Tree testing, end to end
By UXbeam, information architecture research since 2021 · Updated July 28, 2026
Tree testing checks whether people can find things in your navigation structure, before you invest in design and build. This page covers the method itself and how UXbeam supports each step, from writing tasks to iterating on results.
What tree testing is, and when to run one
The idea
Tree testing shows whether key information, features, and content sit where users look for them, before anything is designed. You give participants your structure (labels in a hierarchy, stripped of visual design) plus a realistic task ("You want to change how often you get email updates"), and watch where they go.
When to use it
Before a redesign or IA change, when deciding between candidate structures, after a card sort has suggested groupings you now need to validate, or when a live site shows symptoms like heavy search use and support tickets asking "where do I find…".
What you learn
Per-task findability evidence: who went straight to the right place, who wandered and recovered, and who ended up somewhere else entirely, plus first clicks and the actual paths people took.
Tree testing fits best early in the design process, right after you have gathered (and sometimes card sorted) the requirements from stakeholders, users, and competitors. Waiting until the hi-fi prototype or development stage makes navigation problems far more costly and complex:
- the fixes are slow, since the team may have to remove or rearrange screens and relabel paths;
- the problems tangle with other design issues like layout, contrast, and illustration.
The 3 ingredients of a tree test
The tree
Write the tree as an outline of categories and subcategories, naming and connecting the screens or pages that make up an early prototype or an existing site you plan to redesign. Done thoroughly, that outline doubles as a planning tool. It shows how many screens and features you will need to wireframe, design, and build, and it gives an early sense of how long and involved your main user journeys are.
The tasks
Write a short set of task instructions, each asking the participant to work down the tree to find a page or a location. Cover both the destinations people clearly expect for this kind of product (for a photo-sharing app, save a photo to a folder), and the ones unique to your product (schedule your vacation photos to be shared at the end of the quarter). Each task instruction should be:
- Short: no more than two clauses.
- To the point: a scenario written to sound natural can pull participants in conflicting directions. A task with several focal points is as flawed as a double-barreled survey question, like "how likely are you to buy or borrow this app?"
- Non-leading: avoid reusing words that appear in the tree.
| Category | Task 1 | Task 2 | Task 3 |
|---|---|---|---|
| Wellness mobile app (B2C) | You've meditated every day this week and want to check how many days in a row you've kept it up. | You just finished a workout and want to record what you did so it is saved for later. | You keep going to bed too late and want the app to nudge you at a set time each night. |
| Field service management mobile app (B2B) | You've just finished a service call and want to mark it done so your manager can see it is handled. | An urgent ticket comes in and you need the customer's address and contact before heading over. | You are short a part mid-repair and need to order a replacement to finish the job. |
| Ride-sharing mobile app (B2C) | You need to get to the airport and want to see how soon a car could reach you. | Your meeting ran long and you just want to get home. | You have a lot of luggage and want a car with extra space. |
| CRM platform web app (B2B) | You just got off a sales call and want to jot down what was said before you forget. | A prospect replied, and you want to record that they are now further along. | You have a review meeting coming up and want a summary of this quarter's numbers. |
| Online retail web app (B2C) | You want to find a gift for under $25. | You've read the reviews and want to know whether the item is still available before buying. | You want to give a friend something from this store but let them choose it themselves. |
| Streaming media web app (B2C) | You are in the mood for a thriller and want to see what the app suggests. | You stopped an episode halfway yesterday and want to pick up where you left off. | You've found a show you want to watch later and want to save it for then. |
| Driver's license renewal portal (Gov) | You have just turned 18 and need to renew your license. | You have moved recently and want to make sure the address on file is your new one. | You got a reminder that your license is expiring and want to book a time to sort it out. |
| City maintenance request portal (Gov) | There is a pothole on your street and you want to report it. | A streetlight on your road is out and you want to send a photo so it gets fixed sooner. | There is graffiti on a nearby wall and you want to ask the city to clean it up. |
The participants
Recruit a representative audience: people who match the profile of your real users. Tree tests are quick and unmoderated, which makes them well suited to quantitative analysis, and clear patterns start to emerge between 10 and 50 participants.
The UXbeam workflow
Here is how those steps look inside UXbeam.
-
Bring your tree
Paste an indented text outline, upload a CSV or TXT export (delimiters and indentation are detected automatically), or import a structure straight from a UXbeam card sorting study. Malformed indentation can be repaired without changing any label text.
-
Write tasks and mark correct destinations
Correct answers live in the tree itself: you mark destination nodes, and a task can have several valid destinations, the same way real sites often have more than one legitimate place to complete a task. UXbeam recommends about 10 tasks, with a maximum of 20. Beyond ten or so, people tire and start skipping or rushing answers just to finishO'BrienTree Testing for WebsitesHow many tasks?. For how to phrase a task so it tests findability and not reading, see the Job-Shaped Task.
