Definition
Card sorting is a UX research technique where participants organise topics or items into groups that make sense to them — used to inform information architecture, navigation and labelling.
There are three flavours: open card sort (users name their own groups), closed card sort (users assign cards to predefined groups), and hybrid (users can use the predefined groups OR add their own). Open sorts surface the user's mental model; closed sorts validate yours.
The output isn't 'do whatever the average user did'. The output is a similarity matrix and a dendrogram — which items users group together, with what confidence. Outliers and disagreements are often more revealing than consensus.
Origin
Used in library science from the 1980s for cataloguing systems. Adopted in UX through Donna Spencer's 2009 book 'Card Sorting: Designing Usable Categories'. Now standard in IA practice.
How it works
- List 30–80 cards (any more and participants tire).
- Recruit 15–30 participants representative of the audience.
- Run sorts remotely (Optimal Workshop, UserZoom) or in person.
- Analyse: similarity matrix shows which cards cluster.
- Render as a dendrogram to see hierarchical groupings.
- Cross-reference with qualitative reasons users gave.
When to use it
Use when
- Before a major navigation redesign.
- When the team disagrees about how to organise content.
- After significant content additions — old IA may not fit.
Skip when
- On tiny content sets (under 20 items). The exercise has too few cards to be useful.
- When the issue is wording, not structure. Tree testing is the better tool there.
Key metrics
- Agreement strength (how often two cards land in the same group).
- Cluster confidence (how tight the groupings are).
- Outlier items (cards that nobody knew where to put).
- Group naming consistency (in open sorts).
Examples
- Card sorting revealed nobody thought of 'Resources' as a category — they'd been searching by format (PDFs, videos).
- We ran an open sort with 24 users and the dendrogram suggested two top-level categories, not the seven we had.
- The card sort confirmed our IA was wrong, but it didn't tell us what was right. Tree testing did.
In practice at Makreate
Makreate's UX research engagements use card sorting to inform navigation on marketing sites and product menus. A recent SaaS client was about to ship a 12-item top nav. We ran an open sort with 22 customers; the dendrogram showed they grouped features by job-to-be-done, not by feature category. We rebuilt the nav around three jobs and a 'See all features' link. Bounce rate dropped 18% on the homepage and feature-page traffic redistributed in line with what the sort predicted.
UX Research →Common mistakes
- Too many cards. Over 80 and participants tire and start grouping randomly.
- Recruiting people unfamiliar with the domain. They group by guesswork.
- Acting on the average. Variance and outliers are usually more informative.
- Skipping the qualitative follow-up. 'Why did you put it here?' is half the value.
Frequently asked
Open or closed card sort?
Open if you don't know the right structure. Closed if you have a hypothesis to validate. Hybrid is the most common in practice.
How many participants?
15–30 is the consensus minimum. Below 15 the patterns are noisy; above 30 you usually see diminishing returns.
Card sorting vs tree testing?
Card sorting designs the IA; tree testing validates it. Run them in that order.