Card sorting is a UX research method where participants organize labeled cards into groups that make sense to them — giving you a map of how users actually think about your product, not how you assume they do. There are three flavors: open sorts (participants create their own categories), closed sorts (you define the categories), and hybrid (both). Most teams skip it because it sounds academic.
The ones moving fast on information architecture use it to validate navigation changes in days instead of waiting months for a redesign cycle to confirm what they could have known in a week. You need 20–50 participants to see meaningful patterns, and a single round can settle navigation debates your team has been having for quarters.
Card sorting is one of the oldest UX research methods, and it's still one of the most useful ones you're probably not running.
Here's how it works: you give participants a set of cards — digital or physical — labeled with product features, help articles, pricing tiers, whatever you're trying to organize. You ask them to sort those cards into groups that make sense. That's it.
Why does this matter? Because your information architecture is a hypothesis. You think your users understand your menu structure the way you've arranged it. Card sorting proves whether you're right — or shows exactly where you're wrong. Asana moved their research from quarterly cycles down to 2–3 days by running card sorting inside sprints instead of planning cycles.
Open sort: participants create their own categories from scratch. You get their mental models, their language, their logic. If 40 of 50 people independently create a "Billing & payments" group but your current menu calls it "Invoices & finance," that's actionable data. Best for redesigning navigation completely. Requires 20–30 participants.
Closed sort: you define categories upfront. Participants assign cards to your existing structure. If 60% of participants put an item in a different category than where you have it, your labels need work. Best for validating a structure you've already built. 15–20 participants is usually enough.
Hybrid sort: participants sort into your categories but can create new ones too. Validates and explores simultaneously. Requires 25–40 participants.
Great Question's card sorting tool supports all three formats, with agreement matrices, AI-powered analysis, and results that flow directly into your research repository — no export, no manual triage.
Run card sorting if you're:
Skip card sorting if you're:
One important note: card sorting tells you how users would organize things. It doesn't tell you whether they can find things in the structure you build from it. That's what tree testing is for — and the two methods pair naturally. Run the card sort first to build the structure, then validate it with a tree test before you ship.
1. Define what you're testing and why. Write down the decision this sort informs and what you'll do with the data before you start. If you can't answer "what changes based on the results?", you're not ready.
2. Choose open, closed, or hybrid. Use the framework above. When in doubt, start with a closed sort — it's faster to analyze and answers a tighter question.
3. Select your cards. 30–50 items is the sweet spot. More than 60 creates decision fatigue. Cards should be specific — "Zapier integrations," not "Integrations."
4. Recruit participants. Closed sorts: 15–20. Open sorts: 20–30. Recruit people who actually use the thing you're organizing. You can recruit from your own CRM or pull from a panel — the critical thing is recruiting representative users, not whoever is easiest to reach.
5. Run the sort. Remote is faster and cheaper. Use our card sorting template to get set up quickly. Instructions are simple: "Sort these cards into groups that make sense to you" (open) or "Sort these into the categories provided" (closed). Takes 10–15 minutes per person.
6. Analyze results. For closed sorts, measure agreement -70%+ means users understand it, below 50% means it's confusing. For open sorts, look at similarity matrices and what people named their categories. Standardization score above 0.6 means clear patterns exist.
For closed sorts: cards with 70%+ agreement are clear. Cards below 60% need relabeling or repositioning.
For open sorts: build a similarity matrix showing how often items got grouped together. Use the highest-consensus name for each category — that's the label your users already have in their heads.
Standardization score measures overall agreement (0 to 1). Above 0.6 = clear patterns. Below 0.4 = significant disagreement — something is ambiguous, your participants aren't representative, or the task was poorly framed.
Once you've rebuilt your structure from the data, store your findings in a research repository so future navigation work can build on this round instead of starting from scratch.
Too many cards. Keep it under 60. Aim for 30–45.
Vague card labels. "Integrations" is not a card. "Zapier integrations" or "REST API documentation" are.
Too few participants for quantitative claims. 12 participants gives you useful qualitative direction. It doesn't let you say "70% of users prefer X." Use 20+ for that. If participant recruitment is the bottleneck, build a standing panel rather than recruiting from scratch for every study.
Card sorting without follow-up validation. Always pair with tree testing to confirm the structure actually works for real navigation tasks. The two methods are designed to be run in sequence.
Asana validates navigation changes in 2–3 days using card sorting — down from quarterly planning cycles. One help center with 200 articles ran an open sort with 25 participants, rebuilt from 10 categories to 5 using the exact names participants suggested, and watched search success jump from 30% to 67%.
What's the difference between card sorting and tree testing?
Card sorting asks: how would users organize this? Tree testing asks: can users find things in this structure? Run both for confidence — card sorting builds the IA, tree testing validates it.
How many participants do I really need?
Closed sorts: 15–20. Open sorts: 20–30. Hybrid: 25–40.
How long does it take?
Recruiting: 3–7 days. Running: 10–15 min per person. Analysis: 30 min to 3 hours. Total: 5–14 days.
Can I do this on a tight budget?
Yes. You can run a basic study with a spreadsheet and your existing users. But if you're running card sorting regularly, a purpose-built tool pays off fast — the analysis alone (similarity matrices, standardization scores) takes hours manually vs. minutes with the right tool. See how Great Question's card sorting works and give it a try.
What if my standardization score is below 0.4?
Low agreement means ambiguous labels, unrepresentative participants, or a poorly framed task. Run qualitative follow-ups before redoing the sort.
Start small: 20 participants, 30–40 cards, one sort type (probably closed), one week. You'll have actionable data on navigation your team has been debating for months.
Use our card sorting template to set up your first study, or read our tree testing guide to plan the validation round that follows.
Your navigation should be invisible. Card sorting gets you there faster than iterating in the dark.
