Card Sorting in UX: How to Organise Content With Users

TL;DR: What You Will Learn

• Card sorting is a UX research method where users sort content items into groups that make sense to them, revealing how they expect your site to be organised.
• Use it early, before design, when you are building or restructuring a website, app, or e-service and need a solid information architecture (IA).
• The payoff is better findability: menus and categories that reflect how users think, which means fewer abandoned tasks and less frustration.
• It comes in three flavours (open, closed, and hybrid), and you can run it in person or online with tools you likely already have access to.

What Card Sorting Is and Why It Matters for UX

Let me start with the simplest definition I can give you. Card sorting is a research method where you give participants a set of labelled cards, each representing a piece of content, a feature, or a page, and ask them to group those cards into categories that make sense to them.

You are not telling them how to organise things. You are watching how they naturally do it.

That distinction is the whole point. When your users sort the cards their way, they hand you a map of how they think, and that map is far more useful than any structure you dreamed up in a meeting room.

What Card Sorting Reveals About Mental Models

A mental model is simply the picture in someone’s head of how something works. Your users already have expectations about where a service should live before they ever click.

Card sorting surfaces those expectations. It’s exactly what Nielsen Norman Group defines the method as doing: uncovering users’ mental models of a product’s information architecture. When ten people all drop “reset PIN” under “security”, you have learned something you could never guess from a spreadsheet.

Donna Spencer, author of Card Sorting: Designing Usable Categories, still the field’s go-to reference on the method, frames the payoff the same way: the whole point of a sort is to learn “what things people think go together”, not to confirm the structure you already had in mind. That is exactly why the technique catches blind spots a stakeholder meeting never will.

It also exposes your blind spots. Teams often organise content by internal departments, but users organise by the task they came to complete, and card sorting makes that gap impossible to ignore.

Where It Fits in The UX Process

I always run card sorting before wireframing, not after. Once you have committed to a sitemap and started drawing screens, changing the structure becomes expensive and emotionally painful.

Card sorting sits early, feeding directly into your information architecture (IA) and navigation design. It is a research activity, and it pairs closely with understanding the difference between UI and UX, because good structure is pure UX work that happens long before anyone picks a colour or a button style.

Get this right, and your wireframes almost draw themselves, because the menu already reflects how people think.

The Business Payoff of Better Findability

Here is where it becomes a business case, not just a research nicety. Imagine a bank like DBS restructuring a dense self-service menu, or an HDB e-service reorganising the crowded list of options a resident faces when applying for a flat or a grant.

When services are grouped the way users expect, people find what they need on the first try instead of the fourth.

That directly lightens the load on your support desk. Every call or chat that starts with “I couldn’t find where to…” is a findability failure, and card sorting quietly removes a chunk of those before launch.

For a Singapore team weighing time and budget, that is the payoff worth investing in:

• Fewer misdirected support tickets and phone calls
• Higher task completion on the first attempt
• A navigation structure your whole team can defend with evidence

That is why smart teams sort cards before they commit to a single line on the sitemap.

Open, Closed, and Hybrid Card Sorting Explained

Not all card sorting studies work the same way, and picking the wrong variant wastes your participants’ time and your budget.

The three main types are open, closed, and hybrid. Which one you choose depends almost entirely on what stage your project is at.

Let me walk you through each using a scenario I see often: a homeware SME that has been selling on Shopee for a couple of years and now wants to launch its own standalone website.

Open Card Sorting

In an open sort, participants group the cards however they like and then name each group themselves.

This is your go-to when you are building information architecture from scratch and have no fixed categories yet.

For our Shopee homeware seller, this is the starting point. They give participants cards like “ceramic mugs”, “bamboo trays”, “linen tablecloths”, and “storage baskets”, then watch how real shoppers cluster them and what labels they invent.

Open sorting reveals the mental model your customers already have, which is exactly what you need before you commit to a menu structure.

The trade-off is messiness. Because everyone names groups differently, the results take longer to analyse, and you will need to spot patterns across sessions rather than read off a clean answer.

Closed Card Sorting

In a closed sort, you provide the categories in advance, and participants slot each card into the group they think fits best.

