skip to content

In information architecture, what is the difference between open and closed card sorting, and when would you use each?

level: middleimportance: must knowfreq 48%

answer

  1. who names the groups?
  2. generative versus evaluative
  3. reveals people's own vocabulary
  4. similarity matrix and clusters
  5. grouping is not finding

basics

~20 s

In an open card sort participants group items and name the groups themselves, revealing their mental model and vocabulary; in a closed sort they place items into predefined categories, testing whether a proposed structure makes sense to them.

solid answer

~50 s

Card sorting asks people to group content items written on cards. In an **open** sort they create and **name** their own groups, so it is **generative**: it shows how they think the content fits together and which words they use — ideal before a structure exists or when replacing one that fails. In a **closed** sort the categories are fixed and participants place cards into them, so it is **evaluative**: it checks whether items land consistently in a proposed structure and exposes cards that split across categories. A **hybrid** sort lets people add a category when none fits. For a museum ticketing kiosk, an open sort of items like “annual pass renewal”, “timed exhibition entry” and “school group booking” might show visitors grouping by what they want to do rather than by department. Card sorting is then followed by tree testing, because grouping items is not the same as finding them.

go deeper

for a junior

Recall who names the groups: participants in an open sort, the team in a closed sort, and that open is generative while closed is evaluative.

for a middle

Explain how to write cards that do not cue groupings, how a similarity matrix is read, and why card sorting is followed by tree testing.

for a senior

Show how you sequence open sort, drafted structure, closed sort and tree test, and how you handle cards whose agreement splits across categories.

for a principal

Decide how much structural research a restructure warrants against its risk, and when existing evidence makes a fresh card sort unnecessary.

## What card sorting is **Card sorting** is an information-architecture method in which participants organise content items — each written on a physical or on-screen card — into groups that make sense to them. It makes people's **mental models** visible: how they believe the content fits together and what they would call the groups. It is a structural method, not a test of screens; participants never see the interface. ## Open, closed and hybrid | | Open sort | Closed sort | Hybrid sort | |---|---|---|---| | Categories | Created and named by participants | Fixed in advance by the team | Fixed, but participants may add new ones | | Nature | Generative — discovers structure | Evaluative — checks a structure | Mostly evaluative, with a discovery outlet | | Best moment | Early, or when the current structure fails | After drafting categories | When confident about most categories, unsure of some | | Main output | Groupings and people's own labels | Placement agreement per card | Both, in part | An **open sort** is the tool when you do not yet trust any structure: it tells you which items people see as related and gives you candidate **labels** in their words. A **closed sort** comes later: you already have categories — perhaps derived from an open sort — and want to know whether each item has an obvious home in them. ## Running a sort well 1. **Choose representative items** — a few dozen is common practice — covering the main content, not every page. 2. **Write cards as content, not as categories**, in plain descriptive words, one concept per card. 3. **Avoid shared words that cue grouping**: cards reading “Member ticket”, “Member renewal” and “Member guest pass” get grouped by the word rather than by meaning. 4. **Decide moderated or unmoderated**: moderated sessions reveal the reasoning behind groupings; unmoderated ones reach more people cheaply. 5. **Recruit enough participants** for patterns to stabilise; commonly cited guidance for open sorts is around fifteen or more, as a convention rather than a rule. ## Reading the results - **Similarity matrix** — for every pair of cards, the share of participants who put them in the same group; high-agreement clusters suggest categories. - **Cluster diagrams** — hierarchical clustering of the matrix, showing which groups merge at which level of agreement. - **Label analysis** — in open sorts, the words people chose for groups; recurring terms are candidate labels. - **Placement agreement** — in closed sorts, the share of people placing each card in each category; cards split across two categories flag overlap or an ambiguous item. - **Comments and hesitation** — in moderated sorts, why people struggled is often more useful than where they put a card. The output is **input to design**, not the design itself. A structure still has to be drafted, labelled and validated. ## Limits of card sorting - **Items are out of context**: people group cards without the task that would bring them to that content. - **Surface similarity**: participants may group by shared words or format rather than by how they would look for things. - **Grouping is not finding**: a grouping people agree on can still fail when they must navigate it top-down under labels. - **No navigation design**: it says nothing about screens, order or wayfinding. That is why teams pair it with **tree testing**, which asks people to find items in a proposed hierarchy. ## A museum kiosk example A museum's kiosk menu mirrors its departments: “Admissions”, “Membership Office”, “Learning”, “Development”. An open sort with visitors, using cards such as “print my online booking”, “renew my annual pass”, “book a school visit” and “add a donation to my ticket”, might show three strong clusters that visitors label roughly “Tickets”, “Members” and “Groups”, with donation split between tickets and members. The team drafts a task-based top level from those clusters, runs a closed sort to check that remaining items find a home, and tree-tests the result before any screen is designed. The same method works for a website or a native mobile app; only the content changes.

  • Why do card labels need care before a sort?
    Words shared between cards cue grouping: cards that all start with “Member” get grouped by the word, not by what they mean. Internal jargon confuses participants and gets sorted by guesswork. Writing cards as plain descriptions of content, one concept each, keeps the results about meaning.
  • What does a similarity matrix show in card sorting results?
    For each pair of cards, the share of participants who placed them in the same group. Pairs with high agreement suggest items that belong together; cards with middling agreement against several groups are candidates for relabelling, splitting or cross-listing in more than one place.
  • Why follow a card sort with a tree test?
    A card sort measures grouping with all items visible at once; a tree test measures whether people can find an item by clicking down through labels. A structure people sort consistently can still fail when a top-level label, read on its own, does not predict what lies beneath it.

saying these in an interview costs you the question

  • An open card sort hands you the final navigation to build.
  • A closed card sort discovers the category names people would use.
  • Cards should include category words so participants sort them accurately.
  • If most people grouped two cards together, users will find them in that menu.
  • Card sorting shows whether people can navigate the finished screens.