Card sorting information architecture research is how you find out where users expect things to be, before you commit to a structure. Here’s a rule of thumb worth pinning above your desk: if your project involves more than 25 pieces of information that need to be arranged into a menu, screen, or navigation system, run a card sorting study before you commit to a structure.
The outcome of a card sorting exercise is a data-backed categorization and prioritization of information for your website or app. In other words, card sorting is how you get your information architecture from real users instead of internal guesswork. And it isn’t limited to top-level mega-menus. Card sorting helps you group different types of content into sections, decide what belongs inside each section, and even prioritize the information within those groups, whether that’s a SaaS settings panel, a hospital patient portal, or an e-commerce filter system. With modern remote card sorting tools, it’s also one of the least expensive research methods you can run with real users.
Card Sorting Information Architecture: Why It Matters
If you’re on the team building the product, your mind is already conditioned to group information in a certain way. You know the org chart, the database schema, the internal jargon. Your users know none of this. They arrive with limited domain knowledge and a completely different set of expectations about where things should live.
This gap between the team’s mental model and the user’s mental model is where navigation fails. Users don’t file a bug report when they can’t find something. They leave. That’s why information architecture decisions should involve users during the design phase of a new product, or before a redesign of an existing interface, not after the structure has been locked in.
Card sorting closes that gap. As Nielsen Norman Group puts it, the method uncovers users’ mental models of your information architecture. Instead of guessing how users think, you watch them organize your content themselves and let the patterns emerge from the data.
A Real-Life Card Sorting Example: The Two-Hour Unicorn
One of our colleagues wanted to buy a toy unicorn as a gift for his friend’s six-year-old. His requirements were simple: taller than 24 inches, from a trusted seller, with a few good reviews.
Browsing Shopee, one of Southeast Asia’s most popular retail apps, the search for “unicorn toy” surfaced the relevant category, Soft Toys, which contained over 6,000 items. The retailer had built filters, but the ones that mattered for this purchase were missing. There was no way to filter by size, the single most important attribute for this buyer. What should have been a five-minute purchase took two hours of scrolling.
This is an information organization failure, not a content failure. The products existed; the structure around them didn’t match how a real shopper thinks. A card sorting study with actual toy buyers would have surfaced “size” as a category users expect, long before 6,000 products were dumped into a single bucket.
Card Sorting Methods: Open, Closed, and Hybrid
Every card sort involves a set of cards (granular pieces of information) that participants are asked to group. What varies is how much structure you give them. Depending on your product domain and where you are in the project, you’ll choose between open, closed, and hybrid sorts. (For a deeper breakdown of each type with best practices, see our complete guide to card sorting.)
Open Card Sorting Method
In an open card sort, there are no predefined categories. Participants group the cards however makes sense to them, then name each group in their own words.
This is the exploratory workhorse. It reveals users’ mental models before you’ve committed to any structure, and the category names participants invent tell you the exact terminology your audience uses, which often differs sharply from your internal labels. The trade-off: results are less consistent across participants and take more effort to analyze, which is why open sorts work best in early discovery stages.
Closed Card Sorting Method
In a closed sort, the categories are predefined and fixed. Participants place cards into your existing structure until the cards run out.
Use this to validate a structure you already have. For example, you can test whether users can correctly place new content into your current navigation. Be aware of the built-in bias: because participants are limited to the categories you provide, a closed sort can’t tell you whether those categories were right in the first place. Cards get forced into the least-bad option, which doesn’t necessarily reflect genuine preference.
Hybrid Card Sorting Method
A hybrid sort gives participants predefined categories and the freedom to create their own when a card doesn’t fit anywhere.
For the information architecture of new products and for redesigns, this is the most popular choice among UX designers and researchers: you get the analytical convenience of a mostly-fixed structure, plus a clear signal whenever your proposed categories fail. If participants keep inventing the same new category, that’s your IA telling you something is missing.
Open vs. Closed Card Sorting: Which Should You Choose?
Use an open card sort during early discovery, when you want to learn users’ mental models and terminology without imposing any structure. Use a closed sort when you already have a structure and need to validate it. When you’re between those two points (most redesigns live here), the hybrid method gives you the best of both.
