) in which study participants group individual labels written on notecards according to criteria that make sense to them. This method uncovers how the target audience’s domain knowledge is structured, and it serves to create an information architecture that matches users’ expectations.
Part of making a site easy to use is organizing information so that people find what they’re looking for. Too often, content is structured based on what makes sense to the company, not to the users. One of the primary ways to figure out an organization scheme that best matches users’ mental model is through card sorting.
Card sorting is a highly useful technique in information architecture; it is used to understand how users think about your content.
It can help you organize content so that it suits your users’ mental models, rather than the point of view of your company. Card sorting can be supplemented with other information-architecture methods to identify issues in your category structure.
Let’s imagine that you’re designing a car-rental site. Your company offers around 60 vehicle models that customers can choose from. How would you organize those vehicles into categories that people can browse to quickly find their ideal car rental? Your company might use technical terms such as family car, executive car, and full-sizeluxury car. But your users might have no idea of the difference between some of those categories. This is where card sorting can help: ask your users to organize vehicles into groups that make sense to them, and, then, see what patterns emerge.
Hertz.com: In recent user testing on ecommerce websites, participants saw a dropdown list of Rental Car Types, but they weren’t sure what categories such as Dream Cars or Prestige Collection meant. Fortunately, the site included a photo and a brief description of each category, but comparing the different car types still required a fair amount of effort. Card sorting can reveal what kinds of cars users expect to find on a car-rental site.
Conducting a Card Sorting
Generally, the process works as follows:
Choose a set of topics. The set should include 40–80 items that represent the main content on the site. Write each topic on an individual index card.Tip: Avoid topics that contain the same words; participants will tend to group those cards together.
User organizes topics into groups. Shuffle the cards and give them to the participant. Ask the user to look at the cards one at a time and place cards that belong together into piles. Some piles can be big, others small. If the participant isn’t sure about a card, or doesn’t know what it means, it’s ok to leave it off to the side. It’s better to have a set of “unknown” or “unsure” cards than to randomly group cards.Notes:
There is no preset number of piles to aim for. Some users may create many small piles, others may end up with a few big ones. It all depends on their individual mental models.
Users should be aware that it’s OK change their mind as they work: they can move a card from one pile to another, merge two piles, split a pile into several new piles, and so on. Card sorting is a bottom–up process, and false starts are to be expected.
User names the groups. Once the participant has grouped all the cards to her satisfaction, give her blank cards and ask her to write down a name for each group she created. This step will reveal the user’s mental model of the topic space. You may get a few ideas for navigation categories, but don’t expect participants to create effective labels.Tip: It’s important to do this naming step after all the groups have been created, so that the user doesn’t lock herself in to categories while she’s still working; she should be free to rearrange her groups at any moment.
Debrief the user. (This step is optional, but highly recommended.) Ask users to explain the rationale behind the groups they created. Additional questions may include:You can also ask the user to think out loud while they perform the original sorting. Doing so provides detailed information, but also takes time to analyze. For example, you might hear the user say, “I might put card Tomatoes into pile Vegetables. But wait, they are really a fruit, they don’t really fit there. I think Fruits is a better match.” Such a statement would allow you to conclude that the user did consider Vegetables a decent match for Tomatoes, even though Fruits was even better. This information could push you into crosslinking from Vegetables to Fruits or maybe even assigning the item to Vegetables if there are other reasons leaning in that direction.
Were any items especially easy or difficult to place?
Did any items seem to belong in two or more groups?
What thoughts do you have about the items left unsorted (if any)?
If needed, ask the user for more-practical group sizes. You should not impose your own wishes or biases upon the participant during the original sorting (steps 1–3), but once the user’s preferred grouping has been defined, and after the initial debrief, you can definitely ask the participant to break up large groups into smaller subgroups. Or the opposite: to group small groups into larger categories.
Repeat with 15–20 users. You’ll need enough users to detect patterns in users’ mental models. We recommend 15 participants for card sorting: with more, you’ll get diminishing returns for each additional user; with fewer, you won’t have enough data to reveal overlapping patterns in organization schemes.
Analyze the data. Once you have all the data, look for common groups, category names or themes, and for items that were frequently paired together. If you see that some items were frequently left off to the side, determine whether it’s because the card labels weren’t clear or the content seemed unrelated to the rest of the topics. Combine the patterns you see with your qualitative insights from the debrief, and you’ll be in a better position to understand what organization system will be most successful for your users. (There’s a lot more to the analysis of card-sorting results, but that’s a topic for another article.)
Remote Card Sorting
5 steps for conducting a card sorting study remotely, to discover how users group items together. This is useful when designing your IA.