SEO & Search

Tree Testing for B2B Navigation: Check the Route Before Building the Menu

A menu can look settled while its labels still leave the destination unclear.

The design team has chosen a typeface. The header fits the screen. Nobody has checked where a buyer would look for permission requirements.

Put the hierarchy into text before investing further in the interface. Give someone a reason to find a resource. Record the branch they choose and where they stop.

Nielsen Norman Group describes tree testing as evaluating a hierarchy through resource-finding tasks. It isolates categories and labels from visual design and content. The proposed worksheet below applies that method to a fictional B2B library. Every task, participant, path, and outcome is invented. No customer research or website improvement is claimed.

Name the Decision the Test Will Inform

Write the decision in one sentence: should permission requirements sit within setup, or does the proposed route leave prospective users looking elsewhere?

Include the entire candidate hierarchy in the study. Participants must be able to explore neighboring branches. Showing only the intended branch removes meaningful alternatives.

Use the information architecture audit when the starting question concerns an existing site's structure. Here, the work begins with a candidate hierarchy that the team can still revise.

Freeze a version before collecting observations. Give it a short identifier such as T01. Record every label and parent-child relationship. Save the task wording beside that version so a later reviewer can reconstruct what the participant saw.

Do not rename a category during a session to help someone finish. A changed label creates a changed condition. Keep the original result and prepare a separate revision for a later exercise.

Build a Fictional Resource Tree With Complete Destinations

This example concerns a file-transfer service. Its resources are public instructional topics. It contains no customer accounts, confidential files, or actual product commitments.

The tree has four top-level branches and twelve leaf destinations. Each leaf represents a proposed resource, not a functioning application control.

Resources
  Evaluate
    Supported file formats
    Example import file
    Explanation of file fields
  Set up
    Required account permissions
    Connection requirements
    First transfer checklist
  Operate
    Error message reference
    Recovering a failed transfer
    Stopping scheduled transfers
  Manage agreement
    Understanding the monthly bill
    What each service package includes
    Service agreement terms

The expanded tree above is the researcher's inventory. In this proposed exercise, participants first see only the four top-level branches. Reveal a branch's children when they open it. For a paper session, use separate sheets and let participants return to the branch list. Reset that view before each task, while recording any exposure from earlier tasks. NN/G describes expandable categories as the tree-testing presentation. Showing every leaf at once changes this exercise into a scan of the full inventory.

Keep the tree deep enough to reach the intended answer. Choosing “Set up” alone cannot establish whether the participant distinguishes permissions from connection requirements.

Check that each leaf has a defined meaning in the inventory. In this example, permissions describe access needed for a transfer. Connection requirements describe the technical prerequisites. Do not let an unexplained overlap become a hidden scoring rule.

Leave search boxes, promotional links, page excerpts, and button styling outside this exercise. Record that boundary in the study brief. Those features need their own evaluation in the functioning interface.

Write Tasks Without Supplying the Path

NN/G advises avoiding terms that reveal the answer and defining the acceptable destination for each task. Use that distinction when preparing the participant and researcher versions.

The participant receives the situation and information need. The researcher retains the destination key. Keep both versions short enough to read without searching through a story.

For example, “Find required account permissions under Set up” repeats the destination and its parent. Replace it with a task about the decision someone needs to make.

Use the following four synthetic tasks for the fictional tree:

Task Participant wording Intended leaf
T1 Before choosing this service, you want to inspect a sample of the file you would send. Where would you look? Example import file
T2 A colleague will run the first exchange. Find what access their login needs beforehand. Required account permissions
T3 Yesterday's exchange stopped before finishing. Find instructions for getting that same exchange through. Recovering a failed transfer
T4 You want future automatic exchanges to stop while keeping the agreement in place. Where would you look? Stopping scheduled transfers

These prompts avoid repeating the branch names or the full leaf labels. They still contain task language. Review whether that language describes the participant's work or merely disguises the menu wording.

Ask a pilot participant to explain the task in their own words before interpreting a surprising route. If they read T3 as canceling future activity, task comprehension needs investigation alongside the hierarchy.

Define the Recording Rules Before the Pilot

Use a small qualitative pilot to identify questions for revision. Define its intended audience, recruitment limits, session format, and assistance rules before interpreting any results.

For this proposed exercise, record the first top-level branch opened. Record the leaf selected as the answer. Preserve any return to an ancestor and the next branch opened. Keep an unfinished task unfinished.

Use “direct” only when the participant reaches the intended leaf without returning to an earlier branch. Use “indirect” when they reach it after a return. These are worksheet definitions, not universal labels across research products.

Keep the raw path even when the software supplies a summary. Write the tool name, version if available, imported tree, navigation controls, and any behavior you could not verify. If the tool omits intermediate actions, mark backtracking unknown.

Separate moderator help from ordinary navigation. If the researcher explains a category or suggests a branch, mark the outcome assisted. Do not silently count an assisted destination as independent discovery.

For the fictional paper exercise below, a slash means moving deeper into the tree. An arrow means returning to the root and opening another branch. The researcher reads no hints.

