Card Sorting in UX Research: A Complete Guide

Card sorting is a UX research method in which participants group a set of cards -- each representing a piece of content, a feature, or a category -- into arrangements that make intuitive sense to them. The output is a window into users' mental models: how they naturally organise information, what they expect to find grouped together, and what vocabulary they use for things designers often name differently. Information architects and UX designers use card sorting to build navigation structures and content hierarchies that reflect the way users think, rather than the way internal teams think about their own products.

The method has been a standard part of the UX research toolkit for decades. The Nielsen Norman Group has published extensive research on card sorting as an information architecture tool, and it remains one of the clearest examples of research that directly drives a structural design decision.

Want to practise card sorting on a real client brief? UX Academy (myuxacademy.com)'s Intermediate UX Design course includes a hands-on IA research module where you run live card sorts and present findings to stakeholders. Or try the free masterclass first.

What Problem Does Card Sorting Solve?

The core problem card sorting addresses is this: the people who design and build a product think about its content differently from the people who use it. Internal teams organise things by department, product line, or business function. Users organise things by task, goal, or how they understand the world.

When those two mental models diverge -- and they almost always do -- users cannot find what they are looking for. They either give up, contact support, or leave. Card sorting gives you evidence to close that gap before you build a navigation structure that makes sense only to the people who created it.

It sits within the broader category of UX research methods that inform information architecture decisions. For a fuller picture of how IA fits into the design process, see information architecture in UX design.

Open, Closed, and Hybrid Card Sorting

Open Card Sorting

In open card sorting, participants receive a set of unlabelled cards and create their own category groups. They name those groups themselves, using whatever label feels natural to them.

This is the variant to reach for when you are designing a new information architecture from scratch and do not yet have a proposed structure. Open card sorting tells you two things: how users group content together, and what they call those groups. Both are valuable. Users' own category labels often surface vocabulary that should feed directly into navigation labels and copy.

Use it when: You are starting a new project or undertaking a significant redesign and have no existing structure to validate.

Closed Card Sorting

In closed card sorting, participants sort cards into categories you have already defined. The categories are fixed; participants cannot create new ones.

This is the variant to use when you have a proposed or existing navigation structure and want to test whether users can predict where things live within it. It answers a different question from open sorting -- not "how do users naturally group this?" but "does our proposed grouping match how users think?"

Use it when: You want to validate an existing or proposed IA before committing to it, or when you are testing whether a redesign has improved findability.

Hybrid Card Sorting

Hybrid card sorting gives participants a set of predefined categories but allows them to create new ones if a card does not fit anywhere. It is a middle ground: you get data about your proposed structure while also learning what gaps exist in it.

Use it when: You have a proposed structure but suspect there may be content that does not fit cleanly anywhere within it.

When Should You Use Card Sorting?

Card sorting belongs at the generative end of the research process -- early, before you have built a structure. The ideal sequence is:

1. Run an open card sort to understand users' mental models and vocabulary.
2. Draft a proposed information architecture based on the findings.
3. Run a tree test to validate that users can actually navigate the proposed structure.

(Tree testing is the evaluative counterpart to card sorting -- it tests whether people can find things in a structure you have already designed, without the visual design getting in the way. It is worth thinking of the two methods as a pair.)

Card sorting is particularly valuable for:

• Website navigation redesigns -- especially on sites with large content libraries where the current navigation grew organically rather than by design.
• Product menus and settings -- particularly in B2B tools where users need to find specific functionality quickly.
• Content categorisation -- blog taxonomies, help centre structures, e-commerce category hierarchies.

It is less useful for task-based or flow-based design problems, where usability testing or user interviews will give you more relevant insight.

How to Run a Card Sorting Session

Step 1: Define Your Content Set

Select the cards carefully. Aim for 30 to 100 cards -- enough to surface meaningful groupings, few enough that participants do not become fatigued. Each card should represent a single, clearly defined piece of content or feature. Avoid overlap between cards and avoid jargon that might confuse participants before they have even started sorting.

If you have more than 100 pieces of content to sort, consider running multiple studies with different subsets, or prioritising the content that is most frequently accessed.

Step 2: Choose Your Delivery Method

Remote, unmoderated: Tools like Optimal Workshop's OptimalSort or Maze let participants complete the sort online, at their own pace. This approach scales well -- you can reach 20 to 30 participants quickly -- and removes the logistical overhead of scheduling sessions. The trade-off is that you cannot ask follow-up questions when a participant does something unexpected.

