Usability testing means watching real people try to complete real tasks on your site or app while they think out loud, so you can see where they get stuck.
For most teams, five participants per round is enough to surface the biggest problems, as long as you write neutral tasks, stay quiet while they work, and run another round after you fix things.
Treat what you find as hypotheses: the fixes still need an A/B test before you can claim a lift.
This guide covers the questions to ask, a script you can copy, how to run a session, and which method to pick.
Each step is backed by tests from our database, where problems spotted by watching users turned into measured results (and a few that didn't).
User testing questions: what to ask before, during and after
The best user testing questions are open, neutral and about past behavior.
"How do you usually buy running shoes?" tells you more than "Would you use a size finder?", because people are poor at predicting what they would do.
Split your questions into three groups:
- Screener questions (before the session) filter for people who match your customers. Ask about behavior ("When did you last buy X online?") rather than identity, and don't reveal which answer qualifies.
- Warm-up questions (first 3-5 minutes) set context: "Tell me about the last time you shopped for this." "What did you compare before deciding?"
- Follow-up questions (after each task, and at the end) dig into what you saw: "What did you expect to happen when you tapped that?" "What, if anything, nearly stopped you?" "If you could change one thing, what would it be?"
Avoid leading questions ("Wasn't that easy?"), double questions ("Was the price clear and fair?") and questions that ask people to design for you ("What color should this button be?").
For more on wording, see our guide to survey questions, since the same rules apply.
Usability testing questions that produce findings you can test
Tasks are the core of a usability test, and a good task reads like a real goal, not an instruction.
"You need a waterproof jacket for a hiking trip next week, under $200. Find one and get to the point where you'd pay" works.
"Click the Jackets menu and use the filter" gives away the answer.
Here are task and probe questions that tend to produce testable findings:
| What you want to learn | Task or question | What to watch for |
|---|---|---|
| Can people find things? | "Find a gift for a 5-year-old under $30." | Menu vs search, back-tracking, dead ends |
| Is the offer clear? | "In your own words, what does this company sell, and who is it for?" | Wrong guesses, hesitation, scrolling to find out |
| Is the product page enough? | "Decide whether this is right for you. Tell me when you've decided." | Questions they ask out loud that the page doesn't answer |
| Does checkout flow? | "Buy this using the test card on the sheet." | Field errors, confusion about why data is needed |
| Trust | "Would you buy from this site? Why or why not?" | Doubts about legitimacy, shipping, returns |
The "what to watch for" column matters more than the answers.
In one test from our database, Oliver Kenyon's team found through user testing and behavior analysis that mobile shoppers were frustrated by small, hard-to-tap menu categories.
A full-screen mobile menu with larger tap targets then won an A/B test worth +$50,568 per month.
Trust questions can be just as productive. Ruben's team found through user research that the brand was the market leader and inventor of its product, but the site made it look like a dropshipper.
A trust bar under the navigation stating "Patented since 2014" and "34,000+ Happy Customers Worldwide" produced +55% conversions.
Usability test script: a template you can copy
A script keeps every session the same, so differences come from the participant and not from you.
Read it word for word at the start; improvise only in follow-ups.
Intro (2 minutes). "Thanks for joining. We're testing the website, not you, so there are no wrong answers.
If something is confusing, that's exactly what we need to hear.
I didn't design this, so you won't hurt my feelings."
Think-aloud instruction. "As you go, please say what you're looking at, what you're trying to do, and what you're thinking.
I may stay quiet, because I want to see how you'd do this without help."
Consent. "Is it OK if we record your screen and voice? We only use it internally."
Warm-up (3-5 minutes). "Tell me about the last time you bought [category] online."
Tasks (20-30 minutes). Read one task at a time from a card.
After each: "How did that go? Was anything unexpected?"
Wrap-up (5 minutes). "What was the most frustrating part?
What would make you more confident buying here? Anything else you'd like to add?"
Close. "That's everything. Thank you, that was really useful."
When participants ask "What should I do?", answer with a question: "What would you do if I weren't here?"
When they go quiet, a neutral "What are you thinking right now?" brings them back.
The think-aloud method is what makes this work: participants say their thoughts as they use the interface, and Nielsen Norman Group notes that most people need regular prompting to keep talking, and that clumsy facilitator questions can easily bias behavior.
How to conduct usability testing, step by step
- Pick one goal. "Why do mobile visitors drop between product page and cart?" is a goal. "See what people think of the site" is not.
- Write 3-6 tasks tied to that goal, plus a success definition for each (for example, "reached the payment step with the right size in the cart").
- Recruit 5 people per distinct user group who match your customers. Use your own customer list, a recruiting panel, or an unmoderated platform.
