Activity⚡ Medium

Card Sorting

Let users organise content to reveal their mental models. Card sorting gives you the categories and labels that match how users think — not how your team built it.

Information ArchitectureNavigation DesignContent CategorisationSettings StructureSearch TaxonomyOnboarding Flow

01 — TL;DR

Two sentences.

Card sorting is a research method where participants organise a set of content items — written on cards or displayed digitally — into groups that make sense to them, and optionally label those groups. The output reveals the mental models users bring to your content: which items they expect to find together, what they call those groups, and where they expect to look for specific things.

The method has been used in information architecture research since the 1990s and was popularised by Jakob Nielsen and others as a low-cost, high-signal way to ground navigation and taxonomy decisions in user data rather than internal logic. A card sort with eight participants can surface the dominant mental model for a navigation structure in a day, replacing weeks of assumption-based IA debate with empirical grounding.

02 — When to Use

Apply this when…

→You are designing a navigation structure for a new product or major redesign and want to know how users expect content to be organised

→Users consistently report being unable to find things in an existing navigation

→You are building a help centre or documentation site and need a taxonomy that matches how users search and browse

→A settings panel has grown organically and users complain about finding options

→You need to choose between competing IA proposals and want user data to break the deadlock

When NOT to apply it

Skip it when the content set has fewer than ten items (too trivial) or more than 80 without scoping (produces fatigue). Also skip when the problem is findability within a known category rather than category definition — use tree testing for that. And when you need to validate a finalised structure rather than inform an open one.

03 — How It Works

The mechanism

Card sorting works by externalising users' implicit mental models into observable grouping behaviour. Because the task is concrete and low-stakes, participants reveal their genuine expectations rather than performing approval of a presented structure.

Mental models drive navigation behaviour

Users navigate by matching their internal category expectations against what they find — not by reading and evaluating each option systematically. When categories do not match mental models, users spend disproportionate time searching. Card sorting surfaces the dominant mental model before the navigation is built, so the structure can be designed to match expectation rather than requiring users to adapt to an internal logic.

Open, closed, and hybrid sorts serve different questions

An open sort asks participants to create their own groups and labels — it reveals natural categories. A closed sort gives predefined categories and asks participants to place cards into them — it validates a proposed structure. A hybrid combines both. Open sorts are best early when categories are unknown; closed sorts are best when a structure needs validation before build.

Agreement between participants is the signal

A common mistake is treating the most frequent grouping as the correct answer. The real signal is agreement — how consistently participants group the same items together. High agreement means a strong shared mental model; low agreement means the item is genuinely ambiguous and sits at a boundary between categories. Boundary items are design decisions, not statistical errors.

Dendrograms and similarity matrices reveal the structure

A similarity matrix counts how often each pair of items was grouped together across participants. Items grouped together by more than 70% represent high-confidence categories. Items at 30-50% represent boundary items that may need to appear in multiple locations. Remote tools like Optimal Workshop generate these analyses automatically.

Card sorting is input, not output

Card sorting tells you how users expect to find content — it does not tell you what the navigation should look like, what to call the categories, or how deep the hierarchy should go. Those are design decisions informed by the data, not produced by it. A card sort that shows users group "Billing" and "Account" together is a data point, not a mandate to merge them.

04 — Real Example

Shopify's settings navigation and the merchant mental model

Shopify's admin settings panel was originally organised around internal product architecture — groupings that made sense to engineers but not to merchants. Card sorting with merchants revealed a job-based mental model: they thought in terms of "things I do when I get an order," "things I set up once," and "things I manage regularly" — not in terms of feature modules.

The card sort data directly informed a reorganisation that reduced support tickets for "how do I find X" within months. The deeper principle: the organisation logical to builders is almost never natural to users. Card sorting makes that gap visible before it becomes a support cost.

Shopify · Settings Navigation

Task-based mental model replaces feature-based architecture

Task-based reorg → fewer support tickets

05 — In the Wild

Test yourself & see real examples

No examples yet — be the first.

Spotted a navigation that clearly does not match how users think — or one that feels immediately intuitive? Submit what you observed.

✓ Reviewed before publishing✓ Your name on every example you submit✓ Violation or fix — both welcome

Seen Card Sorting violated in a real product? Help grow the evidence base.

06 — Common Mistakes

Where teams go wrong

✗Running a card sort on too many or too few cards. Fewer than ten produces trivial results. More than 80 without scoping produces fatigue. The practical range is 20 to 60 cards. For larger sets, run multiple focused sorts on meaningful subsets.

✗Treating the most common grouping as the definitive answer. If 60% group two items together, that is a signal to consider — not a mandate. Business constraints, legal requirements, and platform conventions all factor in. The card sort informs the decision, not replaces it.

✗Using card sorting to validate a structure already decided. This produces pressure to interpret ambiguous data favourably. Use card sorting early while decisions are open. If you need to validate a finalised structure, use tree testing.

✗Writing card labels in internal terminology. Cards labelled "CMS Settings" or "Entity Management" produce groupings that reflect label confusion, not mental models. Use language participants already understand.

07 — Variations & Related Principles

Connected ideas

Card sorting sits within the IA research toolkit alongside tree testing and first-click testing. Understanding the differences helps you choose the right method for your question.

The most important pairing is card sorting with tree testing. Card sorting without tree testing produces a structure grounded in user data that has never been validated under task pressure. Together they form a complete IA research loop: generate from mental models, then validate under realistic conditions.

08 — 10-Min Exercise

Run it right now

⏱ 10 minutes · Solo · No prep

The Instant Sort

1. Open the settings or navigation of a product you use regularly. Write down every top-level item you can see. Aim for 15 to 30 items.

2. Without looking at the existing structure, write each item on a separate line and group them however feels natural — as if organising for someone who had never used the product. Label each group in plain language.

3. Compare your groupings to the existing structure. Note every item in a different group and every label that differs from the product's label.

4. Each discrepancy is a hypothesis worth testing. If you have a colleague nearby, repeat the exercise with them and compare. Agreement between two people on a grouping is the beginning of card sort signal.

10 minutes