know. If “Other” still contains half the library, do not shrink the font on that card. Return to the meanings your proposed structure could not accommodate.







