Card Sorting: How to Design Better Navigation by Listening to Your Users

TL;DR: Card Sorting

• Card sorting is a UX research method where participants group labeled cards to reveal how they naturally categorize information.
• There are three types – open (exploratory), closed (validating), and hybrid – and each serves a different stage of IA design.
• Most teams skip analysis best practices; dendrograms and similarity matrices are where the real insights hide – not just the groupings.
• AI is reshaping card sorting: from auto-generating card labels to synthesizing patterns across hundreds of responses in seconds.
• If participant recruitment is slowing you down, tools like Articos run research with synthetic personas in under 30 minutes – no scheduling, no no-shows.

You’ve redesigned the nav three times. Users still can’t find the pricing page. Or the support docs. Or, mysteriously, the ‘Contact Us’ link that’s been sitting in the footer this whole time.

The issue usually isn’t the design – it’s that the structure of your information reflects how your team thinks, not how your users think. Card sorting fixes that.

This guide covers everything: what card sorting is, when each type applies, how to run one properly, how to analyze results without losing your mind, and where AI is starting to change the game for UX researchers who need answers fast.

What Is Card Sorting in UX?

Card sorting is a UX research method where participants organize labeled cards into groups that make sense to them. Each card represents a piece of content – a page, a feature, a concept – and participants decide how those pieces belong together.

The output isn’t just groupings. It’s a window into your users’ mental models – how they expect information to be organized before they ever encounter your navigation menus.

As Nielsen Norman Group explains, card sorting “provides insight into how users naturally categorize information on your site and helps teams develop an information architecture that best matches users’ mental models.” That’s the core promise: build the structure around your users, not around your org chart.

It’s been used in UX since the 1990s, borrowed from cognitive psychology, and it still works because the underlying problem hasn’t changed – people organize information differently than product teams assume they do.

Why Use Card Sorting in UX Research?

Most navigation issues don’t show up in analytics until it’s too late. By then, users have already bounced, failed to convert, or given up finding what they needed. Card sorting catches these problems upstream.

Here’s where it actually earns its place:

• It exposes the gap between internal language and user language. Your team calls it ‘Enablement Resources.’ Your users look for ‘Help.’ A card sort surfaces that disconnect before it becomes a UX debt problem.
• It removes guesswork from IA decisions. Instead of debating menu structures in a room full of opinions, you have data showing how real users actually group content.
• It’s faster and cheaper than most alternatives. Digital card sorts can be run unmoderated, asynchronously, at scale – no lab required.
• It works at any stage. Building from scratch? Use it to define your IA. Already live? Use it to spot navigation problems that usability tests keep flagging but not explaining.

One thing most articles skip: card sorting also validates terminology. The names participants give their groups are often more valuable than the groupings themselves. Those label choices tell you exactly what language to use in your navigation.

Types of Card Sorting

Open Card Sort

Participants receive a stack of cards with no predefined categories. They sort the cards into groups that make sense to them and name each group themselves.

This is a generative method – it’s exploratory by design. You’re discovering how users naturally structure information, not testing whether your existing structure works.

Best for:

• Designing a new information architecture from the ground up
• Discovering what terminology users actually use
• Identifying content groupings you hadn’t considered

Vaida Pakulyte, UX Researcher at Electrolux, put it well:

“An open sort exercise is good at the beginning of the project because it helps you to understand how people naturally categorize information.”

Closed Card Sort

Participants receive both cards and predefined category labels. Their job is to assign each card to the category that fits best – no creating new groups allowed.

This is an evaluative method. You’re validating whether your existing categories work, not discovering new ones.

Best for:

• Testing whether your current navigation categories make sense to users
• Comparing two different IA structures head-to-head
• Following up an open card sort with validation

Worth noting: NNG recommends tree testing over closed card sorting when you want to evaluate navigation effectiveness. Closed sorts can introduce bias because users feel pressure to fit cards into your predefined boxes. Tree testing is often cleaner for validation work – but more on that later.

Hybrid Card Sort

Participants get some predefined categories but can also create new ones. It’s the middle ground – useful when you’re confident about some of your structure but uncertain about the rest.

Best for:

• Redesigning a live website where you want to retain proven categories
• Testing partial structures without starting from scratch
• Complex IA projects where some groupings are settled but edge cases remain

The tradeoff: predefined categories anchor users’ thinking. Even in a hybrid sort, the categories you provide tend to pull other groupings toward them. That’s why IxDF recommends using hybrid sorts sparingly and only when you’re genuinely confident in at least some of your existing structure.

Bonus: Reverse Card Sort / Tree Testing

Tree testing (sometimes called reverse card sorting) isn’t technically a card sort – but it often follows one. Once you’ve used a card sort to design an IA, tree testing checks whether users can actually navigate it to find specific things.

Think of it this way: card sorting builds the structure, tree testing stress-tests it.

When to Choose Each Card Sorting Method

Here’s a quick decision framework rather than another generic pros/cons list:

Step-by-Step Guide to Conducting a Card Sort

Running a card sort isn’t complicated. But the details matter – bad card labeling, too many cards, or wrong participant selection will sink the quality of your results regardless of which method you choose.

