---
title: "UX research methods: card sorts, tree tests and prototype…"
description: "Choosing the right research method for the question Heatmaps, recordings and surveys (previous lessons) tell you how people behave on your live site and…"
url: https://optimizeall.com/learn/conversion-rate-optimization/ux-research-methods
updated: 2026-10-05
---

Conversion Rate Optimization (CRO) · Conversion research · lesson 7 of 20 · 8 min

# UX research methods: card sorts, tree tests and prototype testing

## Choosing the right research method for the question

Heatmaps, recordings and surveys (previous lessons) tell you how people behave on your live site and why they hesitate. But many conversion problems start earlier — in **navigation, labelling, information architecture and flows** — and some ideas are too expensive to build just to test. UX research methods let you test structure, prototypes and comprehension *before* development, often with small samples and low cost.

| Question you have | Method | Typical sample (common guidance) | Tools (examples) |
|---|---|---|---|
| "How do customers group our products/services?" | **Open card sort** | Around 15+ participants per segment | Optimal Workshop (OptimalSort), Maze, Lyssna, UXtweak |
| "Can people find X in our proposed menu?" | **Tree test** (text-only navigation) | Often 50+ for stable success rates | Optimal Workshop (Treejack), Maze, UXtweak |
| "Where would people click first to do X?" | **First-click test** | 30–50+ | Optimal Workshop (Chalkmark), Lyssna, Maze |
| "Is the value proposition clear?" | **Five-second / first-impression test** | 20–50 | Lyssna, Maze |
| "Can people complete this flow in the new design?" | **Prototype usability test** (moderated or unmoderated) | ~5 per round, several rounds | Figma prototypes + Maze, UserTesting, Lookback, Lyssna |
| "How does behaviour unfold over days?" | **Diary study** | 10–20 | dscout, simple forms + WhatsApp/email prompts |
| "How usable is it overall, over time?" | **Benchmark** (task success, time, SUS questionnaire) | 20+ per round | Any testing platform + spreadsheet |

The "five users" guidance (popularised by Nielsen Norman Group) applies to **qualitative** usability rounds that find problems, not to measuring rates. When you need numbers — success rates, click accuracy — use larger samples.

## Card sorting and tree testing: fixing navigation with evidence

Navigation labels written by internal teams often use internal language ("Solutions", "Resources", "Lifestyle range"). Customers look for "Office chairs" or "Prices".

1. **Open card sort**: participants group 30–60 product or content cards and name the groups. Look for common groupings and labels.
2. **Draft a new tree** from the results.
3. **Tree test** the new tree (and the current one as a baseline) with realistic tasks: "You need a desk that fits a small bedroom. Where would you look?" Measure success rate, directness (no backtracking) and time.
4. Only then design and build the navigation — and, if traffic allows, A/B test it on the live site.

## Prototype testing before you build

For big changes — a new checkout, a redesigned configurator, a booking flow — test clickable prototypes (for example in Figma) with 5 participants, fix the issues, and test again. Two or three rounds usually remove the serious problems before a developer writes a line of code. Then A/B test the built version to measure its impact.

## Hands-on: an unmoderated test plan

```markdown
Goal: Can first-time mobile visitors book a home-cleaning slot in the new flow?
Prototype: Figma link (mobile frame), starting at the service page
Participants: 5 per round, UAE residents who booked a home service in the last 6 months;
              screen out UX/marketing professionals
Tasks (scenarios, not instructions):
 1. "You want a 3-hour deep clean for a 2-bedroom apartment next Saturday morning. Book it."
 2. "Before paying, check whether cleaning supplies are included."
 3. "Change the booking to Sunday."
Success criteria: task completed without help; note time and wrong turns
Follow-up questions: "What, if anything, was confusing?" "How confident are you the booking is confirmed? (1-5)"
Analysis: issue list with severity (blocker/major/minor) x frequency (n/5); fix; retest
```

## AI in UX research (useful, with limits)

AI now appears throughout research tools: drafting discussion guides and tasks, transcribing and summarising sessions, clustering notes, and even running AI-moderated interviews or tests. Treat these as accelerators:

- Use AI to **prepare** (guides, screener questions) and **process** (transcripts, first-pass clustering).
- Do **not** use "synthetic users" (AI personas answering as customers) as a substitute for real participants; they reflect training data, not your customers. At most, use them to pressure-test a discussion guide.
- Keep a human reading raw quotes and watching key moments before findings are written.
- Obtain consent for recording and AI processing; check where your research platform stores data.

