Card Sorting: Methods and Best Practices

Card sorting is a structured user research technique used to inform the design of navigation systems, category structures, and labeling within information architectures. The method generates empirical data about how target users mentally group and name content, reducing reliance on designer assumptions when building taxonomies or site hierarchies. It occupies a defined role within the broader information architecture process, typically deployed before structure is finalized and after initial content audits have identified the scope of items requiring organization.

Definition and scope

Card sorting is a research method in which participants organize labeled items — written on physical cards or presented through software — into groups that make sense to them, and optionally assign names to those groups. The technique belongs to the participatory design tradition formalized in human-computer interaction research and is documented as a standard elicitation method in usability engineering literature, including Jakob Nielsen and colleagues' work published through the Nielsen Norman Group.

Within information architecture, card sorting produces data used to build or validate taxonomy structures, controlled vocabularies, and labeling systems. Its scope extends across digital products — from enterprise intranets to e-commerce catalogs — wherever the organization of content items affects user findability. The method does not test interface design or visual presentation; it isolates the structural and categorical logic users apply to content independently of interface cues.

Card sorting produces two primary data types: grouping data (which items users place together) and label data (what names users assign to those groups). Quantitative analysis of grouping data typically uses cluster analysis or similarity matrices, while label data informs the vocabulary used in navigation menus and category headings.

How it works

Card sorting follows a discrete sequence of phases regardless of method variant:

  1. Item selection — Content items are identified and written as discrete topic labels, typically 30 to 100 items per session. Exceeding 100 items introduces participant fatigue and degrades data quality, a threshold documented in usability research by Donna Spencer in Card Sorting: Designing Usable Categories (Rosenfeld Media).
  2. Participant recruitment — Representative users from the target audience are recruited. A minimum of 15 participants is the commonly cited threshold for open card sorting before patterns stabilize, based on research published by Jakob Nielsen and Kara Pernice (Nielsen Norman Group).
  3. Session execution — Participants sort cards into groups using their own logic, working independently to avoid social influence on categorization decisions.
  4. Data collection — Groupings and labels are recorded, either through observation (physical sessions) or automated logging (digital tools such as Optimal Workshop's OptimalSort or Maze).
  5. Analysis — A similarity matrix tabulates how frequently each pair of items was placed in the same group across all participants. Cluster analysis then identifies stable groupings. Agreement scores quantify the consistency of categorization across participants.
  6. Synthesis — Findings are mapped to candidate site maps and hierarchies, with high-agreement clusters informing primary categories and low-agreement items flagged for further investigation or alternative placement.

Common scenarios

Card sorting applies across several distinct contexts within professional information architecture practice:

Open card sorting — Participants create their own group names. Used when the category structure is undefined or being built from scratch. Generates vocabulary data alongside structural data. Appropriate for new product launches, content migrations, or intranet design where existing navigation has been identified as problematic.

Closed card sorting — Participants sort items into predefined categories. Used to validate an existing structure or test whether proposed categories accommodate a content inventory. Appropriate when major navigation labels are fixed but sub-category placement is uncertain. Frequently applied in e-commerce information architecture where top-level categories are business-driven but subcategory placement drives findability.

Hybrid card sorting — Participants sort into predefined categories but may create additional groups for items that do not fit. Useful when a partial structure exists and designers need to identify gaps without reopening the entire taxonomy.

Remote versus in-person — Remote unmoderated sessions scale participant recruitment but eliminate the ability to ask follow-up questions. Moderated sessions — remote or in-person — allow probing of participant reasoning, generating qualitative rationale alongside quantitative grouping data. The user research for IA discipline treats this tradeoff as a standard methodological decision.

Decision boundaries

Card sorting is not a universal solution for structural problems, and its deployment has defined limits.

The method measures mental models, not task success. A card sort revealing that users group "account settings" with "billing" does not confirm that a navigation structure combining them will support task completion. Tree testing is the complementary validation method used after card sort findings are implemented — it tests whether a proposed hierarchy enables users to locate specific items under realistic task conditions. The two methods occupy adjacent but non-overlapping roles: card sorting informs structure creation; tree testing validates structure performance.

Card sorting produces unreliable data when item labels are ambiguous, technical, or unfamiliar to participants. If participants cannot interpret a card label within 5 to 10 seconds, the resulting grouping reflects label confusion rather than genuine categorical logic. Item preparation — ensuring labels match the vocabulary of target users — is a prerequisite for valid results.

The method is also constrained by content scale. Enterprise systems with thousands of content types cannot be addressed through a single card sort. IA for enterprise systems practice typically applies card sorting to defined subsections of a content inventory rather than the full corpus, with findings extrapolated within bounded content domains.

Card sorting aligns with mental model research as documented in the broader mental models in information architecture literature, where understanding user conceptual schemas is treated as foundational to structural design decisions.

References

Read Next

Navigation Design in Information Architecture This page covers navigation typologies, the mechanics that govern their effectiveness, the professional standards that inform... The Information Architecture Design Process Step by Step The process spans discovery through governance, touching research methods, structural deliverables, and iterative testing. Taxonomy: Organizing Information for Clarity This page covers the definition, mechanical structure, classification logic, and practical tensions of taxonomy as applied in...