Tree testing

Tree testing presents a structure as plain nested text and asks participants where they would look for a specific item. No layout, no visual design, no search: just labels and hierarchy.

Stripping the interface away is the point. It isolates whether the structure and its labels work, which a full prototype cannot do, because a striking visual design can carry a user past a structure that does not make sense.

The complement to card sorting

Card sorting generates a structure from how users group things. Tree testing validates whether a proposed structure can be navigated. They answer opposite questions, and running only the first is the common half-measure: a structure derived from sorting still has to be tested against real tasks.

What it measures

Success rate per task, whether the correct answer was reached directly or after backtracking, and where people went instead. That last one is the most useful output, because a task failing consistently into the same wrong branch tells you exactly which label is misleading.

Practicalities

It is cheap and unmoderated, so samples of 30 to 50 participants are realistic, which gives findings more weight than a handful of moderated sessions.

In practice

Write tasks as goals rather than as labels. “Find out how much it costs to return an item” tests the structure; “Find the Returns Policy page” tests whether people can read.