Card sorting is one of the oldest research methods in user experience, and it remains one of the most useful methods that most teams are not running. The premise is deceptively simple. You give participants a set of labeled cards, each representing a piece of content, a feature, or a concept, and you ask them to organise those cards into groups that make sense to them. What emerges is a map of how users actually think about your product, rather than how your team assumes they think. That distinction is the entire value of the method, because the gap between the mental model in the product team's head and the mental model in the user's head is where navigation breaks, where users get lost, and where products quietly lose customers.
The stakes are higher than the low-tech simplicity of the method suggests. A global survey by Airship found that 57% of users decide whether to delete an app after just one or two uses, with poor navigation cited as a top reason for abandonment. An information architecture that does not match how users think is not a minor usability inconvenience; it is a direct driver of churn. This guide explains what card sorting is, the three forms it takes, when and how to run it, how many participants you genuinely need, how to interpret the results, and how it fits alongside the related method of tree testing.
What Card Sorting Actually Reveals
Card sorting reveals users' mental models, the internal logic by which they expect information to be organised. When a participant decides that "billing history" belongs with "payment methods" rather than with "account settings," they are exposing an expectation that your navigation either matches or violates. Aggregate enough of these decisions across enough participants and clear patterns emerge: items that users consistently group together, labels that users consistently misunderstand, and categories that mean something different to your users than to your team.
This insight is most valuable early in the design process, when you are figuring out how to structure information before committing it to a build. It is equally useful before adding new sections to an existing product, or when you suspect that your current navigation is not serving users well. A common and instructive scenario involves large e-commerce sites struggling with poor navigation where users cannot find products and sales suffer as a result. Running an open card sort with a substantial group of participants, each sorting a hundred or so product cards into categories that make sense to them, repeatedly reveals groupings the team never anticipated, and the restructured navigation that follows aligns the site with how customers actually shop rather than how the merchandising team happens to think.
The Three Types of Card Sorting
Card sorting comes in three forms, and choosing the right one depends entirely on whether you are trying to discover a structure, validate one, or do both at once.
An open card sort asks participants to create their own categories and to name them. You hand over the cards and let participants group them however they like, then label each group in their own words. This is the discovery method. It is the right choice when you are building information architecture from scratch or when you want the deepest possible understanding of how users conceive of your content, because it imposes no structure and lets the user's genuine mental model surface. The labels participants invent are often as valuable as the groupings, since they reveal the vocabulary your users actually use, which frequently differs from your internal terminology.
A closed card sort flips this around. You define the categories in advance and ask participants to assign each card to your existing structure. This is the validation method. It is the right choice when you have already built a structure and want to confirm whether it works. The interpretation is direct: if a significant share of participants place an item in a different category than where you have it, your structure or your labels need work. As a practical reading, if sixty percent of participants put an item somewhere other than where you placed it, that is a strong signal that your labelling is failing them.
A hybrid card sort combines both. Participants sort cards into your predefined categories but retain the freedom to create new categories of their own when none of yours fit. This validates and explores simultaneously, which makes it valuable when you have some categories you are confident about but remain uncertain how the rest of the content should be organised. The cost of this richer insight is that hybrid sorts require more participants to produce reliable patterns, because the additional freedom introduces more variation in the responses.
There is also a methodological choice that cuts across all three types: whether to moderate. A moderated card sort places a facilitator alongside the participant, in person or online, who can ask why a participant grouped cards a certain way, clarify confusion, and probe the reasoning behind the structure. This adds qualitative depth. An unmoderated card sort lets participants work alone at their own pace, which sacrifices the real-time probing but allows far greater scale and speed.
How Many Participants You Actually Need
One of the most common questions about card sorting is how many participants are required to produce trustworthy results, and the answer depends on the type of sort and on whether you are running a qualitative or quantitative study. As a working set of guidelines, closed sorts typically need around fifteen to twenty participants, because validating an existing structure requires less data than discovering a new one. Open sorts need more, generally twenty to thirty, because the freedom to create categories introduces more variation that requires more responses to resolve into clear patterns. Hybrid sorts, with the most freedom of all, typically need twenty-five to forty participants.
