Tree test

What it is

The technique was formalised in the mid-2000s and popularised by tools like Treejack (now Optimal Tree). The tree is presented as a nested menu; participants read each task and click through until they reach the item they think answers it, or give up. The tool records the path, the time, and whether they picked the right end-point.

The core metrics are success rate (did they land on the right item?), directness (did they go straight there or backtrack?), and first-click accuracy (did the first top-level click head toward the right branch?). First-click accuracy predicts overall success remarkably reliably — Spool's research found a correlation of around 2× over chance.

Why it matters

An IA that looks sensible to the team can still fail users. Tree testing isolates the IA from visual design, copy and search — if users can't find content in the tree, they won't find it in the finished product either. It's the cheapest, fastest way to catch structural problems before you build.

When to use it

After card sorting (to validate the structure that emerged), before IA changes to a live site, during consolidation or migration projects, and whenever navigation is being redesigned. Not useful for validating individual pages or small content additions — tree testing is a navigation-level tool.

What it looks like in practice

A typical tree test uses 10–20 tasks with 30+ participants. Tasks should mirror real user intents expressed in user vocabulary (not internal jargon). Results are usually visualised as pie charts (where did people land?) and path diagrams (how did they get there?).

Pair with card sorting to design the structure, then with first-click testing on the final visual design to validate the visual hierarchy supports the IA.

Below 70% success on a task is a signal the IA is wrong for that content. Between 70–85% is "needs attention". Above 85% is fine.

What people get wrong

• Writing tasks in internal vocabulary. If the task says "Access the SKU-level reporting dashboard" and users would search for "product sales", the test is measuring your task-writing, not your IA.
• Testing too few tasks. 5 tasks don't cover enough of the navigation to surface structural problems.
• Over-interpreting single-task failures. One failing task may be a labelling problem; a cluster of failing tasks points to a structural problem.
• Testing a tree that doesn't exist on the final site. Only test what you'll actually build.

Learn next

Tree testing is the quantitative companion to card sorting and the validation step for any IA change. Related to first-click testing (narrower scope) and findability studies more broadly.

Reviewed 1 October 2026 · Editorial standards