A card sort asks participants to group representative content items. Open sorting explores groupings and labels; closed sorting examines placement into proposed categories; hybrid sorting combines both. Plan around an architecture question, prepare understandable cards, recruit relevant participants, and analyze agreements, disagreements, rationales, and segment differences. Results inform structure but do not produce a complete navigation system.
Who this is for: UX researchers and information architects studying how relevant participants group and label a digital content collection.
- Select open, closed, or hybrid sorting according to the structural uncertainty the study must address.
- Use clear, representative cards that do not reveal categories through wording, numbering, or unequal detail.
- Interpret similarity and placement with participant explanations, task context, content constraints, and study limits.
Choose the study question
An open sort can reveal grouping ideas and vocabulary when structure is unsettled. A closed sort can show where a proposed set of categories creates hesitation or disagreement. A hybrid sort allows supplied categories plus new ones. State which decision the selected format can influence before preparing cards.
Card sorting studies conceptual relationships, not complete findability. It removes navigation depth, visual hierarchy, search, permissions, and real task pressure. Plan later tree testing or usability testing for those questions. Do not use a card sort to settle business ownership disputes that participants cannot observe.
Prepare cards and participants
Sample the content domain deliberately, including common items, consequential edge cases, and items that expose structural tensions. Write each card at a consistent level of detail in language participants understand. Remove codes and wording patterns that disclose the team's current category or preferred answer.
Recruit people with relevant goals and domain knowledge, noting meaningful role or experience differences. The number of participants should reflect variation, study format, analysis needs, and diminishing new information, not a universal rule. Record the sample and avoid converting placement rates into claims about the entire customer base.
Run consistent sessions
Explain that there is no test of the participant and that cards may be ambiguous. In open studies, ask for group names after sorting so labels reflect the participant's structure. Allow uncertain piles and changes. Avoid explaining card meanings unless the study protocol defines how clarification will be handled.
Observe which cards move repeatedly, which groups overlap, and what rationale participants give. A moderated session provides richer explanation; an unmoderated study may support more distributed participation but needs especially clear instructions. Pilot the tool, card set, timing, accessibility, and export before launch.
Analyze without overclaiming
Review common pairings, category agreement, labels, outliers, hesitation, and comments. Similarity matrices and dendrograms can reveal patterns but depend on method and interpretation. Compare relevant participant groups where the product genuinely serves different tasks, while keeping small subgroup results descriptive.
Translate patterns into architecture hypotheses, unresolved items, and label options. Reconcile them with content ownership, metadata, policy, and known tasks. Then test the proposed hierarchy through tree testing. Preserve raw exports and analysis decisions so teammates can understand how the recommendation arose.
Sort a hypothetical support center
Hypothetical scenario: a software company is reorganizing help articles for account administrators.
- Choose an open sort because the team's current categories follow internal departments rather than administrator goals.
- Prepare hypothetical article cards with consistent titles covering setup, access, billing, recovery, and reporting.
- Recruit administrators with varied recent tasks, pilot the tool, and capture grouping rationale and uncertain cards.
- Use common patterns to draft categories, preserve disputed items as cross-link candidates, then plan tree testing.
Card sort protocol
Use this record from study choice through architecture handoff.
- Architecture decision, uncertainty, format choice, excluded questions, and planned follow-up method.
- Content universe, sampling logic, card wording, detail level, edge cases, randomization, and pilot findings.
- Participant criteria, relevant variation, recruitment, consent, accessibility, compensation, and sample limits.
- Instructions, facilitation rules, uncertain-card handling, tool, observation notes, and exported evidence.
- Pairings, categories, labels, disagreements, segment context, hypotheses, constraints, and tree-test plan.
Common mistakes
- Including current category names in card titles and interpreting the resulting grouping as independent confirmation.
- Treating a clustering visualization as the finished site structure without participant rationale or task testing.
- Mixing cards from unrelated levels, such as broad departments and individual help articles, without a clear purpose.
Try one
Participants place one article in several different groups. Does that mean the study failed?
No. A strong analysis examines why the item crosses contexts, whether its wording is ambiguous, and whether the content genuinely supports several tasks. The architecture might use clearer naming, metadata, cross-links, or duplicate entry points to one maintained source. The team should test proposed retrieval paths rather than forcing agreement that the evidence does not support.
Sources
- Nielsen Norman Group card sorting guideGuidance on using open and closed card sorting to study how participants group content.
- Nielsen Norman Group card sorting analysisGuidance on planning card-sort participation and interpreting patterns with appropriate limits.