Watch people try real tasks on your website, web app or prototype while they think out loud, and see where they get stuck, what confuses them and what works.
At a glance
- 3 prompts to the Project Agent
- ~25 minutes per interview
- 10-20 users
- 10 minutes to set up
Start here
You’ll give the Discuss Project Agent three short prompts. It designs the interview around your product and tasks, and you can change anything before it’s built.
Make your own copy first: File > Make a copy. Then use Edit > Find and replace to fill each blank everywhere at once, and every prompt is ready to paste.
You’ll also need: a link people can open without logging in (for a prototype, a public share link) and two or three tasks with a clear end point. Screen recording only works on a desktop or laptop, so to test a mobile app, use a desktop prototype or emulator link.
How to set up the interview
Step 1: Fill in the blanks
| Replace | With, for example |
|---|---|
| [PRODUCT TYPE] | website, web app or prototype |
| [PRODUCT DESCRIPTION] | a clickable prototype of our new online store |
| [TASK AREA] | shop for running gear online |
| [TASK 1] | find running shoes in your size and add them to your cart |
| [TASK 2] | find out when an order would arrive at your address |
| [TASK 3] | check out as a guest, stopping before you pay |
| [ACTION] | use, sign up for or buy from |
| [TARGET AUDIENCE] | people who run at least once a week |
| [AGE RANGE] | 25 to 55 |
| [COUNTRY] | the US |
| [TIME PERIOD] | 3 months |
Step 2: Tell the agent what you’re testing
In your project: Activities > AI Interview > Chat with the Project Agent. Paste Prompt 1. The agent sums it up and may ask a question or two. Answer them, then click Add to Canvas.
Prompt 1 · Research goal
I want to test how easy our [PRODUCT TYPE] is to use for people who want to [TASK AREA]. In short, it is: [PRODUCT DESCRIPTION]. I want to see where people get stuck, what confuses them, what works well, and whether they can finish the main tasks on their own. The results will help us decide which usability problems to fix before launch.
Step 3: Tell it who to talk to
Paste Prompt 2, answer any questions, then click Add to Canvas.
Prompt 2 · Target audience
I want to hear from [TARGET AUDIENCE], aged [AGE RANGE], living in [COUNTRY]. They should have tried to [TASK AREA] in the past [TIME PERIOD], and be comfortable sharing their screen and thinking out loud on a desktop or laptop. Please exclude people who work in UX, design, marketing or market research, or in our industry.
Step 4: Ask it to design the interview
Paste Prompt 3. The agent designs the questions and explains why it picked each one. Check it against the must-haves and ask for any changes, then click Add to Canvas.
Prompt 3 · Study structure
This is an AI-led interview of about 25 minutes, with 10 to 12 questions. Everyone takes part on a desktop or laptop, and spoken answers should be audio only. Participants open the [PRODUCT TYPE] through a link, look around briefly, then try these tasks in order: [TASK 1]; [TASK 2]; [TASK 3]. I want to watch each attempt while they think out loud and see where they struggle.
Please design the interview using the methods you think fit best. These are the must-haves:
A short warm-up about how they [TASK AREA] today, before they open the link.
A Screen Recording Question for the first look around, and one for each task, asking them to think out loud.
Tasks worded as goals. Don’t name the buttons or menus they should use.
A Single Select Question after each of the first two tasks on how it went: completed easily, completed with some effort, not sure, or couldn’t complete.
Open-ended questions after the tasks on the most frustrating moment and the most useful part, then a 5-point rating on how likely they’d be to [ACTION] it.
To change something, tell the agent. For example:
- “Task 2 mentions the Filters menu. Reword it as a goal.”
- “Add a check-in question after Task 3 as well.”
- “Drop Task 3 to keep it to about 20 minutes.”
- “Make the warm-up one question instead of two.”
Step 5: Check it and launch
In the Activity Builder:
- Must-haves: the warm-up comes first, each task has its own Screen Recording Question, the check-ins follow the tasks, and the rating comes near the end.
- Your link: paste it into the URL field of every Screen Recording Question so respondents get a button to copy it. If the site needs a login, add test account details to the first task.
- Follow-ups: 3 on each task. 2 on the other open-ended questions. 1 on the warm-up.
- Last check: search the activity for “[” to catch a missed blank, and read each task aloud, since the AI reads it word for word. Then preview it on a desktop and do every task yourself.
Then launch the activity.
