When a website grows, the problem is no longer writing content but making it findable. Menu items appear one at a time, often mirroring the company's org chart, and after a few years nobody knows whether the refunds page lives under «Support», «Orders» or «My account». Card sorting and tree testing are two complementary techniques that help build a structure based on how users think, not on how the publisher is organized.
Card sorting
In card sorting, the site's topics are written on cards, physical or digital, and participants are asked to group them in a way that makes sense to them. The goal is to uncover mental models: which content people see as related and what they would call the groups.
Open, closed, hybrid
- Open: participants create the groups and name them. Useful at the start, when no structure exists or it needs rethinking from scratch.
- Closed: categories are predefined and participants only place the cards. Used to check an existing structure.
- Hybrid: some categories are offered but participants can create new ones. Often a practical compromise.
Preparing the cards
The quality of the cards determines the quality of the results. Some rules:
- pick between thirty and sixty representative topics: fewer are too few to show patterns, more make the task tiring;
- word every card in plain language, avoiding internal jargon;
- avoid having several cards share the same keyword: if three labels start with «Management», participants will group them by wording, not by meaning;
- if needed, add a short description for ambiguous terms.
Analyzing the results
With fifteen or twenty participants, reliable trends start to appear. Online tools usually produce a similarity matrix, showing how often two cards were placed in the same group, and a dendrogram showing clusters at different levels.
But numbers are not enough. Reading the names participants gave to their groups is often the most instructive part: it reveals the real vocabulary. If many call «Bills» what the company calls «Accounting documents», the menu label has a better candidate.
Tree testing
Card sorting helps build a structure; tree testing checks it. Participants see only the navigation tree, as text, with no visual design, no search and no content, and are asked where they would look for a given piece of information.
A typical task is: «You received a damaged product and want to return it. Where would you go?». The participant explores the tree by clicking items until choosing the one they believe is right.
What is measured
| Measure | What it shows |
|---|---|
| Success rate | How many reach the correct item |
| Direct path | How many get there without backtracking |
| First click | Where most people go at the first step |
| Time | How hard it was to decide |
The first click is especially revealing: if most people start in the wrong section, the problem lies in the top-level label, and the whole rest of the path suffers.
Writing the tasks
As in usability tests, tasks must not contain the menu's words. «Find the Returns section» measures nothing; «The sweater you bought is too small, what do you do?» does. Eight to ten tasks per session are enough, focused on the most important information and on what you suspect is hard to find.
A complete cycle
An effective process combines the two techniques:
- inventory of existing content;
- open card sort to understand spontaneous groupings;
- proposal of one or two alternative structures;
- tree test on each, comparing results;
- fixes and a second tree test on the improved version;
- only then, visual design of the navigation.
Separating structure and visuals is the main advantage: it prevents a beautiful menu from hiding a confusing organization, or a solid organization from being judged poorly because of a temporary style.
Limits to keep in mind
Card sorting tells you how people group content, not how they navigate. Tree testing tells you whether they find items, not whether they understand the page once they arrive. Neither technique replaces the site's internal search, which for many users remains the first choice: a clear architecture and good search complement each other.
In short
- Card sorting reveals how users group and name content.
- Tree testing checks, without visuals, whether a structure lets people find information.
- The first click and the spontaneous group names are among the most useful data.
- Structure and visual design should be designed and tested separately.
