Information architecture · Card sorting

Card Sorting Guide: Types, Setup, and Analysis

By UXbeam · Information architecture tools since 2021 · Published September 20, 2026

Card sorting asks people to group content or place it in supplied categories. It shows how participants would group the content, rather than forcing them into the structure your team already uses. This guide covers the method: sort types, cards, participants, analysis, and what to do with a result. Want to run one? Open the free card sorting tool.

What a card sort measures, and what it does not

A sort produces two kinds of evidence. Grouping: which cards people put together, and how consistently. Naming: participant-created category names in open and hybrid sorts. Grouping tells you about structure; naming gives you evidence about the labels participants naturally reach for.

A sort does not measure findability. Participants see every card and every group at once; in the product, someone sees one level of labels and has to guess. Agreement in a sort is a good sign for a structure and no guarantee of it. That question belongs to tree testing. Use card sorting to propose the structure, then tree testing to test findability.

You provide

15 to 60 cards, one concept each, with an optional one-line description. For closed and hybrid sorts, a list of categories.

The participant does

Drags cards into groups in a browser. No account, no install. In an open sort they name the groups; in a closed sort they use yours; in a hybrid sort, both.

You get

Every individual sort, a similarity dendrogram, a similarity matrix, the category labels people wrote, the cards that never settled, and the pattern summary for the sort type: mental models, category-use patterns, or organization patterns.

When to run one

Open, closed and hybrid sorts

The three sort types differ in who supplies categories and whether participants can add them, and that changes the question the study can answer, the instruction participants see, and how to read the result.

The pattern summaries are named differently on purpose. A closed sort constrains what participants can express, so differences can reflect how participants interpret the supplied categories. UXbeam calls these category-use patterns.

Open card sorting

An open sort gives participants the most room to surprise you. Participants get the cards and nothing else: they decide how many groups to make, what goes in each, and what to call them. The result reflects the organizing principle they reach for, by topic, by task, by who it is for, by where it happens. That organizing principle can change the entire structure that follows, and a closed sort cannot reveal it because you have already supplied the categories.

Read the labels first. Participants' category names show the language they reached for while organizing the material. Two groups can produce the same groupings under different names, or different groupings under the same name; both matter; hybrid sorts can also include participant-created groups.

Watch for participants who make one group per card, or one group for everything; UXbeam shows every individual sort so you can review unusual responses. Split placements can reflect wording, overlapping categories or different organizing principles. Inspect the sorts before choosing a fix. And when the groupings split into coherent patterns, resist averaging them; the next sections cover why.

The full protocol, from scope and card preparation to the handoff into analysis, is on the open card sorting page.

Writing the cards

1. One concept per card, 15 to 60 cards

   UXbeam accepts 15 to 60. Larger sets increase participant effort, so sample a very large inventory or split it into focused studies.

2. Write in the user's language

   Order tracking, not OMS status view. If a card needs explaining, add a description after ||; participants can reveal it, and UXbeam reports how often they did, which can flag card names to review.

3. Keep category words out of card names

   Billing settings answers the question for the participant. Name the thing and let people decide where it lives.

4. One level of granularity

   Mixing broad topics with specific features can make the sorting task unclear.

5. Randomize the order

   Cards, and in closed and hybrid sorts, categories too. A fixed order can influence grouping.

How many participants

There is no universal number; the practical test is stability. Check whether important groupings change as responses arrive. Choose the sample around the question and the audiences you need to understand. Deciding between two competing structures, or trusting a minority pattern, asks for more.

Recruitment is yours. UXbeam gives you a study link for your own users, customers or a panel you already work with. Participants need no account, and responses are not capped.

Reading the results

UXbeam builds the analysis from your results as they arrive:

• Similarity dendrogram with a grouping threshold that can be adjusted for eligible open and hybrid results: the combined structure across all participants. Cards often sorted together sit close.
• Similarity matrix. How often any two cards shared a group, for checking a specific pair.
• Category labels. The names participants wrote, with counts; in closed and hybrid sorts, how your categories were used and what was added.
• Grouping diagnostics. Inspect group agreement and cards with uncertain or contrasting placements.
• Every individual sort, to read one participant's logic or exclude a careless one.

Read the dendrogram in two passes. Find a threshold that gives a useful number of top-level groups and look at what each group holds. Then move the threshold and watch which cards move. A grouping that stays together across nearby cuts is less sensitive to the chosen threshold. One that moves as the threshold changes is worth inspecting: inspect its pair counts and individual placements before deciding why.

For the order in which to read these views, and what each one can and cannot settle, see How to analyze card sorting results.

Mental Model Detection

A dendrogram pools every sort into one tree. When participants organize the same cards in two coherent ways, it can summarize them in a structure that matches neither grouping. Mental Model Detection keeps those distinct mental models separate. Mental Model Detection identifies recurring ways participants organized the cards. It analyzes individual participant sorts, so its findings depend on the cards and responses included. Use it to investigate different mental models that a pooled similarity matrix or dendrogram may obscure.

When more than one mental model appears, open each as its own candidate sitemap with Sitemap + and test the alternatives against the same tasks, instead of averaging them.

From grouping to sitemap to tree test

• Pick a groupingThe dendrogram at a threshold, or one detected mental model.
• Sitemap +Opens it as an editable sitemap, pre-populated. Rename, merge, move.
• Save a versionThe candidate structure you intend to test, kept apart from the sort data.
• Tree Test thisThe sitemap opens as a tree test. Write tasks, share the link.
• Read task successWhere people went, where they gave up, which labels misled.

If the goal is findability, follow the sort with a tree test. A sort shows that people agree where a card belongs when they can see all the options; a tree test shows whether they can find it when they cannot. A tree test measures task performance in the candidate hierarchy. See the tree testing guide, and the examples for a sort carried through to a two-structure tree test.

In this guide

See also: Organizing principles in information architecture · Tree testing vs card sorting · The UXbeam card sorting tool

Frequently asked questions