Step 1: Define Your Research Objective

Before touching a single card, be specific about what question you’re trying to answer. ‘Improve our navigation’ is too vague. ‘Understand how our target users would group our 35 product features’ is a research objective.

Also decide: do you need qualitative insight (why users group things the way they do), or quantitative validation (how many users agree on a grouping)? That affects your participant count and method choice.

Step 2: Write Your Cards

This is where most teams go wrong. Card labels need to be:

• Specific enough to be distinct, but not so technical that they require domain expertise to understand
• Written in user language, not internal product language
• 30–50 cards total – fewer than 30 leaves gaps; more than 50 causes fatigue and loose groupings

One thing to avoid: don’t use labels that share obvious keywords (e.g., ‘Invoice Settings’ and ‘Invoice History’) – users will automatically group them by the shared word rather than by actual conceptual similarity. That skews your data.

Underused tip: Run a quick pilot with 2-3 people before your full study. This almost always catches confusing card labels you’d missed.

Step 3: Recruit Participants

Participant quality matters more than participant quantity here. For qualitative card sorting, 15-30 participants who represent your actual users gives you enough signal. For quantitative analysis, you want 30+ participants.

Most teams default to whoever they can recruit fastest – colleagues, friends, or generic panel participants. This is a real problem for B2B products or niche audiences where the mental models of general consumers don’t apply. A healthcare SaaS sorting exercise conducted with tech workers will produce bad IA. Domain-appropriate participants are non-negotiable.

Step 4: Run the Session

Moderated: Have participants think aloud as they sort. Ask open-ended follow-up questions: ‘You put those two together – can you tell me more about why?’ The transcripts from moderated sessions are often more valuable than the sort data itself.

Unmoderated: Use digital tools (OptimalSort, Maze, UXTweak, or similar). Provide clear written instructions. Randomize card order for each participant to avoid anchoring effects – if everyone sees the same first card, early groupings influence everything that follows.

Step 5: Analyze Results

This is where most guides give you vague advice about ‘looking for patterns.’ Here’s what that actually means in practice:

• Similarity matrix: Shows how frequently any two cards were grouped together. High co-occurrence = strong relationship. Start here.
• Dendrogram: A hierarchical cluster diagram that visualizes which cards naturally belong together. Read it from the bottom up – tight clusters at the bottom are clear groupings, looser ones at the top are ambiguous.
• Popular placements matrix: For closed sorts – shows which category each card landed in most often.
• Group naming analysis: Catalogue what participants called their groups. Recurring terms are your navigation labels.

The dendrogram is the most underused tool in card sort analysis. Most teams glance at their top categories and call it done. The dendrogram shows you the cards everyone agreed on and the cards that split participants down the middle – which is arguably more useful, because that’s where your navigation decisions are genuinely hard.

Step 6: Translate Into IA Recommendations

Card sort data doesn’t directly become a sitemap. It informs one. Take your clusters and naming patterns, then map them against your existing content inventory. Where card sort groupings conflict with your current structure, that’s a design decision you now have data to support.

Always follow up with tree testing before you commit to a new IA. Card sorting tells you how users want to group things – tree testing confirms they can actually find things in that structure.

What Card Sorting Can’t Tell You (And What to Do About It)

Card sorting answers the question of how users expect information to be organized. It doesn’t answer how well they navigate under real conditions – with real page layouts, visual hierarchy, and competing links.

Some specific limitations most guides skip over:

• It assumes users have enough domain familiarity to sort meaningfully. For highly technical B2B products, novice participants produce groupings that reflect their confusion, not their mental model. You need domain-appropriate participants, or your IA will be built on noise.
• It doesn’t account for cross-categorization. Many content items belong in multiple places – a product page might fit under both ‘Features’ and ‘Use Cases.’ Card sorting forces single placement. Your IA shouldn’t necessarily impose the same constraint.
• Participant fatigue distorts results. After card 40+, groupings get sloppier and ‘Miscellaneous’ piles multiply. Keep your card set focused.
• It’s still based on stated behavior, not observed behavior. What users say they’d do in a sorting exercise doesn’t always match what they do when navigating under time pressure. Complement card sorting with analytics and usability tests.

How to Use AI for Card Sorting UX

AI hasn’t replaced card sorting. But it’s genuinely changing how fast teams can run studies and how much they can do with the results.

Here’s where AI is actually useful in a card sorting workflow:

• Generating card labels. Feed your content inventory or sitemap to an AI and ask it to generate 40-50 card candidates. This is a solid starting point – you’ll edit it, but it’s faster than staring at a blank spreadsheet.
• Synthesizing open sort data. When you have 50+ participants using different language for essentially the same category, AI can cluster synonymous group names. ‘Help,’ ‘Support,’ ‘Get Help,’ and ‘Contact Support’ are all the same intent – AI spots that quickly.
• Writing analysis summaries. After you’ve done the hard work of reading dendrograms and similarity matrices, AI can help you turn raw data observations into a stakeholder-ready findings document.
• Running synthetic card sorts. For early-stage exploration – before you’ve defined your real participant panel – AI-powered synthetic personas can give you a directional sense of how different user types might group your content. Not a replacement for real users, but useful for hypothesis generation.