What a good guide looks like
Compare what the agent builds against this. The order matters more than the exact wording.
| Part | What you learn | Sample question |
|---|---|---|
| Warm-up | How they handle [TASK AREA] today, before they see anything | “Think about the last time you tried to [TASK AREA]. Walk me through what you did.” |
| First look | What they think it’s for, before any tasks | “Look around for about a minute. What do you think this is for?” |
| Tasks | Where they get stuck, on screen and in their own words | “Using the link, try to [TASK 1]. Think out loud as you go.” |
| Check-ins | Whether they think they finished each task | “How did that task go for you?” |
| Wrap-up | The biggest frustration, the most useful part, and likelihood to use | “If you could change one thing to make this easier to use, what would it be?” |
Read the results
Start with the task check-in charts to see who finished each task easily and who couldn’t or wasn’t sure (“not sure” usually means the product didn’t confirm success). With 5 to 8 people, use them to pick which recordings to watch first, and look for long pauses, back-and-forth clicking, and moments where people expected something else.
Then ask the Insights Agent:
- “Where did people get stuck in [TASK 1], and what did they expect to happen instead?”
- “What did people think this [PRODUCT TYPE] was for at first glance? Did that match what we intended?”
- “What were the most common frustrations across all tasks, ranked by how often they came up?”
- “What separates people who finished the tasks easily from those who struggled?”
- “What changes did people suggest most often?”
Go further
Comparing two designs
Give each person only one design. Duplicate the finished activity, swap in the second link, and send each copy to its own group, keeping the tasks the same so the check-ins compare. If task order matters, make another copy with the tasks swapped, since later tasks benefit from what people learned in earlier ones.
Extra questions to ask the agent for
- Ease rating: “Add a 5-point rating after each task on how easy or difficult it was.”
- Comparison: “Add a question on how this compares to other websites they use to [TASK AREA].”
- Checkout or sign-up flows: “Add a question on what, if anything, would make them hesitate to enter their details.”
Using Quick Composer instead
Quick Composer runs the same fixed questions every time. To use it, go to AI Interview > Quick Composer, paste the two blocks below, turn Generate Follow-up Instructions on, and click Import into Activity Builder. Then finish with step 5.
Copy into Research Goal
We want to understand how easily [TARGET AUDIENCE] can use our [PRODUCT TYPE] to [TASK AREA]: where people get stuck, what confuses them, what works well, and whether they can finish the main tasks on their own. The findings will help us decide which usability problems to fix before launch.
Copy into Questions for AI-Led Conversations
- Warm-up
Type: Open Ended Question: Audio
Question: To start, tell me a little about yourself and how you usually [TASK AREA]. - Recent experience
Type: Open Ended Question: Audio
Question: Think about the last time you tried to [TASK AREA]. Walk me through what you did. - First impression
Type: Screen Recording Question
Question: Open the link and spend about a minute looking around, without clicking too far in. As you look, say out loud what you think this [PRODUCT TYPE] is for and what you would do first. - Task 1
Type: Screen Recording Question
Question: Using the link, try to [TASK 1]. Please think out loud as you go, saying what you’re looking for and what you expect to happen. Stop when you think you’re done, or if you would give up. - Task 1 check-in
Type: Single Select Question
Question: How did that task go for you?
Options: I completed it easily / I completed it, but it took some effort / I’m not sure if I completed it / I couldn’t complete it - Task 2
Type: Screen Recording Question
Question: Now, using the same link, try to [TASK 2]. Keep thinking out loud as you go. Stop when you think you’re done, or if you would give up. - Task 2 check-in
Type: Single Select Question
Question: How did that task go for you?
Options: I completed it easily / I completed it, but it took some effort / I’m not sure if I completed it / I couldn’t complete it - Task 3
Type: Screen Recording Question
Question: Finally, try to [TASK 3]. Keep thinking out loud as you go. Stop when you think you’re done, or if you would give up. - Most frustrating
Type: Open Ended Question: Audio
Question: Thinking about everything you just did, what was the most confusing or frustrating moment? - Most useful
Type: Open Ended Question: Audio
Question: What, if anything, did you find most useful or easy to use? - Likelihood to use
Type: Rating Question
Question: Based on what you’ve seen, how likely would you be to [ACTION] this [PRODUCT TYPE]? - One change
Type: Open Ended Question: Audio
Question: If you could change one thing to make this easier to use, what would it be?