Card Sorting vs. Tree Testing: What’s the Difference?
Card sorting tells you how users would build your structure. Tree testing tells you whether users can find things inside the structure you built. The strongest IA workflows use both: run a card sort to generate the structure, then run a tree test to validate that people can navigate it. If you’re new to the second half, our tree testing guide walks through the full process.
How to Conduct a Card Sorting Exercise
Card sorting exists in physical and digital forms, and digital studies can be moderated or unmoderated. The right choice depends on your time, budget, and factors like information sensitivity.
Physical vs. Remote Card Sorting
Physical (in-person) card sorting gives you the definitive advantage of direct interaction: you can probe why a participant hesitated over a card. But the logistics are heavy: booking a venue, recruiting and scheduling participants, facilitating skillfully, then manually digitizing piles of index cards before you can analyze anything. Unless the stakes are high and you have both the time and an experienced facilitator, a physical workshop is rarely worth the overhead today.
Moderated vs. Unmoderated Card Sorting
Moderated remote sorting keeps the conversational depth while removing the venue. The researcher attends each session and can ask questions as participants act. It’s time-intensive, since you’re present for every session, so it works best piggybacked onto ongoing research where you already have participants’ time.
Unmoderated remote sorting is where card sorting scales. Participants complete the exercise on their own time, the tool records every move, and results are consolidated automatically. This is the right approach when you need feedback from a wide variety of users in larger numbers, which, for statistical confidence in your IA decisions, is most of the time.
Card Sorting Process: 6 Steps
- Define the objective. What decision will this study inform? “Restructure the account settings area” is an objective; “learn about our users” is not.
- Select and write the cards. Aim for roughly 30-60 cards. Fewer than 20 rarely produces meaningful groupings; more than 60 causes fatigue. Write each card in plain language your audience actually uses.
- Choose the sort type. Open for discovery, closed for validation, hybrid for most new-structure and redesign work.
- Recruit the right participants. Screen out anyone who doesn’t match your target audience, and use survey questions to establish each participant’s context and domain knowledge. If you don’t have a user pool of your own, a participant recruitment panel can source matched respondents.
- Pilot the study. Run it with 2-3 colleagues outside the project team first. You’ll catch ambiguous card labels and unclear instructions before they contaminate real data.
- Launch, then analyze. Look at the similarity matrix (which cards were grouped together most often), the category names participants created, and the agreement level across participants. Strong, repeated patterns become your structure; weak agreement flags content that needs rethinking or renaming.
Card Sorting Sample Size: How Many Participants Do You Need?
Card sorting is a quantitative-leaning method: you’re looking for patterns strong enough to base structural decisions on. The right sample size matters: running a sort with 8-10 respondents usually won’t produce distinguishable patterns, and ambiguous results make decision-making harder, not easier.
Classic research on the topic (Tullis & Wood’s widely cited study) found that agreement scores stabilize somewhere in the 20-30 participant range, and Nielsen Norman Group’s analysis of card sorting sample sizes generally recommends at least 15 for an open sort. Our practical recommendation: aim for a minimum of 30 participants per audience segment for unmoderated studies. Remote unmoderated tools make this affordable in a way physical sessions never were.
Tips for an Effective Card Sorting Study
Align internally first. Before the study goes live, get stakeholder agreement on the objective, the card list, any predefined category names, and the project timeline. Disagreements are much cheaper to resolve before data collection than after.
Ask participants to prioritize, not just group. Once the cards are sorted, ask participants to arrange the categories themselves in order of importance. You get an information hierarchy and a priority ranking from the same session.
Set the context explicitly. In a remote study, your instructions do the facilitator’s job. State the scenario clearly (“Imagine you’re managing your company’s billing…”) so every participant sorts from the same frame of mind.
Use screener and survey questions. Filter out mismatched participants before the sort, and capture their background so you can segment results afterward.
Common Card Sorting Mistakes to Avoid
Card sorting is a mature, well-documented method, but the same mistakes keep appearing:
Too few participants. As above: small samples produce noise, and noise gets misread as insight.