Use this when you already have a proposed structure and want to validate whether people can find things in it.

Once our homeware seller has drafted categories from their open sort, say “Kitchen”, “Dining”, “Living”, and “Storage”, they run a closed sort to test it. If shoppers keep dropping tablecloths into “Living” instead of “Dining”, the structure needs a rethink.

The trade-off is that closed sorting cannot tell you about categories you never thought of. You only learn whether your existing buckets work, not whether better ones exist.

Hybrid Card Sorting

Hybrid sorting sits in the middle. You offer some predefined categories but let participants create and name new groups when nothing fits.

This is ideal for refinement, when you have a rough structure but suspect there are gaps.

Our seller might run a hybrid sort as a final check before launch, giving the four main categories but allowing participants to add a new group if, say, “Gifts” or “Sale items” feels missing.

Here is a simple way to match method to stage:

1. Starting fresh with no categories: run an open sort.

2. Validating a structure you already have: run a closed sort.

3. Refining a near-final structure and hunting for gaps: run a hybrid sort.

Most substantial projects, such as this website rebuild, benefit from running an open sort first and a closed sort later to confirm the result holds up.

Planning a Card Sorting Study

Before you drag a single card, you need a plan. In my experience, the difference between a card sorting study that reshapes your information architecture and one that produces confusing noise is decided almost entirely at the planning stage.

Good planning determines whether your results are usable. Rush it, and you will end up with clean data that answers the wrong question.

Defining Your Research Objective and What to Test

Start by writing down one clear question. Are you validating a proposed menu structure, or discovering how users would group content from scratch?

For an open sort, you are exploring how people mentally group items. For a closed sort, you are testing whether your existing categories make sense to real users.

Say you are redesigning a public-sector portal where citizens apply for services and check statuses. Your objective might be to find out whether people expect payments, applications, and appeals to sit together or apart.

Keep the scope tight. Testing 50 unrelated cards across five different goals gives you tangled results that satisfy no one.

Choosing and Writing Card Content

Each card is one piece of content, usually a page or a task. Choose items that represent the real breadth of your site, not just the easy or obvious ones.

Write labels the way your users speak, not the way your internal teams do. Avoid jargon, department names, and acronyms that only make sense inside your organisation.

For that government portal, a card labelled “Renew Passport” is far clearer than “ICA e-Service Module 3”. If your participants have to guess what a label means, you are testing comprehension, not grouping.

Aim for roughly 30 to 60 cards for most studies. Optimal Workshop’s own card sorting guidance lands in that range. Fewer than 30 rarely reveals structure, and more than 60 starts to exhaust participants.

Deciding Participant Numbers and Recruiting the Right Users

For card sorting, 15 to 20 participants is a common benchmark for stable patterns; the Nielsen Norman Group recommends testing at least 15 users per group for reliable results. Beyond that point, groupings tend to repeat rather than surprise you.

Recruit people who actually resemble your users, not just colleagues who are easy to grab. For a Singapore public-sector portal, that means spanning age groups so your findings reflect Singapore’s digitally diverse population.

Include seniors who log in with SingPass and may navigate very differently from younger, mobile-first users. A structure that feels obvious to a 25-year-old can baffle someone less confident online, and public services must serve both.

When you collect participant details, respect the PDPA. Get clear consent, explain how you will use their data, collect only what you need, and store it securely.

• Tell participants what the session involves and how long it takes.
• Record consent before you begin.
• Anonymise results when you report them.

Understanding how research shapes design decisions is exactly the kind of thinking that comes up in UX designer interview questions hiring managers ask, so treat your planning as practice for the job too.

Running a Card Sorting Session Step by Step

When I run my first card sorting session with a new team, I always remind them that the session lives or dies by how little the facilitator interferes. Your job is to set the stage, then get out of the way.

The Session Flow From Briefing to Grouping

Start with a short briefing. I tell participants there are no right or wrong answers, that we are testing the content and not them, and that I want their honest, instinctive groupings.

Then I hand over the cards and let them sort. The temptation to nudge is strong, but the moment you say “wouldn’t that fit better under Payments?” you have contaminated the data.