-
Check the setup before launch
Before launch, you can run an optional check on your task wording. If a task repeats a label from your tree, especially a top-level menu label, it gets flagged as potentially leading, with suggested rewrites that avoid every word in your tree. Why that matters →
-
Test with your participants
Share the participant link with your own users, customers, or a panel you already work with. It works in any browser with no account and no install, on desktop or mobile.
-
Read results and iterate
Outcome breakdowns, first clicks, top destinations, and full path exploration per task. When you iterate, UXbeam pre-selects the tasks that need work (those where fewer than half of participants went direct, or where wording was flagged) and carries your tree and tasks into the next version.
Reading tree test results
Start with the success rate: the share of participants who reached a correct destination, direct or indirect. Albert and Tullis, reviewing 98 tree tests, set a rough scale for reading it:
| Above 90% | Excellent |
|---|---|
| 80-90% | Very good |
| 61-80% | Good |
| 41-60% | Fair |
| Below 40% | Poor |
The median across those studies was 62%, so anything in the "good" band already beats most tasks. Hold mission-critical or revenue tasks to a higher bar, above 90%.
A success rate alone doesn't show how people got there. Every attempt lands in one of three outcomes:
| DIRECT | Reached a correct destination without using Back or starting over. The strongest signal your structure works. |
|---|---|
| INDIRECT | Reached a correct destination, but backtracked along the way. Success, with hesitation worth understanding. |
| ELSEWHERE | Confidently ended somewhere other than a correct destination. Often the most instructive outcome: it shows where your labels actually send people. |
UXbeam also shows the top first clicks per task (where intuition pointed people first), the top destinations they ended at, grouped navigation paths, a per-task path explorer over your full tree, and a per-participant table.
The questions to ask, and where to look
- Did people head down the right path first? Look at the first clicks. The first click is the strongest single predictor of whether someone eventually succeedsBailey & WolfsonMeasuringUThe Importance of the First Choice in Website Navigation, so it is the thing to optimize. A weak or misdirected first click is the earliest warning you get.
- Where did people expect the content to be? Look at the top destinations. When a lot of people confidently land somewhere that is a fair home for the task, that is a signal to mark it as a correct alternate destination, not to write it off as a miss.
- Which labels are muddying the choice? Look at how the first clicks spread. When they scatter across more than three places, people have no clear scent to follow, and the task wording or the top-level labels are what to rework.
Results also re-check task wording: if participants converged on a destination whose label appeared in the task text, UXbeam flags it, because that convergence may reflect word-matching rather than findability.
From card sorting to a testable tree
Tree testing validates a structure; card sorting helps you discover one. In UXbeam the two connect directly: run a card sort, review how participants grouped and named things, shape those groupings into a sitemap in the built-in tree editor, then send that sitemap straight into tree testing setup with one click.
Taking results out of UXbeam
Participant CSV
Every participant × task outcome in long format, ready for Excel or your own analysis.
Tree & tasks text file
Your full structure and task set as plain text, for tickets, wikis, and version history.
Sitemap screen cards
A ZIP of SVG menu cards, one per section of your tree, for walking stakeholders through a proposed structure.
Frequently asked questions
How many tasks should a tree test have?
UXbeam recommends about 10 and enforces a maximum of 20. Additional tasks increase participant fatigue and drop-off, which degrades the data for every task after the fatigue sets in.
How do I recruit participants?
UXbeam gives you a study link to share with your users, customers, or an external panel you already work with. Participants don't need an account, and testing with your own participants isn't capped.
Does it work on mobile?
Yes. The participant experience is built for phones as well as desktop, with full-width touch targets and a compact navigation layout on small screens.
Can a task have more than one correct answer?
Yes. You can mark several destinations as correct for the same task, and one destination can serve several tasks, matching how real information architectures work.
Can participants stay anonymous?
Yes. The participant ID field is optional: use it when you want to match results to a panel ID or email, skip it when you don't. Study links use encrypted identifiers rather than raw database IDs.
Can I randomize task order?
Yes. Task order can be randomized per participant to reduce order effects, while your results still map back to the original task list.
How does iteration work?
Each version of a study keeps its own results. When you start the next version, UXbeam carries over your tree and tasks and pre-selects the tasks that need attention (those below 50% direct or flagged for leading wording). Version comparison highlights where a new structure measurably improved.
References
- Dave O'Brien, "How many tasks?", in Tree Testing for Websites (companion wiki).
- Jeff Sauro, "The Importance of the First Choice in Website Navigation", MeasuringU, reporting Bob Bailey and Cari Wolfson's first-click research.
- Bill Albert and Tom Tullis, Measuring the User Experience, Morgan Kaufmann. Task-success rubric and the 98-study tree-testing benchmark as reported by Nielsen Norman Group.