Antoine Pezé
FR

How to run a Card Sorting workshop

TL;DR

Summary

Card sorting is a user research method where participants sort and group content items to reveal how they mentally structure information. It’s the go-to method for building or evaluating an information architecture.

Goals

Design a navigation and content tree that matches users’ mental models rather than the organization’s internal logic.

What is card sorting?

Card sorting is a participatory design technique that consists of presenting content items on individual cards and asking users to organize them into groups that make sense to them.

The method has been around since the 1980s, originally in cognitive psychology, then adapted to interface design in the 1990s by the pioneers of information architecture. The principle is simple: instead of guessing how users think, you ask them to show you.

Card sorting starts from an observation I see regularly in the field: organizations structure their content based on their org chart, not based on how users think. HR puts job openings under “Human Resources.” Communications puts news under “Communications.” But users don’t know your org chart. They look for “a job” or “the latest news.” Card sorting helps build a structure that reflects user logic.

When to use card sorting?

Card sorting fits particularly well in the following situations:

  • Building a new site or application: to build the navigation tree from scratch
  • Redesigning an existing site: to understand why users can’t find what they’re looking for
  • Adding a large amount of content: to figure out where to put it
  • Merging multiple content sources: for example during a company merger or a multi-site redesign
  • Validating a proposed tree: to check that your proposal matches user expectations

The 3 types of card sorting

There are three variants of card sorting. Each addresses a different need at a different point in the project.

Open card sort

In open card sorting, participants receive cards and must group them freely, then name the groups they’ve created. No category is imposed.

When to use it: at the start of a project, when you don’t yet have a tree or you want to start over. Open sorting is exploratory. It surfaces users’ mental categories without influencing them.

What it reveals: the category names participants use are often very different from the names used internally. It’s a goldmine for naming your sections in a way users can understand.

Concrete example: you’re designing a company intranet. You prepare 40 cards with content like “Request time off,” “Find the cafeteria menu,” “Book a meeting room,” “Check my paystub,” etc. Participants sort them freely. You discover they group “Request time off” and “Check my paystub” in a group they call “My space” rather than “Human Resources.”

Closed card sort

In closed card sorting, the categories are predefined by you. Participants only need to place the cards in the existing categories.

When to use it: when you already have a tree and you want to validate it or compare it with an alternative. Closed sorting is evaluative. It tests an existing structure.

What it reveals: cards that are systematically misfiled (placed in an unexpected category) are warning signals. The card may be poorly named, the category may be ambiguous, or the structure may not match users’ mental models.

Hybrid card sort

Hybrid sorting offers predefined categories but allows participants to create new ones if they feel a card doesn’t fit any existing category.

When to use it: when you have a base tree that you want to improve. It’s a good compromise between the openness of open sorting and the structure of closed sorting.

What it reveals: new categories created by participants point to gaps in your tree. If 8 out of 15 participants create a “Help and contact” category, that category is missing from your proposal.

Preparing a card sorting workshop

Define the scope

First, define what you want to sort. The key is to choose a coherent scope and a homogeneous level of detail.

  • A full site: you sort first- and second-level navigation sections
  • A site section: you sort the content of a specific section
  • A product catalog: you sort product categories

Don’t mix levels. If you’re sorting top-level navigation sections (“Our products,” “Contact,” “About”), don’t put sub-sections in the same deck (“Women’s running shoes”). It throws participants off.

Pick the cards

This is the most important preparation step. The quality of your results depends directly on the quality of your cards.

How many cards?

  • For an in-person sort: 30 to 50 cards max. Beyond that, participants get tired and the quality of results drops.
  • For an online sort (digital tool): you can go up to 60 cards, since participants can take breaks.
  • Below 20 cards, the sort risks being too simple to surface interesting insights.

How to phrase the cards?

  • Use concrete and short terms (3 to 5 words max per card)
  • Avoid internal jargon. Write “Request a day off” rather than “Absence request via HRIS”
  • Each card should represent a single concept. If a card contains “and,” it’s probably two cards
  • Test your cards with a colleague to make sure they’re understandable without context

Where to source the content?

  • From the existing site tree
  • From real content (page titles, feature names, product names)
  • From user research results (terms used by users during exploratory interviews)
  • From internal site search queries (the words users type into the search bar)

Recruit participants

How many participants?

