Information Architecture Validation Reference
Purpose: Validate information architecture (IA) through card sorting, tree testing, and first-click testing. Covers open / closed / hybrid card sort design, tree testing (Treejack / PlainFrame), sample-size guidance for IA studies, dendrogram and similarity-matrix analysis, and first-click success thresholds.
Scope Boundary
- Field
cards: IA validation through structured sort / tree-test methods. Category generation, label testing, navigation-path verification, analysis of dendrograms and first-click paths. - vs Echo: Echo runs persona-based cognitive walkthroughs evaluating UI comprehension within a page or flow. Cognitive walkthrough of a UI → Echo; validating whether the navigation structure itself is findable →
cards. - vs Pulse: Pulse instruments in-product navigation events and measures search / nav funnel performance after launch. In-product nav analytics → Pulse; pre-launch IA validation with moderated or unmoderated participants →
cards. - vs Voice: Voice mines feedback for "I can't find X" complaints on existing products. Post-launch findability complaints → Voice; proactive IA validation before commit →
cards.
Method Selection
| Method | Answers | Use when |
|---|---|---|
| Open card sort | "How do users group these items?" | Early IA work, no existing categories |
| Closed card sort | "Do items fit our proposed categories?" | Validating an existing taxonomy |
| Hybrid card sort | "Do our categories work, and are we missing any?" | Refining a draft IA |
| Tree testing | "Can users find X in this structure?" | Validating a navigation tree |
| First-click testing | "Do users start down the right path?" | Testing specific find-tasks |
| Reverse card sort | "Does our structure match their mental model?" | Late-stage validation before commit |
Default progression: open sort → build hypothesis IA → closed or tree test → first-click on prototypes.
Sample Sizes for IA Studies
| Method | Minimum n | Recommended n | Rationale |
|---|---|---|---|
| Open card sort | 15 | 20–30 | Correlations stabilize around 30 (Tullis & Wood, 2004) |
| Closed card sort | 15 | 20–30 | Same correlation basis |
| Tree testing | 50 | 100+ | Task-based, binary success; needs statistical power |
| First-click testing | 20 | 30+ | First-click success correlates 80%+ with task success |
| Moderated IA interview | 5–8 | — | Qualitative depth, not quantitative |
For multilingual / multi-market IA: run separate studies per market; do not pool.
Card Sort Design
- Card count: 30–60 cards. Under 30 misses structure; over 80 causes fatigue and dropoff.
- Card wording: use real content labels, not feature names. "Reset password" beats "Authentication utility".
- Card granularity: consistent level of abstraction — mixing "Settings" with "Change notification sound" biases grouping.
- Participant brief: "Group these the way they make sense to you" — do not hint at domain expertise.
- Open-sort category cap: none, but flag participants who create >15 categories (often indicates disengagement).
- Closed-sort forced placement: always provide an "Unsure" or "Doesn't fit" bucket; forced placement corrupts signal.
Tree Testing
Tree testing (Treejack, PlainFrame, UserZoom IA) validates a navigation structure without visual design confounds.
| Metric | Threshold | Meaning |
|---|---|---|
| Success rate | ≥80% | Label and structure support the task |
| Directness | ≥70% | Participants found it without backtracking |
| Time on task | Context-dependent | Compare within study, not absolute |
| First-click success | ≥67% | Strongest predictor of overall task success |
Run 15–25 tasks per test, each targeting a specific finding. Randomize task order. Budget ~10–15 minutes total per participant.
First-Click Testing
First-click success is the single strongest predictor of overall findability — tasks where the first click is correct succeed ~87% of the time; wrong first click succeeds only ~46% of the time (Bailey et al., 2009).
- Show a static screen, give a task ("Find your billing history"), record the first click location.
- Analyze via click heatmap and success-rate per task.
- A task with <50% first-click success needs label or layout revision before further testing.
Analysis
Dendrogram (hierarchical cluster analysis)
- Output of open card sort: a tree showing how often pairs of items were grouped together.
- Read cut-points to identify natural category groupings.
- Actionable similarity threshold: typically ≥ 0.60 to 0.70 indicates strong grouping signal.
Similarity matrix
- Heatmap of co-occurrence: how often each pair of cards appeared together.
- Look for dense "blocks" indicating robust groups, and "bridge" cards that associate with multiple groups — those need careful placement or duplication.
Standardization grid (closed sort)
- Percentage of participants who placed each card in each category.
- ≥75% agreement: category is clear.
- 50–75%: ambiguous — investigate label or card wording.
- <50%: card does not belong; rework.
Tool Selection
| Tool | Best for | Notes |
|---|---|---|
| Optimal Workshop (OptimalSort, Treejack) | End-to-end card sort + tree test | Industry standard, full analytics |
| UserZoom / UXtweak | Enterprise IA studies | Strong panel integration |
| Maze | Tree test + first-click + prototype test | Fast, integrates with Figma |
| Miro / FigJam + spreadsheet | Moderated card sort, small n | DIY, cheap, no auto-analysis |
| PlainFrame | Lightweight tree testing | Minimal setup |
Anti-Patterns
- Running tree testing with <50 participants — insufficient for meaningful success-rate CIs.
- Using feature-team language on cards — participants cannot group what they don't recognize.
- Skipping first-click before full usability testing — a wrong first click means you are measuring recovery, not findability.
- Pooling multilingual results — label semantics differ per locale.
- Accepting 60% success as "good enough" for primary navigation — industry norm is ≥80% for top-level tasks.
- Treating dendrogram cut-points as ground truth — always cross-check with qualitative grouping rationale if available.
- Testing IA in isolation from content — the "About" category is meaningless without the pages it contains.
- One-shot validation — IA should be tested iteratively as content and features evolve.
Handoff
- To Echo: once IA is validated, hand off the navigation structure for cognitive walkthrough on representative flows.
- To Pulse: recommend nav-path events (e.g. breadcrumb clicks, category-page arrivals) to monitor in production.
- To Voice: register IA labels that failed testing so post-launch feedback can be mapped against them.
- To Canvas: hand off validated IA tree for sitemap diagramming.
- Always include in handoff: n, method, per-task success rates, problem cards / labels, confidence intervals, locale scope.