Card sorting for beginners: build a menu your customers understand

Every argument about a website menu ends the same way: two people are sure their grouping is obvious, and neither has asked a customer. Card sorting settles it cheaply. You write each page name on a card, hand the pack to people who resemble your customers, and watch how they file them. What comes back is a map of how your audience expects information to be arranged.

It needs no software, no budget and about two hours in total for a small site. This is how to run one from scratch.

Open, closed or hybrid: pick the right kind

An open sort tells you what the groups should be

Participants receive the cards with no categories and create their own piles, then name each pile themselves. Use this when you are starting fresh, restructuring after growth, or genuinely do not know what to call things. The names people invent are often better menu labels than anything written internally.

A closed sort tests groups you already have

You supply the category names and participants place each card into one of them. Use this when the top-level menu is fixed, perhaps because it matches how the business sells, and you need to know where individual pages belong.

A hybrid sort sits between the two

Participants use your categories but may add their own if nothing fits. The extra categories people create are the interesting result, because they show gaps in your structure.

For a first attempt on a small business site, an open sort gives the most useful information per hour spent.

Preparing the cards

Write one card per page or topic, not per menu item. Keep between 30 and 50 cards for a small site; below 20 the exercise is too easy to be informative, and above 60 people start to tire and rush.

Card wording matters more than anything else in the setup:

• Use the customer's likely words, not internal names or product codes.
• Keep every card at a similar level of detail, so a card for one small page does not sit beside a card for an entire section.
• Avoid category-sounding words on cards, since a card called "Support services" nudges people into creating a support pile.
• Do not include the home page, contact page or policy pages, as their position is not in question.
• Add a short explanatory line where a title alone is ambiguous.

Print them on paper, cut them up, and bring a stack of blank cards for pile names and any topics participants say are missing.

Finding participants without a research budget

You need people who resemble buyers, not colleagues who already know the structure. Existing customers are ideal and are often happy to help for ten minutes. Suppliers, neighbouring business owners and friends in the same industry work too. Staff who joined in the last month are the best internal option, since they have not yet absorbed the company's internal logic.

Numbers are smaller than people expect. Around twelve to fifteen participants is enough for patterns to stabilise for a typical small site, and even five will expose obvious problems. Run sessions individually rather than as a group, because in groups the loudest voice becomes the answer.

Running a session

Allow fifteen to twenty minutes per person and follow the same script each time:

1. Explain that you are testing the website, not the participant, and that there are no wrong answers.
2. Shuffle the cards and hand over the pack.
3. Ask them to sort the cards into groups that make sense to them, setting aside anything they do not understand.
4. Stay quiet while they work. Resist explaining what a card means, because the confusion is data.
5. When they finish, ask them to name each pile on a blank card.
6. Photograph the layout, then ask two questions: which cards were hardest to place, and was anything missing?

Record the pile names and contents in a spreadsheet straight after each session, one column per participant. Doing it later from photographs takes twice as long.

If your customers are spread across cities, the same exercise works over a video call with a shared online whiteboard, or with one of the dedicated card sorting tools that run sorts remotely. The analysis is identical.

Reading the results

You are looking for three things.

Cards that travel together. Build a simple grid with every card as a row and column, and count how often each pair landed in the same pile. Pairs that co-occur in most sessions are your natural groups. This sounds laborious but takes about half an hour for 40 cards, and dedicated tools do it automatically.

Cards that scatter. A card that appears in five different groups across twelve participants is either badly named or genuinely belongs in two places. Rename it first and, if it still scatters, plan to link it from more than one part of the site.

Names people chose. Tally the pile names. If eight people out of twelve wrote some version of "Prices" and nobody wrote "Investment", you have your label. Watch for a group where names vary wildly, which usually means the grouping is yours rather than theirs.

Look for agreement, not unanimity. Patterns shared by roughly two-thirds of participants are strong enough to act on.

Turning the sort into a real menu

The result of a sort is an information structure, not a finished menu. Convert it carefully:

• Merge groups that clearly overlap, and split any group that grew beyond about eight cards.
• Choose the label people used most often, keeping it to one or two plain words.
• Decide which groups earn a top-level slot and which live inside another, remembering that a menu with more than about seven top-level items is hard to scan.
• Keep cards that scattered accessible from several routes rather than forcing them into one home.
• Place business necessities such as contact and policy links yourself, since participants never sorted them.

Once the draft structure exists, validate it before building: give a handful of people a task such as "find out what an annual maintenance contract costs" and ask which menu item they would click. Any task where most people head down the wrong branch needs another look.

Making the exercise a habit

Run a sort before a redesign, before adding a new product line, and whenever enquiries suggest people cannot find something. The cards from last time can be reused with the new pages added, which turns a two-hour exercise into a forty-minute one.

If you would rather have the structure planned for you, our team runs this kind of exercise while mapping pages for a new build; see how we approach website design and development before you commit to a sitemap.

Frequently asked questions

How long does a card sorting exercise take in total?

For a small business site, allow about an hour to write and print the cards, fifteen minutes per participant, and one to two hours for analysis. Most teams complete the whole exercise within a week alongside normal work.

Can I run a card sort with my own staff instead of customers?

You can, and it is better than guessing, but staff carry internal knowledge that customers do not have. Treat a staff-only sort as a starting draft and confirm it with at least a few customers before rebuilding the menu.

What if participants create completely different groupings?

Wide disagreement usually means the card labels are unclear or the content genuinely spans several audiences. Rewrite the vague cards and repeat with a few more people, and consider whether different audiences need separate entry points on the site.

Do I need paid software to do this?

No. Paper cards and a spreadsheet handle everything described here. Paid tools mainly save time on analysis and make remote sessions easier, which becomes worthwhile for large sites with hundreds of pages.