Card Sorting for Information Architecture: A Guide

Card sorting helps a team understand how people group concepts and what language they use for those groups. It is useful when a navigation structure reflects internal departments more than customers' expectations. It does not, by itself, prove that someone can find a specific destination in the finished product.

Choose the question and sorting format

Use an open card sort when you need to explore participants' own groupings and names. Use a closed sort when you already have candidate categories and want to understand where items fit. A hybrid sort allows predefined groups plus new ones, which can expose gaps without starting from a blank canvas.

Use the word counter to compare card labels for unnecessary length. Read the labels together to spot inconsistent detail: “Invoices” and “Everything related to managing your organization's employees” operate at different levels. Length checks support editing, not a universal ideal label size.

Write a decision such as “determine whether customers organize help by product area or by task.” That research objective shapes the content sample and recruitment. Do not ask a card sort to decide visual navigation treatment, page layout, or search ranking; those need different evidence.

Build a representative card set

Inventory real content before making cards. Include common destinations, high-consequence items, and items that currently cause confusion. Remove exact duplicates, but retain meaningful distinctions such as changing a payment method versus reviewing a past payment. Those may look similar internally while serving different goals.

Each card should represent one concept at a consistent level of granularity. Mixing “Account settings” with “Change the last digit of a telephone extension” makes grouping difficult for reasons unrelated to your architecture. Include a short neutral description only where a label would otherwise be genuinely ambiguous.

A few dozen cards can be manageable for a focused study, but the right number depends on complexity and participants' domain knowledge. Pilot the set and watch for fatigue. Splitting a very large inventory into overlapping subsets requires a considered analysis plan; it is not equivalent to everyone sorting the same set.

Nielsen Norman Group's card-sorting overview explains the method's role in information architecture. Keep an audit trail from each card to its source content. Your later design work will need that connection when stakeholders ask what a proposed category actually contains.

Run a pilot before scaling recruitment

Choose participants who represent the people making these navigation decisions, including different experience levels where relevant. New customers may organize concepts differently from experts who have learned your existing terminology. Neither group is automatically wrong; the product may need to support both.

Explain that there is no single correct grouping. In an open sort, participants should name groups after sorting rather than forcing every item into an early label. Allow an uncertain group or comments so ambiguity is visible instead of hidden in arbitrary placement.

For a moderated pilot, ask participants to explain difficult choices without suggesting a destination. Notice when two labels are treated as synonyms, when a card appears to belong in several groups, and when participants lack enough context. These observations reveal label ambiguity that a similarity score alone cannot diagnose.

If you revise cards after the pilot, keep those results separate or clearly document the change. Consistent materials matter when comparing patterns. The GOV.UK research planning guidance can help structure recruitment, consent, and practical session arrangements without turning the sort into an informal stakeholder workshop.

Interpret grouping patterns carefully

Start by reading participants' explanations and group names. Then inspect how often pairs of cards appear together. A similarity matrix summarizes co-occurrence, while a dendrogram offers one possible hierarchical representation. Neither is a ready-made sitemap or an objective discovery of the only correct categories.

Look for stable clusters, split clusters, and outliers. A stable cluster suggests a coherent relationship. A split cluster may reflect different task contexts or audience needs. An outlier may be poorly worded, genuinely cross-cutting, or less familiar to the recruited audience.

Avoid averaging away meaningful differences. If workspace administrators group access controls together but occasional contributors place invitations under project collaboration, consider role-specific entry points or contextual routes. Do not immediately create separate navigation trees without understanding the underlying tasks.

Track confidence and exceptions alongside each proposed grouping. A category supported by consistent sorting and clear explanations is different from one assembled mainly to avoid leftover cards. Use the UX research category to plan additional discovery when participants disagree because the content spans fundamentally different goals.

Translate findings into a testable structure

Draft a candidate hierarchy using participants' language where it is clear, accurate, and appropriate for the product. Avoid copying a group name that works only because the participant remembered the full card set. Navigation labels must make sense before someone sees everything inside them.

Document the rationale for each major category and flag cross-listing candidates. A billing address may belong under account settings while also being reachable during checkout. Information architecture can support multiple routes without duplicating the underlying content or creating conflicting versions.

Validate the candidate structure with tree testing or realistic findability tasks. Card sorting measures grouping expectations; tree testing evaluates whether people can navigate the hierarchy to a destination. A promising cluster can still fail if its label is too abstract or if the relevant branch is buried several levels down.

Keep implementation constraints visible but separate from research findings. A content platform's current folder structure is a delivery constraint, not evidence of users' mental models. Nielsen Norman Group's information-architecture guidance helps distinguish the underlying organization from the visible navigation controls. Finish with a decision record describing the chosen hypothesis, rejected alternatives, evidence gaps, and planned validation tasks.

Key takeaways

• Match open, closed, or hybrid sorting to your decision.
• Use real content at a consistent level of detail.
• Pilot labels and instructions before expanding recruitment.
• Interpret numerical clusters alongside participants' explanations.
• Validate the resulting hierarchy with findability tasks.