Remote, moderated: Screen-sharing tools combined with a digital card sort tool allow you to observe in real time and ask participants to think aloud as they sort. You lose some scale but gain much richer qualitative data -- you will hear why participants made the choices they did, which is often as valuable as the groupings themselves.

In person: Physical index cards on a table. Slower to set up and harder to analyse at scale, but some participants find physical cards more intuitive than screen-based tools, and the think-aloud data from in-person sessions is frequently excellent.

Step 3: Recruit Participants

The Nielsen Norman Group recommends a minimum of 15 participants for open card sorting and 20 for closed card sorting. For most projects, 20 to 30 participants is a practical target. If your user base is divided into meaningfully different segments -- for example, administrators and end users in a B2B product -- recruit separately for each segment and analyse the results independently.

Recruit from your actual user base wherever possible. Card sorting results from the wrong population can be actively misleading.

Step 4: Run the Sessions

For moderated sessions, ask participants to think aloud as they sort. Do not guide them -- your job is to observe, not to explain or validate. Note what they say as much as what they do. Participants who hesitate, move a card back and forth, or create an "I do not know where this goes" pile are giving you important signal.

For unmoderated sessions, include a brief open-ended text field where participants can comment on any cards they found confusing or categories they nearly created but did not.

How to Analyse Card Sorting Results

Similarity Matrix

A similarity matrix is the standard starting point for card sort analysis. It shows, for every pair of cards in your study, what percentage of participants grouped them together. A high agreement score (say, 80% or above) means users consistently see those items as belonging together. A low score means there is no consensus -- which might mean the relationship is unclear, or that different user segments think about it differently.

Most dedicated tools (Optimal Workshop, Maze) generate the similarity matrix automatically. You read it as a heat map: darker cells indicate stronger agreement.

Dendrograms

A dendrogram is a tree diagram that visualises the hierarchical relationships between cards based on the similarity data. Items that cluster tightly were frequently grouped together by participants; items that sit far apart were rarely grouped together.

Dendrograms are useful for spotting natural cluster boundaries -- the points in the hierarchy where you might draw the line between one category and another. They are not a prescription: they show you what users did, and you still need to make design judgements about how to translate that into a navigation structure. But they make the patterns in the data visible in a way that a raw matrix does not.

Qualitative Themes

Do not ignore the category labels participants create in open card sorts. Read through all of them. You will often find that several participants independently coined the same label -- or very similar ones -- for the same cluster of content. Those labels are your navigation copy, handed to you by your users.

Common Mistakes in Card Sorting

Too many cards. Participants lose patience around the 100-card mark and start sorting arbitrarily. If you have more content than that to sort, split the study.

Cards that are too similar. If participants cannot tell the difference between two cards, they will group them together by default, not because they belong together conceptually. Make sure each card represents a genuinely distinct piece of content.

Recruiting the wrong people. Card sorting results from participants who do not represent your users are worse than no data at all, because they create false confidence in a structure that will not work.

Treating the dendrogram as a finished architecture. The dendrogram is evidence, not a design. You still need to apply judgment -- particularly around breadth vs depth trade-offs, top-level category count, and labelling -- to turn the research findings into a navigation structure that works.

Skipping tree testing. Card sorting tells you how users group things; it does not validate that they can find things in the structure you build from those groupings. Always follow up with a tree test before finalising the architecture.

Fitting Card Sorting Into the Wider Research Process

Card sorting is a generative method. It sits alongside user interviews and contextual enquiry in the early, problem-definition phase of a project -- helping you understand users' mental models before you commit to a structure. Once you have a proposed architecture, tree testing validates it. Once you have a designed interface, usability testing checks whether people can navigate it in practice.

Understanding where card sorting fits in the sequence is part of becoming a well-rounded UX researcher. For a broader map of the methods and when to use them, UX research methods is the right place to start. And if you are thinking about how to define and reach your users before running any research at all, user persona templates covers the foundations of building a research-grounded understanding of your audience.

Learn Card Sorting by Doing It

Reading about card sorting is useful. Running a real session -- watching a participant hesitate over a card, hearing them say "I would never call it that," seeing a cluster emerge that completely contradicts your assumptions -- is where the learning actually happens.

Natalia Veretenyk, UX Academy's lead instructor and a designer with experience at Adobe, Google, and Canva, includes card sorting as a hands-on exercise in the Intermediate UX Design course. Students run live sessions with real participants, analyse their own similarity matrices, and present findings to a stakeholder audience -- the same workflow they will use in a professional UX role.

If you are not ready to commit yet, the free UX masterclass is a good place to start. You will get a grounded introduction to the discipline and a clear sense of whether UX research is the direction you want to take your career.