Use a Task-and-Path Worksheet

Keep a study header with tree version, task version, participant code, relevant task experience, and session conditions. Use one row for each attempted task.

Field What to record
Task ID Exact prompt version and intended leaf
First branch First category opened, or none
Raw path Every opened branch and selected leaf
Final destination Selected leaf, stopped, or unknown
Backtracking Recorded returns, or unknown
Explanation Participant account, identified as paraphrase
Assistance Hint or clarification, with its timing
Interpretation Researcher's proposed explanation

Keep the participant account distinct from the interpretation. Someone choosing “Manage agreement” does not prove they misunderstood the word “permissions.” Their explanation may concern billing authority, responsibility, or a mistaken reading of the task.

Record task order. If later tasks reuse a branch the person has already explored, preserve that exposure in the notes. Do not treat the later path as an untouched first encounter.

Record pauses without assigning a cause. A participant may pause to reread, think, or handle an interruption. If elapsed time matters to a later study, define timing and exclusion rules separately.

Read a Fictional Qualitative Pilot Without a Benchmark

The following worksheet contains six invented task attempts. Participants P01 and P02 each attempt T1, T2, and T3 in that order. T4 remains untested in this teaching example.

This is a paper pilot with researcher-recorded paths. It offers no population estimate, performance benchmark, or validated tool measurement. All explanations are fictional paraphrases.

Person/task First branch Final leaf Returns Explanation
P01/T1 Evaluate Example import file None Wanted a sample before choosing
P01/T2 Manage agreement Required account permissions One Initially associated access with account ownership
P01/T3 Operate Recovering a failed transfer None Looked for a way to finish yesterday's work
P02/T1 Set up Example import file One Expected a sample beside preparation instructions
P02/T2 Set up Required account permissions None Treated login access as a prerequisite
P02/T3 Operate Error message reference None Wanted to identify the error before choosing a remedy

The raw paths behind those summary rows are:

  • P01/T1: Evaluate / Example import file.
  • P01/T2: Manage agreement → Set up / Required account permissions.
  • P01/T3: Operate / Recovering a failed transfer.
  • P02/T1: Set up → Evaluate / Example import file.
  • P02/T2: Set up / Required account permissions.
  • P02/T3: Operate / Error message reference.

The two indirect paths are P01/T2 and P02/T1. Each returns to the root once before reaching the intended leaf.

The other four attempts stay within their first branch. P02/T3 ends at a different leaf from the answer key. That mismatch deserves a note even though the chosen reference could plausibly help with a real failure.

Record the mismatch without rewriting the key after seeing the answer. Review whether the prompt asks for diagnosis or recovery. A vague task can produce an ambiguous interpretation without establishing a category defect.

Do not turn these six fictional rows into a success percentage. They mix repeated participants and different tasks. In a real qualitative pilot, retain the sampling and session limitations before discussing any completion measure. NN/G also distinguishes qualitative findings from quantitative benchmarking.

Write a Revision Decision With a Testable Reason

Use a decision record to keep proposed changes attached to their evidence. A requested label change and an observed navigation obstacle should occupy different fields.

Field Fictional entry
Observation P01 first opened Manage agreement for T2
Explanation Associated login access with account ownership
Hypothesis The setup route may not signal access prerequisites clearly
Candidate revision Examine “Prepare access and connections” as a branch label
Alternative Keep Set up and clarify the resource label
Open question Does the revised label confuse people seeking the first-transfer checklist?
Next check Run neutral permission and checklist tasks on the revised tree
Status Proposed, not accepted

Keep the alternative visible. One participant's detour does not establish the best repair. A longer parent label could clarify one resource while obscuring its neighbors.

Version the revised tree as T02. State exactly which label or relationship changed. Avoid changing several unrelated branches before checking the disputed route. Otherwise the next observation leaves the cause of any difference harder to examine.

If comparing alternatives, plan the comparison before recruitment. NN/G recommends giving each participant one tree version to avoid learning from an earlier tree. A later session with the same person needs an explicit exposure limitation.

Hand the Hierarchy to the Interface Team With Its Limits

Deliver the versioned tree, exact tasks, destination key, raw paths, participant explanations, study limitations, and revision record. Name the untested T4 route. Do not let an empty worksheet row look like a completed check.

Then evaluate the proposed hierarchy inside the actual website. A text tree does not test keyboard interaction, menu behavior, layout, page comprehension, or the visibility of a link. Tree-test completion establishes no general website usability result.

Keep search evidence separate. Internal site search analysis examines submitted queries and available answers. No-results page planning examines recovery when search yields no match or fails. Neither problem is resolved by a plausible text hierarchy.

The next decision should name the evidence still required. If the access route remains unclear, revise and test that route. If the text route works but the menu is difficult to operate, inspect the interface behavior.

Check the route before building around it.

Sources

Topic map: Customer and usability research

Articles in this cluster, then the 3 clusters that share the most topics with it.

This cluster: Customer and usability research

Neighbor: Topical authority and keyword strategy

Neighbor: Content production with AI

Neighbor: Technical SEO, crawling, and migrations