  • For an open sort: 15 to 20 participants minimum. Open sorting generates a lot of variability. You need enough participants for patterns to emerge.
  • For a closed sort: 15 participants is enough. Variability is lower since categories are fixed.
  • For a hybrid sort: aim for 15 to 20 participants.

These numbers come from research in information architecture. With fewer than 15 participants, your results will be too variable to draw reliable conclusions.

Who to recruit?

Participants should match your target users. If you’ve built personas, use them to define your recruitment criteria. The key is to recruit people who’ll actually use the site or app, not colleagues from the project team.

Plan for compensation: a nice 20 to 30 euro gift for a 30 to 45 minute sort is reasonable.

Choose between in-person and online

In-person:

  • Lets you observe hesitations, spontaneous comments, changes of mind
  • Richer qualitatively
  • Heavier logistically (venue, scheduling, travel)
  • Ideal for open sorts where observing thinking processes is valuable

Online:

  • Lets you recruit more participants, more easily
  • Results directly analyzable by the tool
  • Less rich qualitatively (no direct observation)
  • Ideal for closed sorts or sorts with a large number of participants

Running an in-person card sort (1h - 1h30)

Materials

  • Cardboard cards or index cards (business card format, around 7x10 cm). Write one item per card, in capital letters, with a black marker
  • A large table or a clear work surface
  • Blank cards and a marker for participants (open and hybrid sorts, to name the groups)
  • A rubber band or paper clip per card group (to collect the results)
  • A camera
  • Something to take notes (a dedicated observer is ideal)

Standard run-through

Introduction (5 minutes)

Greet the participant and put them at ease. Explain the principle:

“Thanks for being here. I’m going to give you a stack of cards. Each card contains a content item. Your mission is to sort them into groups that make sense to you. There’s no right or wrong answer. What I’m interested in is your own logic.”

For an open sort, add: “You’re free to create as many groups as you want. Once your groups are made, I’ll ask you to give them a name.”

For a closed sort, add: “Here are the categories where you should place the cards.” Place the categories on the table. “If a card doesn’t seem to fit any category, set it aside.”

For a hybrid sort, add: “Here are some starter categories. You can place cards in them, but if you feel a category is missing, you can create a new one with these blank cards.”

Specify: “While sorting, please think out loud. Tell me what you’re thinking, your hesitations, your associations of ideas. It really helps me understand your reasoning.”

The sort (20 to 40 minutes)

  1. Hand the cards to the participant, shuffled randomly.
  2. Let them work at their own pace. Don’t interrupt unless they’re stuck on a card for more than 2 minutes.
  3. Take notes on hesitations, comments, and cards that cause trouble. These are valuable qualitative data.
  4. If the participant hesitates between two groups for a card, note it. It’s a sign that the card is ambiguous or that the categories overlap.
  5. If the participant doesn’t understand a card, note it too. That means the label needs rework.

Never guide the participant. If they ask “Can I put this card here?”, always answer: “It’s up to you, there’s no wrong answer.”

The debrief (10 to 15 minutes)

Once the sort is done:

  1. Ask the participant to confirm they’re satisfied with their groups. “Are you happy with everything? Want to change anything?”
  2. For an open or hybrid sort, ask them to name each group. “If this group were a menu section, what would you call it?”
  3. Ask questions about edge cases: “I noticed you hesitated on this card. Can you tell me more?”
  4. Ask if any cards were missing: “Were there things you expected to find that weren’t in the deck?”
  5. Photograph the result before putting everything away.
  6. Note the groups and their contents (or number the cards beforehand to make data entry easier).

Between two participants (5 minutes)

Shuffle the cards again. The order in which the cards are presented influences the sort. By shuffling them, you reduce that bias.

Running an online card sort

  • Optimal Workshop: the reference. Offers open, closed, and hybrid sorts, with built-in analysis tools (dendrograms, similarity matrices). The tool is in English, but participants don’t need to understand English if you write your cards in their language.
  • Maze: more recent, modern interface, integrated with Figma. A good option if you already use Maze for user testing.
  • UsabiliTEST: a simpler, cheaper alternative.

Setup

  1. Create your study in the chosen tool.
  2. Enter your cards (and your categories if closed or hybrid sort).
  3. Write a clear introduction for participants.
  4. Test the link with a colleague to make sure everything works.
  5. Send the link to your participants by email with a participation deadline (usually 1 week).

Tip: add an open question at the end of the online sort to gather comments. “Were there cards you found hard to classify? Why?” This compensates for the absence of direct observation.