A useful broader rule is that you generally need somewhere between twenty and fifty participants to see meaningful patterns, with the more diverse your audience and the more complex your content, the more participants you should include. For a purely qualitative study aimed at understanding rather than measuring, as few as fifteen participants can be sufficient, while a quantitative study aimed at statistically reliable agreement scores benefits from thirty or more. The encouraging reality is that these numbers are achievable quickly. Recruiting typically takes three to seven days, and each participant usually needs only ten to fifteen minutes to complete a sort, which means a single round of card sorting can settle navigation debates that a team has been having for quarters, in the span of a single week.
How to Run an Effective Card Sort
Running a card sort well begins with selecting the right content to test. Choose cards that represent the actual items users will navigate, written in clear, neutral language, and avoid overloading a single study with too many cards, since fatigue degrades the quality of sorting decisions. Randomising the order in which cards are presented is necessary to avoid biasing participants toward a particular grouping, though it is worth being aware that randomisation can occasionally confuse participants, which is one more reason to keep card labels clear and self-explanatory.
Recruiting a diverse group from your genuine target audience matters as much as the number of participants. A card sort completed by people who do not resemble your real users produces a mental model that is irrelevant to your actual product. Once the study is running, the mechanics are straightforward: participants sort the cards, and in moderated sessions the facilitator observes and asks clarifying questions where useful. The session length should match the number and complexity of cards, generally falling between five and thirty minutes.
Interpreting the Results
The output of a card sort is a set of patterns rather than a single answer, and reading those patterns is where the method's value is realised. The central tool is the agreement matrix, which shows how frequently any two cards were grouped together across all participants. High agreement reveals strong, shared mental models that your information architecture should respect. Low agreement, where participants scatter an item across many different categories, signals either an ambiguous item or a genuine divergence in how different user segments think, both of which deserve attention.
It is worth holding one important caveat clearly in mind. Card sorting tells you how users would organise things, but it does not tell you whether users can actually find things in the structure you eventually build. These are different questions. A structure that emerged from a card sort still needs to be validated for findability, and that is precisely what tree testing does. In a tree test, you give participants the proposed navigation structure and ask them to locate specific items within it, measuring whether they can actually reach the right destination. The two methods pair naturally: card sorting helps you design the structure based on how users think, and tree testing confirms whether the structure you designed actually works in practice. Running a card sort without a follow-up tree test leaves the most important question, can users find what they need, unanswered.
Card Sorting in a Modern Research Workflow
Historically, card sorting required a dedicated specialist tool, and the results lived in isolation from the rest of a team's research. Modern platforms have changed this. Card sorting now sits naturally within an integrated research workflow, where the same platform that runs the sort also runs the follow-up tree test, analyses the agreement patterns, and stores the findings in a searchable repository that connects to the team's other research. This integration matters because the insights from a card sort are most valuable when they can be cross-referenced with usability tests, surveys, and interviews that touch on the same navigation questions.
Fred supports all three forms of card sorting, open, closed, and hybrid, alongside tree testing, within a single platform. The agreement analysis is generated automatically, and because card sorting and tree testing live in the same workspace, the natural pairing of the two methods requires no exporting, no manual triage, and no stitching together of results from separate tools. A team can design an information architecture from an open sort, validate it with a tree test, and have both sets of findings flow directly into the same analysis and repository. For product teams that treat information architecture as a recurring concern rather than a one-time project, having these methods in one place turns card sorting from an occasional academic exercise into a fast, repeatable part of how the product evolves.
Card sorting deserves a place in every product team's toolkit precisely because it answers a question that no analytics dashboard can: not what users did within the structure you built, but how they would have built the structure themselves. Closing the gap between those two is among the most reliable ways to make a product feel intuitive, and a single week of card sorting can prevent quarters of navigation problems that would otherwise drive users away.
Fred supports open, closed, and hybrid card sorting plus tree testing in one platform, with automatic agreement analysis and results that flow straight into your research repository. EU-hosted and GDPR-native. Try it with a 15-day free trial, no credit card required. Start your 15-day free trial →
Related reading: