Skip to content
Tutorials Center
My learning

Topics

Lesson 16 of 36 ยท Information Architecture

Card Sorting

Learn card sorting for information architecture: open, closed and hybrid sorts, how to run one, how to read similarity results, and a runnable stand-in that counts how often cards are grouped.

How should an online shop group its products? Which headings belong in the help menu? The team will have opinions, but the people who use the site may group things differently. Card sorting is a method where participants organize labeled cards into groups that make sense to them. It shows you the users' mental model of your content (see Mental Models).

How it works

  1. Write each content item or feature on a card (physical index cards, sticky notes, or an online tool).
  2. Ask participants to sort the cards into piles.
  3. Optionally ask them to name the piles.
  4. Combine results from many participants and look for agreement.

Usually 15 to 30 participants give stable patterns for a quantitative analysis; fewer can still yield useful qualitative ideas, especially if you ask people to think aloud.

Types of card sort

  • Open sort: participants create their own groups and name them. Use it when designing a new structure, to learn how people think and which words they use.
  • Closed sort: you give the categories; participants place cards into them. Use it to test whether an existing or proposed structure works.
  • Hybrid sort: participants use your categories but may create new ones. A good middle ground.

Sorts can be moderated (you watch, listen and ask why) or unmoderated (online, many people, less insight into reasoning).

flowchart LR
  A["Have content, no structure"] --> B["Open sort"]
  B --> C["Draft structure"]
  C --> D["Closed sort or tree test"]
  D --> E["Refine"]

Open sorts generate structures; closed sorts and tree tests check them.

Preparing good cards

  • Use 30 to 60 cards; fewer does not reveal patterns, more tires people.
  • Use real items, in the users' language, not internal jargon.
  • Keep cards at the same level of detail. Mixing "Shoes" and "Men's running shoes size 44" confuses the grouping.
  • Avoid words that repeat and tempt matching ("Account settings", "Account help"): people may group by the shared word and not by meaning.
  • Include a few cards you expect to be hard, such as "Gift cards".

Analyzing results

Count, for every pair of cards, how many participants put them in the same group. Show the counts in a similarity matrix. High counts mean cards belong together. Tools then draw a dendrogram, a tree showing which cards cluster at which level. Look for:

  • Strong clusters where most people agree.
  • Cards that nobody agrees on: these may need a new label or two homes (cross-links).
  • Group names people used, which are label candidates.

Also read the comments: the reasons behind groups matter more than the arithmetic.

Choose the right sort

Open People make and name groups
Closed People fit cards to your groups
1 / 2

A stand-in: counting co-occurrence

This stand-in takes three participants' sorts and counts how often two cards were grouped together. Real tools do the same with dozens of people and then cluster the result.

JS Similarity counts (stand-in)

Press Run (or Ctrl + Enter) to see the console output here.

No output.

Here "Running shoes + Yoga mat" and "Lamp + Pillow" are grouped by everyone, while "Gift card" is placed differently by each person: that card needs a decision, perhaps a cross-link from several sections.

From results to a structure

  1. Draft categories from the strongest clusters, using participants' own group names.
  2. Decide where disputed cards live, and add cross-links for items that belong in more than one place.
  3. Draw the draft site map.
  4. Validate with a closed sort or tree test with new participants.

Limits of card sorting

  • It shows how people group, not how they find. A good grouping can still fail in a real task, which is why you also test navigation.
  • It depends on the cards you chose. Bad cards give misleading results.
  • Participants might sort by superficial words.
  • Results may differ by user group; recruit each important segment.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. In an open card sort, who creates the groups?

    Open sorts let people create and name their own groups.
  2. 2. Which type of sort checks whether a proposed structure works?

    Closed sorts give the categories and see where people place cards.
  3. 3. What does a similarity matrix show?

    High counts show cards that users think belong together.
  4. 4. What does it suggest when everyone sorts a card differently?

    Disagreement points to ambiguity; consider cross-links or relabeling.

Summary

  • Card sorting reveals how users group and name content.
  • Use open sorts to discover, closed sorts to validate, hybrid in between.
  • Use 30 to 60 clear cards in the users' language.
  • Analyze by pair counts and clusters, and read participants' reasons.
  • Validate the result with a tree test or closed sort.

Next: Site Maps and Navigation Patterns