Interpretation boundary
A card sort is evidence for a draft, not approval for a navigation.
Short answer: a card sort observes how participants group and label a selected card set under a specific task. It does not test whether users can find real content in a hierarchy, understand navigation in context, complete tasks, use the interface with assistive technology, or succeed after implementation.
What card sorting can illuminate
- Card pairs often or rarely grouped together under the issued task.
- Where cards land within supplied categories.
- Raw language participants use for created groups.
- Cards with dispersed placements or method-sensitive cluster membership.
- Differences between individual groupings worth discussing qualitatively.
What it does not directly test
| Claim | Why the card sort is insufficient | Possible follow-up |
|---|---|---|
| Users can find item X | Grouping does not require navigating a hierarchy under a realistic task. | Tree testing or task-based usability testing |
| The label is clear | A participant-created or frequently used phrase can remain vague in navigation context. | Comprehension, first-click, or usability testing |
| The hierarchy is accessible | The sort does not test implemented semantics, focus, announcements, zoom, or assistive technology. | Accessibility review and assistive-technology testing |
| The structure scales | The study uses a bounded selected card set, not every future content type or governance change. | Content modelling and governance stress test |
| The result generalizes | Sample, recruitment, task, language, and inclusion choices bound the evidence. | Research design review and additional studies |
| The cluster is correct | Linkage and threshold choices can change cluster membership. | Method-sensitivity reporting and follow-up validation |
Why the candidate structure is separate
The field lab copies one cluster cut into a new editable structure. Renaming, moving, parking, splitting, and merging cards changes only the draft. The source analysis fingerprint remains attached. This prevents a polished draft from being mistaken for the exact computed result.
A stronger decision chain
- Frame the question and recruit for the actual audience.
- Run and review the card sort.
- Compare raw evidence and analysis methods.
- Draft an information structure with content, business, and governance constraints.
- Test findability with realistic tasks.
- Prototype and test navigation in context.
- Review accessibility, content quality, analytics, support evidence, and operational ownership.
- Document what changed and why.
Avoid outcome language the study cannot support. Say “seven included responses grouped A and B together in six cases” rather than “users expect A and B together.” Say “candidate draft” rather than “validated IA.”