Most importantly: AI doesn’t help you think of the right questions to ask. Defining what cards to include, what research objective to pursue, and what decisions the data needs to support – that’s still your job.

How Articos Accelerates UX Research (Including Card Sorting Prep)

Card sorting is only as useful as the research it’s embedded in. The bigger problem for most product teams isn’t the sort itself – it’s everything around it: finding the right participants, analyzing data quickly enough to matter, and fitting research into a two-week sprint cycle.

That’s where Articos comes in. It’s an AI-powered research platform that lets you run user research with synthetic personas – without the 6-8 week recruitment process that usually makes research impractical in fast-moving product cycles.

Instead of waiting weeks for participants to schedule time:

• Define your user type (e.g., B2B SaaS admin, 50-person company, mid-level ops role)
• Articos generates realistic synthetic personas matching your target demographic
• Interviews run simultaneously – no scheduling, no cancellations, no incentives to manage
• Results in under 30 minutes, with synthesis built in

For card sorting specifically: use Articos to run a rapid round of user interviews before your card sort. Those conversations help you understand what your users actually care about – so you’re writing better card labels and testing the right structure from the start.

It’s particularly useful for product teams without a dedicated research function – the people who need validation fast but don’t have the infrastructure or budget for traditional recruitment.

Try Articos free →

Card Sorting Across Different Industries: What Changes

The core method stays the same, but what you’re testing – and who you’re testing it with – varies a lot by context.

• E-commerce: The priority is product taxonomy. How shoppers browse and group products often differs dramatically from supplier category structures. A card sort for an apparel retailer might reveal that users organize by ‘occasion’ (work, weekend, formal) rather than by clothing type (tops, bottoms, outerwear). That’s IA-changing insight.
• B2B SaaS: Navigation often mirrors internal product architecture rather than user workflows. Card sorting helps expose that mismatch – where the settings menu contains 40 items that users would logically split between ‘Admin,’ ‘Integrations,’ and ‘Billing.’
• Healthcare / regulated industries: Participant selection is critical. Clinicians and patients have completely different mental models for the same content. Run separate sorts for separate user types and compare – the divergence is usually more informative than the consensus.
• Content-heavy sites (media, publishing, education): Card sorting helps surface how readers/learners categorize topics versus how editorial teams do. Tagging systems, article categories, and learning paths all benefit from this.

Card Sorting vs. Other UX Research Methods: When to Use What

Card sorting doesn’t exist in a vacuum. Here’s how it fits with other methods you’re probably already using:

• Card sorting + user interviews: Interviews tell you why users group things the way they do. Card sorting tells you what the groupings are. Run interviews to build your card labels; run the sort to validate structure.
• Card sorting + tree testing: The natural follow-up. Card sorting designs the IA; tree testing validates it. Learn more about running fast user research without weeks of recruitment.
• Card sorting + usability testing: Usability tests flag navigation problems; card sorting diagnoses them. When your test participants fail a task because they’re looking in the wrong place, a card sort tells you where they expected to look instead.
• Card sorting + AI research tools: If you need faster user research to inform card design or to run supplementary persona-based exploration, AI research platforms can compress weeks of prep work.

Advanced Card Sorting Tips Most Guides Don’t Cover

These show up rarely in standard card sorting guides – which is mostly why navigation redesigns still fail even after running a sort.

• Run bi-directional sorts. Run an open sort with new users and a closed sort with your existing customers separately. The divergence between the two often reveals exactly where your current navigation fails your power users versus new ones.
• Analyze the ‘misfits’ separately. Cards that nobody agrees on – the ones that end up in completely different places across participants – are more valuable than the obvious clusters. Those are the content items that your users are genuinely confused about. Address those first.
• Don’t ignore the ‘left behind’ cards. In unmoderated sorts, participants sometimes leave 2-4 cards unsorted. That’s not laziness – it often means those items don’t fit anywhere in the user’s mental model. Treat unsorted cards as a signal that the content itself may need repositioning, not just recategorizing.
• Test with empty cards. In advanced open sorts, include 3-5 blank cards and ask participants to write in any content they’d expect to find that isn’t already there. This surfaces missing content gaps that card sorting alone can’t catch.
• Run a second sort after your redesign. Most teams run a card sort before an IA redesign and never again. Running a second sort 6-12 months post-launch tells you whether the new structure actually matches how users have evolved – or whether new content has introduced new confusion.

The Bottom Line

Card sorting is one of those methods that’s genuinely simple to explain and genuinely hard to do well. The technique hasn’t changed much in 30 years, but the context around it has: faster product cycles, more content to organize, more diverse user bases, and now AI tools that can handle parts of the workflow that used to take days.

The teams that get value from card sorting are the ones who run it with real users (or realistic synthetic ones when time is critical), write careful card labels, and do the full analysis – not just skim the obvious groupings. The dendrogram is your friend. The misfits are your goldmine.

If participant recruitment is what’s been holding your research back, Articos removes that bottleneck entirely. Sign up free and run your first research session today – insights in 30 minutes, no scheduling required.

FAQs: Card Sorting