## Worked example: a Karachi electronics retailer's navigation

The retailer's menu grouped products by brand and internal categories ("Digital lifestyle"). Site search data showed people searching "earbuds under 5000" and "laptop for students". An open card sort with 20 customers produced groups by **use and budget**. A tree test of the new structure versus the old showed higher task success for finding "a laptop for university under a budget" (illustrative). The team rebuilt navigation around use cases with brand as a filter, then monitored search usage and product-listing conversion after launch.

## Common mistakes

- Testing with colleagues or friends instead of target customers.
- Giving instructions ("click Products") instead of scenarios.
- Using five participants to claim a success *rate*.
- Skipping the baseline tree test, so you cannot tell if the new structure is better.
- Treating AI summaries or synthetic personas as research findings.

## Video lecture: UX research methods: card sorts, tree tests and prototype testing

Lecture coming soon · 13 chapters · about 8 minutes. Read the full transcript below.

1. UX research methods
2. Why UX research in CRO
3. Question → method
4. Sample sizes
5. Card sort + tree test
6. Prototype testing
7. Simple example: Dubai booking prototype
8. Realistic example: Karachi electronics navigation (illustrative)
9. Watch me do it: unmoderated test plan
10. AI in UX research
11. Research + experiments
12. Common mistakes
13. Recap

## Lecture transcript

### UX research methods

What if you could find out whether a new navigation, a new checkout or a new booking flow works, before a developer writes a single line of code? That's what UX research methods are for. In this lecture you'll learn how to choose the right method for your question, how card sorting and tree testing fix navigation with evidence, how to test clickable prototypes in rounds, and how AI is changing research, including where it helps and where it's a trap. Then you'll watch me write an unmoderated prototype test plan.

### Why UX research in CRO

Why add these methods to CRO? Because many conversion problems start before the page. They start in the structure: what the menu is called, how products are grouped, how many steps a flow has. And some ideas are too expensive to build just to A/B test. A new checkout might take weeks of development. If a prototype test with five people shows that three can't find the delivery options, you've saved those weeks. UX research is cheap insurance for expensive changes.

### Question → method

Start with the question, then pick the method. How do customers group our products? That's an open card sort. Can people find something in our proposed menu? A tree test. Where would people click first to do a task? A first-click test. Is the value proposition clear? A five-second test. Can people complete this flow in the new design? A prototype usability test. How does behaviour unfold over days? A diary study. Think of it like a toolbox. A hammer is great, but not for every job.

### Sample sizes

A word on sample sizes, because this confuses people. The famous five users guidance is for qualitative usability rounds, where the goal is finding problems. Five people typically reveal the most serious issues, so you fix them and test again. But when you need numbers, like a success rate in a tree test or click accuracy in a first-click test, you need larger samples. Common guidance is around fifteen or more for card sorts, and fifty or more for tree tests. Match the sample to whether you're finding problems or measuring rates.

### Card sort + tree test

Card sorting and tree testing work as a pair. In an open card sort, participants group thirty to sixty cards, each a product or piece of content, and name the groups. You learn how customers think, and what they call things. Then you draft a new menu, and run a tree test: a text-only version of the navigation, with realistic tasks like you need a desk that fits a small bedroom, where would you look? You measure success rate, directness and time. And always tree test the current menu too, as a baseline. Otherwise you can't prove the new one is better.

### Prototype testing

Prototype testing is for flows. Build a clickable prototype, for example in Figma, and test it with about five target users per round, giving scenarios, not instructions. Fix what you find, then test again. Two or three rounds usually remove the serious problems before development starts. You can run tests moderated, where you watch live and ask questions, or unmoderated, where participants record themselves using a testing platform. Moderated gives depth. Unmoderated gives speed and scale. Then, once it's built, A/B test to measure real impact.

