Card Sorting — Definition & Guide

Card sorting is a UX research method where participants organize content items — written on individual cards — into groups that make sense to them. It reveals how real users categorize information, helping teams build an information architecture that matches users’ mental models instead of internal assumptions.

Why It Matters

One of the hardest problems in product design is deciding how to organize content. Where should “Billing” live — under “Account” or “Settings”? Should “Templates” be part of “Projects” or stand alone? Teams debate these questions endlessly because everyone’s intuition is different. Card sorting replaces opinion with evidence.

The method works because it externalizes how users think about your content. Instead of guessing how users would group items, you watch them do it. The results often reveal surprising patterns — categories your team never considered, labels that users prefer over your internal terminology, and mental model differences between user segments.

Card sorting is also one of the most accessible research methods available. It requires no functioning product, no development time, and no specialized tools. You can run a card sort with index cards on a table or remotely with free online tools. It fits naturally into the early stages of any project where navigation or content structure is being designed or redesigned.

How It Works / Types

Card sorting comes in three main variations, each answering different questions about your information architecture.

Open Card Sorting

Participants receive unlabeled cards with content items and create their own groups and group names. There are no predefined categories — participants organize freely based on their own logic.

Best for: early discovery. When you’re starting from scratch or rethinking your entire IA, open card sorting reveals how users naturally cluster content and what labels they use. It generates ideas you wouldn’t have considered.

What you learn: how many groups users expect, what those groups should be called, and which content items users see as related.

Closed Card Sorting

Participants receive the same content cards but sort them into predefined categories that you provide. The categories already exist — you’re testing whether users can correctly place content within them.

Best for: validation. When you’ve drafted a navigation structure and want to test whether users can find items where you’ve put them. Closed sorting tells you whether your categories make sense and whether specific items are grouped where users expect them.

What you learn: which categories work well, which are confusing, and which content items are difficult to classify — items that get sorted into different categories by different users signal labeling or structural problems.

Hybrid Card Sorting

A combination: participants sort cards into predefined categories but can also create new groups if none of the existing ones fit. This balances the structure of closed sorting with the discovery potential of open sorting.

Best for: iterating on an existing IA. When your structure is mostly set but you suspect some areas need adjustment, hybrid sorting reveals both what works and where users need categories you haven’t provided.

Real-World Example

Imagine a university redesigning its website. The current navigation reflects the university’s organizational chart: “Office of the Registrar,” “Student Financial Services,” “Division of Student Affairs.” But prospective students don’t think in departments — they think in tasks: “How do I apply?”, “How much does it cost?”, “Where will I live?”

An open card sort with 30 content items and 15 prospective students reveals three dominant user-created groups: “Applying” (application, deadlines, requirements, transcripts), “Costs & Aid” (tuition, scholarships, payment plans, financial aid), and “Campus Life” (housing, dining, clubs, events). None of these match the university’s departmental structure.

The redesigned navigation uses these user-generated categories. Support ticket volume drops because students can find information without calling the admissions office. The content is identical — only the structure changed.

How to Apply

Select 30-60 content items. Too few cards don’t reveal enough structure. Too many overwhelm participants. Aim for 30-60 items that represent your most important content pages, features, or menu items. Write each item on a card with a short, neutral label — avoid jargon that would bias grouping.
Recruit 15-20 participants who match your target users. Card sorting results are most useful when participants represent your actual user personas. A card sort for a developer tool needs developers, not marketers. Nielsen Norman Group recommends at least 15 participants for reliable quantitative patterns.
Start with open sorting, then validate with closed. Use open card sorting first to discover user-generated categories. Analyze the results to draft your IA structure. Then run a closed sort with a different set of participants to validate whether users can correctly place items within your proposed categories.
Look for agreement and disagreement. Cards that every participant groups the same way are easy wins — your IA should reflect that consensus. Cards that participants place in different categories reveal problems: ambiguous labels, overlapping categories, or content that doesn’t fit neatly anywhere. Focus your design attention on these disagreements.
Combine card sorting with tree testing. Card sorting generates the structure; tree testing evaluates it. After building your IA from card sort data, run a tree test — give users tasks and ask them to navigate a text-only hierarchy to find items. This validates that the structure works for task completion, not just categorization.

Common Mistakes

Using biased card labels. If a card says “Account Settings & Preferences,” participants will group it with other cards containing “Account” or “Settings” based on keyword matching rather than conceptual fit. Use neutral, descriptive labels. “Change your password” is better than “Account Security Settings” because it describes the task without hinting at a category.

Running only closed sorts. Closed sorting validates your categories but doesn’t reveal whether those categories are the right ones. If the categories themselves are wrong, a closed sort will show decent results because users are forced to choose from what you’ve given them. Always run at least one open sort to discover user-generated structures.

Ignoring outlier groupings. When 14 out of 15 participants group a card the same way, the fifteenth isn’t wrong — they represent a real user perspective. Consistent outlier patterns across multiple cards might indicate a distinct user segment with a different mental model. Investigate these patterns instead of dismissing them as noise.