Instead, I use gentle think-aloud prompts. Something like “tell me what you’re thinking as you move that one” keeps the reasoning flowing without steering the outcome.

Capturing the why matters as much as the groupings. When someone hesitates over where “Transaction History” belongs, that hesitation is a signal, so I jot it down alongside their final label for the group.

Moderated Versus Unmoderated, and When to Use Each

Moderated sessions, whether in person or over video, give you richer insight because you can probe with follow-up questions in the moment. The trade-off is time: you can realistically run only a handful before it eats your week.

Unmoderated sessions scale beautifully. Using a tool like OptimalSort or Maze, you send a link and participants sort on their own schedule, so you can gather thirty or forty responses without booking a single meeting room.

I worked through this logic with a small Singapore fintech redesigning its transfers dashboard. They ran remote unmoderated sorts with users islandwide, from Jurong to Punggol, which saved everyone the commute and gave a broad baseline quickly.

Then they layered in three moderated video sessions to understand the tricky groupings the data raised. Combining scale with depth gave them confidence no single method would have.

Practical Do's and Don'ts for First-Timers

Keep the card count manageable. I aim for somewhere between thirty and fifty cards, because past that point participants fatigue and start dumping items into a “miscellaneous” pile just to finish.

A few habits I follow every time:

• Do use neutral card labels that match your real content, not internal jargon like “SKU management”
• Do stay silent while people sort, and save your questions for the end
• Do pilot the session with one colleague first to catch confusing cards
• Don’t group the cards yourself as an example, since that anchors their thinking
• Don’t ask leading questions such as “does that feel like Settings?”

Watch your own body language too. A raised eyebrow when someone makes an unexpected choice is enough to bias the next few cards, so I keep my reactions flat and my curiosity genuine.

Tools for Running Card Sorts Online and In Person

Once you have decided on an open or closed sort, the next question is what you actually run it with. The honest answer is that card sorting does not need expensive software to be useful, but the right tool saves you hours when you scale up.

Let me walk you through the toolkit I reach for, and how I decide between them.

Physical Tools for In-Person Workshops

For a room of five to eight participants, nothing beats index cards, sticky notes, and a big whiteboard.

I write one content item per card, hand the deck to a participant, and watch how they group things while thinking aloud. That observation is gold, because you catch the hesitation and the muttered “where does this even go” that a screen hides.

The trade-off is analysis. You are photographing groupings and transcribing them by hand, which gets painful past a handful of sessions.

Digital Platforms for Remote and Larger Studies

When you need thirty participants across Singapore, or you are running a distributed team, digital platforms earn their keep.

Optimal Workshop is the specialist most people start with, and UXtweak and Maze cover similar ground for card sorts and tree tests. These tools recruit or link out to participants, run the sort in the browser, and crucially auto-generate the analysis for you.

Instead of building spreadsheets, you get dendrograms, similarity matrices, and agreement scores out of the box, so you spend your time interpreting rather than tabulating.

Miro and FigJam sit in a different lane. They are collaborative whiteboards, so they suit facilitated, moderated sorts where you want to talk with participants live rather than send an unattended study.

FigJam is the natural pick if your team already designs in Figma, since the cards live next to your wireframes and you avoid tool-switching. If you are new to that ecosystem, it helps to understand what Figma is used for in design teams first, and there is a step-by-step Figma tutorial for beginners to get your bearings.

How to Choose for Your Project

Three factors decide it for me: participant count, budget, and whether you are remote or in the same room.

• Small, in-person, low budget: sticky notes and a whiteboard
• Moderated remote sessions with a design team: Miro or FigJam
• Large unmoderated studies needing clean stats: Optimal Workshop, UXtweak, or Maze

Here is a practical Singapore angle. Rather than outsourcing research to an agency, an SME can send a couple of staff for SkillsFuture-funded UX training and bring these tools in-house.

The bigger win is capability, not just a one-off study. Once your team knows how to run and read a card sort, you can repeat it every time the content grows, which is far cheaper than paying per project.