### Simple example: Dubai booking prototype

A simple example. A home-services company in Dubai redesigns its booking flow. Before building, they test a Figma prototype with five recent customers. Task one: book a three-hour deep clean for a two-bedroom apartment next Saturday morning. Three of five can't tell whether cleaning supplies are included, and two think they've booked when they've only added to basket. The team adds a supplies line and a clear confirmation screen. In round two, all five complete the task, and all five are confident the booking is confirmed.

### Realistic example: Karachi electronics navigation (illustrative)

Now a realistic scenario with illustrative details. An electronics retailer in Karachi has a menu grouped by brand and internal categories like digital lifestyle. Site search shows people typing earbuds under five thousand and laptop for students. The team runs an open card sort with twenty customers. Groups form around use and budget, not brand. They draft a new tree and run tree tests on both the old and new structure. The new structure helps more participants find a laptop for university within a budget, and with fewer wrong turns. They rebuild navigation around use cases, keep brand as a filter, and watch search usage and listing conversion after launch.

### Watch me do it: unmoderated test plan

Watch me write an unmoderated test plan. Goal: can first-time mobile visitors book a home-cleaning slot in the new flow? Prototype: the Figma mobile frame, starting at the service page. Participants: five per round, residents who booked a home service in the last six months, and I screen out UX and marketing professionals. Tasks, written as scenarios: book a three-hour deep clean next Saturday morning; check whether supplies are included before paying; change the booking to Sunday. Success criteria: completed without help, with time and wrong turns noted. Follow-ups: what was confusing, and how confident are you the booking is confirmed, from one to five? Analysis: issues by severity and frequency, fix, retest.

### AI in UX research

AI is now everywhere in research tools. It can draft discussion guides and screener questions, transcribe and summarise sessions, cluster notes, and even moderate interviews or tests. Use it to prepare and to process. But be careful with one trend: synthetic users, AI personas that answer as if they were customers. They reflect training data, not your customers, and they can't tell you that your date picker is confusing. Don't use them as a substitute for real participants. Keep a human reading raw quotes and watching key moments. And get consent for recording and AI processing.

### Research + experiments

How does this connect to A/B testing? UX research reduces the risk of what you test. If a prototype fails with real people, you don't build it. If it passes, you build it with more confidence, then A/B test to measure the size of the impact on real behaviour. Research tells you whether something is usable and understood. Experiments tell you whether it changes outcomes. Mature programmes use both, in that order.

### Common mistakes

Common mistakes. Testing with colleagues or friends. Giving instructions like click Products instead of scenarios. Using five participants to claim a success rate. Skipping the baseline tree test. Testing prototypes that are too rough to be realistic, or so polished that nobody wants to criticise them. And treating AI summaries or synthetic personas as findings.

### Recap

Recap. Choose the method from the question: card sorts for grouping, tree tests for findability, first-click tests for the first move, five-second tests for clarity, prototype tests for flows, diary studies for behaviour over time. Match sample size to whether you're finding problems or measuring rates. Use AI to prepare and process, never as a stand-in for real customers. Try this now: pick one flow you plan to change, write an unmoderated test plan using the template in the lesson text, and run one round with five target users before any development starts.

## Key takeaways

- Pick the research method from the question: card sort, tree test, first-click, five-second, prototype test, diary study or benchmark.
- About five participants per round finds serious usability problems; measuring rates needs larger samples.
- Card sorting reveals how customers group and name things; tree tests prove whether a new structure beats the baseline.
- Test prototypes in rounds before building, then A/B test the built version to measure impact.
- Use AI to prepare and process research, but never substitute synthetic users for real customers.

## Try it

Choose one flow you plan to change, build a clickable prototype, write an unmoderated test plan with three scenario tasks, and run one round with five target users. Log issues by severity and frequency.

- [Previous: Voice of customer: surveys, interviews and reviews](https://optimizeall.com/learn/conversion-rate-optimization/voice-of-customer)
- [Next: Writing strong hypotheses](https://optimizeall.com/learn/conversion-rate-optimization/writing-hypotheses)
- [All lessons of Conversion Rate Optimization (CRO)](https://optimizeall.com/learn/conversion-rate-optimization)