Too many cards or categories. Dumping every content item into the study causes participant fatigue, and fatigued participants start sorting without thinking. Sample representatively instead of exhaustively.
Jargon on the cards. People recognize what they already know. If your cards use internal terminology, participants will sort your words, not their concepts. Pull card labels from user interviews, search logs, and support tickets.
Ignoring local language. If you’re running studies across markets like Vietnam, Japan, Indonesia, or Thailand, test in the local language, and have translations reviewed by a native-speaking expert, not just machine translation. Mental models are shaped by language and culture, so a structure validated in English may fail elsewhere. (More on this in our guide to international UX research.)
Running it too late. Card sorting is a proactive method. Running it close to launch means discovering structural problems at the exact moment they’re most expensive to fix.
How to Analyze Card Sorting Results
Unmoderated tools generate a lot of data: similarity matrices, dendrograms, category-agreement scores. (For a sense of how much manual work modern tools save, see digital.gov’s write-up of running card sorting with spreadsheets alone.) Focus on three questions:
- Which cards cluster together consistently? High-agreement clusters are your sections.
- What did participants call those clusters? Their labels become your navigation language.
- Which cards scattered everywhere? Low-agreement cards signal content that’s ambiguous, mislabeled, or trying to serve two purposes at once.
Modern platforms increasingly automate the pattern-finding step; if you’re curious how far this has come, see our piece on using AI in UX research. And if you’re comparing platforms for your first study, we’ve reviewed the leading options in our roundup of the best card sorting tools.
Conclusion: Organize Information Around Users’ Mental Models
Card sorting information architecture work is the most direct way for information architects, designers, and researchers to define or refine the structure of an app or website around users’ actual mental models instead of the team’s assumptions. Run it early, when structure is still cheap to change. Use open or hybrid sorts to discover the structure, closed sorts and tree tests to validate it, and recruit enough participants for the patterns to be trustworthy. With remote, unmoderated tools handling the logistics and consolidation, there’s very little standing between you and navigation your users actually understand.
Design what suits your users’ mental model, right from the start of product creation. Run your first card sorting study with UXArmy.
Frequently Asked Questions
How does card sorting improve information architecture
Card sorting reveals how users naturally group and label content, exposing the gap between your team’s internal logic and users’ mental models. Applying those patterns to your navigation, menus, and content sections produces an information architecture users can navigate intuitively without stopping to think about where things “should” be.
How many cards should be used in a card sort?
Aim for 30–60 cards. Fewer than 20 rarely produces meaningful groupings, while more than 60 causes participant fatigue and careless sorting. If you have hundreds of content items, select a representative sample rather than testing everything.
Is card sorting qualitative or quantitative?
Both. The similarity matrices and agreement scores are quantitative, showing how often participants grouped items together. The category names participants create and their comments are qualitative, revealing the language and reasoning behind their choices. Strong studies use the numbers to find patterns and the labels to name them.
When in the design process should you do card sorting?
As early as possible during the design phase of a new product or before a redesign begins. Card sorting is a proactive method; running it close to launch means discovering structural problems when they’re most expensive to fix.
How many participants do you need for card sorting?
Plan for 15–30 participants per audience segment. Research by Tullis and Wood shows agreement scores stabilize in this range, and beyond 30 participants results rarely become substantially clearer. For unmoderated remote studies, we recommend a minimum of 30 to ensure distinguishable patterns.
What is the difference between card sorting and tree testing?
Card sorting helps you build a structure participants group content into categories that make sense to them. Tree testing validates that structure participants try to find items inside a proposed hierarchy. Use card sorting early to generate your information architecture, then tree testing to confirm users can navigate it.
When should you use open vs. closed card sorting?
Use an open card sort during early discovery, when you want to learn users’ mental models and terminology without imposing structure. Use a closed sort to validate an existing structure for example, checking whether new content fits your current navigation. For redesigns, hybrid sorting combines the strengths of both.
How do you analyze card sorting results?
Focus on three outputs: the similarity matrix (which cards were consistently grouped together), the category labels participants created (your users’ natural terminology), and low-agreement cards (content that’s ambiguous or mislabeled). High-agreement clusters become your sections; scattered cards signal content that needs renaming or restructuring.