Analysing Card Sorting Results

Running a card sorting study is the easy part. The real work begins when you have a spreadsheet full of piles, and you need to turn that mess into decisions your whole team will trust.

Let me walk you through the three tools I lean on most, and how I read them without kidding myself.

Similarity Matrices and How to Read Them

A similarity matrix is a grid where every card appears on both axes, and each cell shows the percentage of participants who placed those two cards in the same group.

I read it by scanning for dark, high-percentage clusters near the diagonal, the same approach Optimal Workshop lays out in its guide to reading a similarity matrix. Cards that sit together at 80 percent or above are telling you something loud.

A strong grouping shows a tight block of high agreement, while a weak one is scattered and inconsistent. Once agreement drops into the 30-50 percent range, treat it as a genuine split in user expectations rather than noise. Those middling scores are the interesting ones, because they signal cards users are genuinely unsure about.

Dendrograms and Cluster Analysis

A dendrogram is a tree diagram that shows how cards join into clusters as agreement loosens (Optimal Workshop breaks down the two scoring methods, AAM and BMM, here). Cards that merge close to the left, at high agreement, form your most confident groups.

Here is a concrete case. A healthcare portal team I worked alongside in Singapore kept finding that “appointment booking” and “billing” merged early in the dendrogram, across almost every participant.

That was not noise. Users mentally linked booking a slot with paying for it, which signalled that the existing navigation, splitting these across separate top-level menus, was fighting user expectations. The team moved them into one “Manage my visit” section, and that decision came from the data, not a stakeholder’s hunch.

Standardisation Grids and Category Labels

In an open sort, participants invent their own category names, and this is gold for labelling. A standardisation grid groups these raw labels into standardised themes so you can count how often each concept appears.

When ten participants independently name a pile “My Records”, “Health History”, and “Past Visits”, you have both your grouping and a strong candidate label handed to you.

I always cross-check these labels against the matrix. If the naming agreement is high but the sorting agreement is low, the label is popular, but the content underneath is not settled.

Reading Agreement Honestly

The biggest trap is over-interpreting a small sample. With eight participants, a single person’s unusual sort can shift a percentage dramatically, so treat low counts as directional, not definitive.

Resist cherry-picking too. It is tempting to highlight the one cluster that confirms the redesign you already wanted, but discipline means reporting the weak groupings alongside the strong ones.

My rule of thumb: act confidently on clusters with consistent high agreement across most participants, flag the ambiguous cards for a follow-up tree test, and never dress up a 40 percent split as a mandate.

Turning Results Into Navigation and Information Architecture

You have run your card sorting sessions, and now you are staring at a wall of clusters. The real work is translating those groupings into a navigation people can actually use.

Here is the honest bit: card sorting gives you strong signals about how users mentally group content, but it is one input, not the final verdict. You still weigh it against your analytics, your business priorities, and stakeholder realities.

Drafting a Sitemap from Clusters

Start by turning your strongest, most consistent clusters into candidate top-level categories. If most participants grouped the same cards, that grouping earns a place in your primary navigation.

Picture a Singapore retail brand rebuilding its site. If shoppers repeatedly clustered store locations, opening hours, and return policy under one mental bucket, that becomes a clear candidate for a “Visit Us” or “Support” section rather than three scattered links.

Loose or contested cards need judgement. When a card keeps landing in different groups, do not force it. Flag it, look at what your analytics say about how often that page is used, and decide with stakeholders whether it deserves prominence or belongs in a footer.

Your sitemap is a hypothesis built from user logic, refined by business context.

Naming Menus in Users' Language

Clusters tell you what belongs together. The labels your participants used, especially in an open sort, tell you what to call each group.

This matters enormously for a tourism brand serving both overseas visitors and locals. An overseas traveller might search for “getting here” while a local looks for “opening hours and tickets”, so your labels have to work for both without insider jargon.

Resist internal naming. Marketing might love “Experiences”, but if users called it “Things to do”, use their words. Clear labels also feed directly into your SEO, which is why I always connect card sort labels to a wider plan for how to build a content marketing strategy that boosts SEO.