Analyzing the results

Analysis is the key moment. You move from raw data to concrete, actionable insights.

The Similarity Matrix

The similarity matrix shows how often two cards were placed in the same group. It’s a two-way table where each cell shows the percentage of participants who associated the two cards.

  • A high score (70% and above) means the two cards are strongly associated in users’ minds. They probably belong in the same section.
  • A low score (less than 30%) means the two cards aren’t related. Separating them in the navigation is coherent.
  • Intermediate scores are the most interesting: they reveal ambiguities to dig into.

Optimal Workshop generates this matrix automatically. If you run the sort in person, you’ll have to build it manually in a spreadsheet.

The dendrogram

The dendrogram is a tree showing how the cards naturally group. The most often associated cards are connected at the bottom of the tree, then the groups gradually merge upward.

To read a dendrogram:

  1. Choose a cutoff threshold (for example 60%). All branches below that threshold form a group.
  2. The higher the threshold, the more groups you get (fine structure). The lower it is, the fewer you have (coarse structure).
  3. Try several thresholds to see which structure emerges.

The dendrogram is particularly useful for open sorts: it surfaces the natural structure without imposing categories.

Group name analysis

In an open sort, the names participants give their groups are a precious source of information. Group similar names together and identify the most frequent terms. They’re serious candidates for your section names.

For example, if 12 out of 20 participants name a group “My account” or “My space,” that’s a strong signal. If 3 say “Settings” and 2 say “Configuration,” those terms are less consensual.

Problematic cards

Identify cards that don’t have a stable group: those classified into 4 or 5 different groups depending on the participant. These cards are ambiguous. Either their label isn’t clear, or they cover a concept that overlaps several categories. You may need to split them into two cards, rename them, or place them in several site sections (with cross-links).

From analysis to tree

Once the analysis is done, build a tree proposal:

  1. Start from the groups that emerge naturally (dendrogram or similarity matrix).
  2. Name the sections with the terms most used by participants.
  3. Place ambiguous cards in the category where they’re most often classified, and plan shortcuts or cross-links from the other categories.
  4. Validate this proposal with a closed sort if you have doubts.

Common mistakes

  1. Poorly worded cards. If your cards contain jargon or are ambiguous, your results will be unusable. Always test your cards with 2 or 3 people before launching the sort.

  2. Too few participants. With 5 participants, you won’t have reliable patterns. Card sorting is a method that needs sufficient volume to be exploitable. Aim for at least 15 participants.

  3. Guiding participants. Card sorting rests on the participant’s own logic. If you help them, you’re measuring your own mental model, not theirs.

  4. Confusing card sorting with tree testing. Card sorting is for building an architecture. Tree testing is for evaluating an existing architecture by asking participants to find content in a navigation tree. The two methods are complementary: do a card sort first to build, then a tree test to validate.

  5. Not analyzing qualitative data. Hesitations, comments, cards set aside are just as important as the quantitative data. Don’t settle for the similarity matrix.

  6. Taking the results as absolute truth. Card sorting gives you trends, not a turnkey tree. You then need to cross-check those results with business, technical, and editorial constraints to build a viable architecture.

  7. Mixing levels of granularity. If you put cards like “Contact us” (top-level section) and “Edit my email address” (third-level feature) in the same deck, participants will be lost and your results unusable.

Standard format for a complete workshop

Here’s a standard schedule for an in-person card sort with 15 participants, over 2 days:

Day 1: Preparation (2h)

  • Selecting and writing the cards (1h)
  • Testing with 2 colleagues and adjusting (30 min)
  • Producing the physical cards and prepping the materials (30 min)

Day 2: Facilitation (full day)

  • You have 45 minutes per participant, with 15 minutes of transition between each session
  • Plan for 5 participants in the morning and 5 in the afternoon if you’re alone
  • If you’re two facilitators, you can double the pace

Day 3: Analysis (half day)

  • Entering results into a spreadsheet or analysis tool (1h)
  • Analyzing the similarity matrix and dendrogram (1h)
  • Writing conclusions and proposed tree (1h)

By the end of this process, you should have:

  • A similarity matrix showing the strong and weak associations between content items
  • A dendrogram surfacing the natural groups
  • A proposed tree to validate
  • A list of problematic cards to rework

Going further

Sources

Want to go further?

I offer individual coaching to dig deeper and apply these topics to your context.

Book a session