What is card sorting in UX?

How a card sort works

You write each piece of content or feature on a separate card: "Returns policy", "Track my order", "Gift cards". Participants then sort the cards into groups that make sense to them. Afterwards you compare the groupings across people and look for patterns.

The point is not to find one correct answer. It is to see where most people agree, where they split, and which items nobody knows what to do with. Those orphan cards are often the most useful finding.

Open, closed and hybrid sorts

In an open sort, participants make up their own groups and name them. Use it early, when you do not yet know what the categories should be. It surfaces the words people naturally use.

In a closed sort, you give them the category names and they drop cards into them. Use it later, when you already have a draft structure and want to see if it holds up.

A hybrid sort lets people use your categories but also create new ones when nothing fits. It is a decent middle path when you have a rough idea but do not fully trust it.

Running one step by step

A basic study follows these steps:

• Pick 30 to 60 items. Fewer than that gives you little to look at, more makes people tired.
• Write each item in plain language, not internal jargon, and avoid starting several cards with the same word because people will group by it.
• Recruit people who resemble your real users, not your colleagues.
• Run it in person with paper, or remotely with an online tool.
• Ask people to think aloud, since the reasoning matters as much as the piles.
• Look for clusters that appear again and again, then use them to draft a structure.

A handful of participants gives you early signals. For a large or complex site you will want more, especially if you plan to count agreement.

What it will not tell you

Card sorting reveals mental models, but it does not prove your navigation works. People sorting cards at a table are not hunting for something under time pressure on a phone. That is why teams usually follow it with a tree test, where participants try to find items in a proposed structure.

It also cannot choose the top level labels for you. Participants often produce vague group names such as "Other" or "Info", and you will still need judgment and a clear content strategy to turn raw piles into a menu.

Mistakes worth avoiding

Using your own org chart as the starting point is the big one. A company may have a "Solutions" team and a "Services" team, but customers see one thing. Another trap is writing cards that give away the answer, like "Billing: invoices" next to "Billing: refunds". And finally, treating a small sample as gospel. If three people disagree, you have a question, not a verdict.