Validating With Tree Testing Before Build

Before anyone writes code, test your proposed structure with a tree test; as NN/g puts it, card sorting generates ideas for an IA; tree testing evaluates them. You strip away the visual design and give users tasks like “find how to get a refund”, then watch whether they land on the right branch.

For that tourism site, you might ask overseas visitors to find airport transfer information and locals to find annual pass pricing. If both groups navigate confidently, your architecture holds.

If they get lost, you fix labels and groupings now, when changes cost minutes rather than a full rebuild.

Then the handoff is clean:

1. Deliver the validated sitemap and label list to your designers.

2. Include tree test findings and any unresolved cards with recommendations.

3. Note where analytics or business needs overrode the raw sort.

That documented trail turns messy sticky notes into a navigation design your team can defend and build with confidence.

Common Card Sorting Mistakes and How to Avoid Them

I have run enough card sorting studies to know that the method fails quietly. You get tidy-looking results, act on them, and only later realise the study was flawed from the start.

Here is what usually goes wrong, and how to catch it before it costs you.

Too Many Cards, and Cards Nobody Understands

The fastest way to ruin a session is fatigue. When I hand someone 80 cards, the last 30 get sorted carelessly because they just want it over with.

Keep your deck lean, roughly 30 to 50 cards for most studies. If you have more content than that, split it across multiple sessions or prioritise the pages people actually visit.

Ambiguous or jargon-heavy labels are just as damaging. A card labelled “e-services portal integration” tells a participant nothing, so they guess, and your data becomes noise.

Write cards in the plain words your users use, not your internal team’s shorthand. When in doubt, test a few labels on a colleague outside your project first.

Leading Facilitation and The Wrong Participants

I have watched facilitators nudge people (“that one usually goes under Services, right?”) without realising it. That single hint contaminates the whole session.

Stay quiet during moderated sorts. Explain the task, then let silence do its work, and only step in if someone is genuinely stuck.

The bigger trap is recruiting the wrong people. I once advised a Singapore public sector team that ran a card sort using only their own staff because it was fast and convenient.

Staff already knew the internal logic, so they produced a neat structure that later confused older citizens who could not find schemes they were entitled to. Once the team recruited a broader, PDPA-compliant group that included seniors and non native English speakers, the resulting categories looked completely different and far more usable.

The fix is simple: recruit participants who mirror your real users, obtain proper consent, and handle their data responsibly under the PDPA.

Treating Results as Final

The last mistake is treating a card sort as the final answer. It is not.

Card sorting tells you how people group and label content, but it does not tell you whether they can complete tasks in your live product. Always validate your proposed structure with follow-up methods.

• Run tree testing to check if people can find items in the new hierarchy.
• Follow with usability testing on a prototype to see real task flows.
• Cross-check against analytics to confirm which pages people genuinely need.

Card sorting complements these methods; it does not replace them. Use the trio together, and you build an information architecture you can actually defend.

Conclusion

If you take one thing away from this guide, let it be this: card sorting is one of the cheapest, highest-signal methods you have for grounding your information architecture in how real people think, not in how your internal teams happen to be organised.

Remember the pain point we opened with. A confusing menu is not a cosmetic problem. When a shopper on your Singapore e-commerce site cannot find the returns policy, or a citizen cannot locate a form on a government portal, you lose the conversion, and you chip away at trust in the same click.

Card sorting fixes that at the root. Instead of guessing which label belongs under which heading, you watch how your actual users cluster and name your content, then you build navigation that matches their mental models.

The workflow is straightforward once you have seen it end to end. You plan your cards and choose open, closed, or hybrid. You recruit a modest group of representative participants. You run the sessions, whether in person or through a tool. Then you analyse the groupings and the labels, and you translate those patterns into a clearer menu and page structure.

You do not need a big budget to start. Pick one messy area of your own site, write out fifteen to twenty cards, and run a small pilot with five or six colleagues or customers this week. The insights will surprise you.

To turn these one-off wins into a repeatable skill, structured training helps enormously. The WSQ User Experience & User Interface (UX-UI) Design Essentials course is the logical next step, and it’s SkillsFuture eligible, so your training costs are largely offset.