User testing questions for a prototype, and where to ask them

User testing questions for a prototype, grouped by before, during and after the session, plus the ones to avoid and how to ask when the tester is alone.

The moment a tester goes quiet over a prototype is the most useful moment in the session, and the easiest one to waste. Ask “what do you think?” and you get an opinion. Ask “what did you expect that to do?” and you find out where the design broke.

Lists of user testing questions are easy to find, and most are fine. This page has one too: 30 questions, grouped by when you ask them, plus the usability testing questions to avoid. But the wording decides less than you'd think. What decides whether a session is useful is where the question gets asked: on the thing, while the person is using it, about one element at a time.

Why does where you ask matter more than the wording?

Because people are poor reporters of their own behavior. Jakob Nielsen of Nielsen Norman Group put the rule in four words in First Rule of Usability? Don't Listen to Users: “Watch what people actually do.” The same article says “Self-reported claims are unreliable, as are user speculations about future behavior.”

A question asked in a debrief, ten minutes after the click, gets a memory of the screen. A question asked while the cursor is over the button gets the screen. So the best questions in a session are short, point at something visible, and ask what the person is doing or expecting. Almost none of them ask what the person thinks of it.

User testing questions to ask before, during and after a session

Swap the brackets for your own product. Thirty is more than one session holds, so pick what fits the hour.

Before the session: screening and context

These find out whether this is the right person, and how they do the job today. Ask about the last real time, never about “usually.”

  1. What is your role, and how long have you been doing it?
  2. When did you last [do the job this prototype is for]?
  3. Walk me through how you did it that time.
  4. What did you use to do it?
  5. Which part took the longest?
  6. How many times did this come up in the last month?
  7. Who else was involved?
  8. What have you already tried to make it easier?

During the session: task prompts

A task gives the person a goal and then gets out of the way. It names what they want, never the button that does it.

  1. Before you click anything: what is this page for?
  2. You want to [goal]. Show me how you would start.
  3. What do you expect to happen when you click that?
  4. You have finished [step]. What would you do next?
  5. Find [a specific thing], and tell me when you think you are done.
  6. Something on the screen just changed. What do you think happened?

During the session: neutral follow-ups

Use these when someone stops, backtracks or says something you didn't expect. Each one is about what is on the screen right now. Two come from Amy Schade's Nielsen Norman Group article on leading questions: “What was easy or difficult about completing that task?” and the habit of pointing at an area and asking what it is, instead of naming it for the tester.

  1. What are you looking for right now?
  2. What did you expect to see there?
  3. What is this area for? (Point at it. Don't name it.)
  4. What is going through your mind?
  5. What was easy or difficult about that?
  6. What would you call this?
  7. How do you know that worked?
  8. What made you choose that one?
  9. Tell me more about that.

After the session: wrap-up

Opinions go last, when they can no longer color the tasks. The final question is the one that most often turns up something you had no task for.

  1. How would you describe what you just used to a coworker?
  2. What stood out, good or bad?
  3. What did you expect to find that wasn't there?
  4. What did you want to do that you couldn't?
  5. How is this different from the way you do it today?
  6. What questions do you still have about it?
  7. What should I have asked that I didn't?

Which questions should you avoid?

Three kinds, and each has a fix. A leading question carries its own answer. A hypothetical asks for a prediction, which is the kind of self-report Nielsen warns about. A yes-or-no question gets a polite yes.

KindInstead ofAsk
Leading“How easy was it to find the export button?”“Show me how you would get this report out of the app.”
Hypothetical“Would you use this every week?”“When did you last need to do this? What did you use?”
Yes or no“Did that make sense?”“What do you think just happened?”

When you can't avoid suggesting an answer, leave the door open. Schade's rewrite of a question about saving time is “How might this affect your efficiency, if at all?” The last three words do the work.

How many people do you need to ask?

Fewer than most teams plan for. In Why You Only Need to Test with 5 Users, Nielsen writes: “The best results come from testing no more than 5 users and running as many small tests as you can afford.” Five people, fix what they found, then five more on the new version.

How do you ask when the tester is alone with the prototype?

Unmoderated testing and async review take the moderator out of the room. Kathryn Whitenton's Nielsen Norman Group guide to unmoderated tests names the cost: “Participants cannot ask for clarification if they don't understand the instructions and you can't ask them to try again if they do the wrong thing.” She adds that task instructions should tell people when to stop, because nobody will be there to move them along.

So the follow-ups above have nobody to say them. “What did you expect to see there?” only works if “there” is on the screen when the question arrives. A form at the end of the session asks it too late, about a screen the tester has already closed.

The fix is to attach the question to the thing. Write each task so it stands alone and names its stopping point. Then give testers a way to answer on the element: where they got stuck, what they expected, in their own words, at the moment it happened.

Let testers answer on the element with Skyelight

Ask your coding agent to publish a single page or component to Skyelight, and it gets its own link with a review badge on it. Testers press C, click the element they mean and write what happened, so every answer arrives attached to the thing it is about.

What does Skyelight cover, and what doesn't it?

Skyelight is a website feedback tool, and it is not a user research platform. It doesn't recruit participants, record sessions or run moderated calls. If you need to watch a stranger's screen while they think aloud, use a tool built for that.

What it does cover is the async part, and the details matter for a test.

  • A hosted prototype. One HTML page or one React component, published by an agent and served at its own link. Free includes one, with any number of versions. Testers sign in with a Skyelight account to comment there, and reviewer accounts are free on every plan. See hosted prototypes.
  • A build with the review badge. Here testers can comment without an account: they type a name and start. That is guest feedback, which is on Pro and above and off until you turn it on. It doesn't work on a hosted prototype.
  • Who sees what. A guest sees only their own threads, so one tester's comments can't steer the next. A reviewer with an account can read the workspace's threads, which suits a team review better than a test. It also means you can pin a question on an element and have reviewers reply to it there. Guests won't see that pin, so put their tasks in the message that carries the link.

Moderated sessions have a place here too. Pin what you saw on the screen where it happened, as the UX research guide describes, and the finding stays next to the element instead of in a deck. For review rounds with a team or a client, the method is in how to annotate a website, and for prototypes built with AI tools, in getting feedback on a vibe-coded app.

Frequently asked questions

What questions should you ask in user testing?

Ask for actions, then for what the person expected. Give a task such as “You want to send this invoice. Show me how you would start,” then follow with neutral prompts like “What did you expect to see there?” Save opinion questions for the last five minutes.

How many users do you need for a usability test?

Five per round is the usual answer. Jakob Nielsen of Nielsen Norman Group wrote that the best results come from testing no more than 5 users and running as many small tests as you can afford. Fix what the first five find, then test the new version with five more.

What is a leading question in user testing?

A leading question carries its own answer, such as “How easy was it to find the export button?” It tells the tester the button exists and that finding it should be easy. Ask for the action instead: “Show me how you would get this report out of the app.”

Can you run user testing without a moderator?

Yes. It's called unmoderated testing: the tester works through written tasks alone. Nobody is there to clarify, so each task has to say what to do and when to stop, and each question has to sit next to the thing it is about.

Is Skyelight a user testing tool?

No. Skyelight is a website feedback tool: testers comment on the element itself, on a hosted prototype or a build. It doesn't recruit participants, record sessions or run moderated calls, so pair it with a research platform when you need those.

Thirty good questions help. One asked on the element, while the person is still using it, tells you more than any of them asked afterward.

Put the question on the element

No card required.