- Run a pilot with a colleague to catch broken tasks and timing problems.
- Run the sessions. One facilitator, one note-taker if you can. Record everything.
- Log every problem with how many participants hit it and how badly (blocked, slowed, annoyed).
- Turn the top problems into hypotheses written as test hypotheses, then rank them with your prioritization framework.
- Fix and retest. Run a new round on the changed design, and A/B test changes where you have the traffic.
How many participants do you need?
The "five users" rule comes from Jakob Nielsen's 2000 article.
It models the share of problems found as:
share found = 1 - (1 - L)^n
where n is the number of participants and L is the share of problems a single participant reveals.
Nielsen reported a typical L of 31%, averaged across many projects. Worked through:
| Participants (n) | Share found if L = 31% | Share found if L = 10% |
|---|---|---|
| 1 | 31% | 10% |
| 3 | 67% | 27% |
| 5 | 84% | 41% |
| 10 | 98% | 65% |
Two caveats matter. First, L is not fixed: on a complex product or with very different users it can be much lower, and five people then miss more than half the problems (see the right column, which is our illustration, not Nielsen's data).
Second, any single group of five varies. In Laura Faulkner's 2003 simulation study, five-person samples found anywhere from nearly 100% down to 55% of the known problems, while 10-person samples found 95% on average.
Nielsen's own advice was to spend a 15-person budget on three rounds of five, fixing problems between rounds, and to test 3-4 people from each group when you have two very different audiences.
The five-user rule is for finding problems, not measuring them: for quantitative usability metrics like task success rate, Nielsen Norman Group recommends about 40 participants (assuming a 15% margin of error at 95% confidence).
When usability testing is the wrong tool
Five people can tell you why a step is confusing.
They can't tell you how much revenue a fix will make or whether a change "works".
A usability finding is a lead, not a result.
Our database shows why. Clean Commit tested a more prominent, guided search with suggestions on a Shopify store, the kind of change that looks like an obvious usability win, and recorded -12.0% conversion rate (not significant, though AOV rose 11.5%).
Their modern image carousel lifted add-to-cart by 4.5% on 84,028 visitors, but orders barely moved.
Testing told them what usability work alone couldn't.
Usability testing methods compared
Most methods sit on two dimensions: moderated or unmoderated, and in person or remote.
Pick based on what you need to learn and how fast.
| Method | Best for | Weakness |
|---|---|---|
| Moderated remote (video call) | Understanding why; follow-up questions; complex flows | Slower to schedule; facilitator can bias |
| Moderated in person | Physical products, mobile hardware, body language | Expensive; small local sample |
| Unmoderated remote | Fast feedback on clear tasks from many people | No follow-ups; testers may rush |
| Five-second and preference tests | First impressions, message clarity | Says nothing about task completion |
| Card sorting and tree testing | Navigation labels and structure | Abstract; no real page context |
| Heuristic evaluation | Cheap expert review before testing users | Experts aren't customers |
| Guerrilla testing | Quick checks with whoever is nearby | Participants may not match your buyers |
For a quick measure of perceived usability after tasks, many teams add the System Usability Scale (SUS), a 10-item questionnaire created by John Brooke in 1986. To score it, subtract 1 from each odd-item answer, subtract each even-item answer from 5, add them up and multiply by 2.5 to get 0-100. Across 500 studies, the average SUS score is 68.
With five participants, treat SUS as a rough signal only.
Pair usability tests with behavioral data. Session replay tools show where real visitors struggle at scale, and usability sessions explain why.
If you're choosing between the two, our guide to qualitative vs quantitative research covers when each fits.
For platforms, compare options in user testing tools or read our reviews of UserTesting, Maze and Lyssna.
What fixing usability problems has done in our database
Small, observable friction points are where usability testing earns its keep.
Blend Commerce added short text labels to icon-only mobile header buttons at Goat Milk Stuff and measured +10.37% conversion rate.
Surefoot grouped email and phone into a labeled block with helper text in BioLite's checkout, explaining why each was needed, for +18% transactions.
Bundled fixes show up too, but read them carefully.
Conversionista shipped 18 research-led checkout fixes at Devold and reported +73% checkout conversion, measured before and after against a year-earlier month, not split tested.
Miller and Smith added links where visitors tried to click and captioned gallery images, reporting +8.5% goal conversions year over year. Neither tells you which single fix mattered.
@oliverkenyon
Blend Commerce
Surefoot
@Rubendgrf
Clean CommitClean CommitBrowse more results in the full test database, or see how usability testing fits with interviews and surveys in our customer research guide.