Create a study

This guide walks you through creating a study from start to finish. By the end, you have a live Researcher Agent ready to interview your users.

​

Overview

Creating a study involves these steps:

1. Start a new study and describe your objective
2. Choose a style and refine your interview guide in the builder chat
3. Configure your Agent, Outreach, Persona, Reward, and Interview Depth in the sidebar
4. Test your study with the embedded preview
5. Go live when ready

​

Step 1: Start a new study

From your dashboard, click New in the sidebar. Enter your study objective in plain language:

Our purchase funnel is: browse, add to cart, enter payment, confirm order. We’re seeing a big dropoff at the payment step. I want to understand what’s causing users to leave at that point and what would get them to complete the purchase.

​

Step 2: Choose the interview type and a style, then refine your guide

Answer two questions before Pillow drafts anything. The first sets the interview type:

How should participants take the interview?

Video interview shows each question on screen and records the answer on camera, opening on a welcome page the builder writes for you. Text chat runs in the chat widget. The second question sets the guide style:

How do you want to build your Interview guide?

Questions give you control over the exact wording your agent uses. Topics let it write its own questions around the areas you set. Pillow then drafts the guide, searching the web for context on your product or market and proposing angles you might not have considered. Use the chat to refine it. Example prompts:

• “Add a question about whether users trust us with their payment info”
• “Remove question 3, we already know about cart abandonment”
• “Make the pricing question more specific to the payment step”
• “Merge the last two questions into one”
• “Reorder the guide to prioritize the dropoff reasons first”

Write your interview guide →

​

Undo, redo and version history

Every change to the study setup is saved as a version, whether you made it in the chat or by editing a section directly. Press ⌘Z (or Ctrl+Z) to undo the last change and ⌘⇧Z (or Ctrl+Shift+Z) to redo it, or use the arrows in the top bar. The chat stays as it is, and the agent is told what you restored. Click History in the top bar to browse every version. Selecting one shows it in place of the current setup with a banner: Revert to this version makes it the current setup again, and the later versions stay listed as undone so you can move forward to them. Your next edit replaces them. Cancel returns to where you were. Your guide decides what the agent covers, so more specific items lead to more focused interviews.

​

Adding images

You can attach up to 4 images to a builder message. Images are saved with the relevant guide items and participants can see them during the interview. Use this for UI screens, design mockups, ads, logos, packaging, or product photos you want reactions to. Click the image icon on an item card to upload participant-visible images directly. Supported formats: PNG, JPEG, WebP. In video interviews, guide images appear alongside the question. Participants can open an image full screen to inspect it before recording their answer.

​

Adding notes

Notes are instructions your agent follows as ground truth, the fastest way to give it constraints, terminology, and context that don’t fit into your guide. Add them in the Research notes panel. See Steer your agent with notes.

​

Step 3: Configure in the sidebar

The right sidebar panel provides quick access to configure your study. These settings are accessible at any time during the building process.

​

Agent

Click Agent in the sidebar to customize your Researcher Agent:

• First name: Choose a friendly, human name (e.g., Sarah, Alex, Jordan)
• Role: Customize the text shown under the agent’s name in the chat header. Use {{org_public_name}} to include your public organization name. For custom roles, turn on Auto-translate role to localize it for supported participant languages.
• Gender: Select Female, Male, or Non-binary (used for grammatical agreement in gendered languages)
• Look: Choose Logo (uses your organization logo), Text (auto-generated from the agent’s name), or Custom (upload your own image or generate with AI)
• Tone of voice: Set the interview style
• Reveal policy: Control how the agent responds when asked if it’s AI
• Formal address: Toggle on for languages with formal pronouns (vous, usted, Sie). When enabled, the agent uses the formal form of address.

​

Interview type

Pick how participants take the interview:

• Text chat: the default. Participants type their answers in the chat widget, and can send voice messages.
• Video interview: participants answer on camera. Each question shows on screen one at a time. They press Start recording, answer, then press Submit answer. Their words are transcribed live while they speak, so the agent asks the next question right after they submit, exactly as in the text chat. The camera records only while they answer, and each answer plays as a video clip in the interview view.

Participants see a welcome page, then check their camera and microphone before the first question, and can pause while recording an answer. The interview type locks when the study goes live. If sending an answer fails, participants can select Try again to resend the same recording or Cancel recording to discard it. The recording stays available while the interview page remains open. Each submitted video answer uploads separately. Once submission succeeds, that clip stays with the interview even if the participant leaves before finishing. Participants should wait for submission to finish before closing the page. A video study has no agent name or look, and sends no outreach message: participants meet the study through the welcome page instead, and the interview opens directly on the first question. The Welcome page section replaces the Agent section and holds the title and subtitle of that first screen. Write them yourself or ask the builder, and use {{minutes}} in the subtitle to show the estimated length, which follows the interview depth and the size of your guide:

Thanks for joining!

We’d love to hear about your online video habits. There are no right or wrong answers, we’re just interested in your honest experience. This should take about minutes.

Empty fields fall back to the placeholder text shown in the builder, in the participant’s language. The same section has an optional Recording consent switch. When it is on, the welcome page shows a checkbox (“I agree to the recording and processing of my interview data for research purposes as described in the Privacy Notice”) that participants must tick before they can continue. The link opens Pillow’s Privacy Notice, since the platform processes the recording. The time of consent is kept with the interview. For a video study, the Question skipping switch at the top of the study page, next to the interview type and depth, is off by default, so every question needs an answer. Switch it on to show a skip button next to the record button, so a participant can move past a question without answering. The Read questions aloud switch below it is off by default. When it is on, a voice reads each question as it appears on screen, in the language the question is written in, and participants get a small Read aloud switch in the top right corner to turn the voice off and back on. The voice stops the moment they start recording, so it never ends up in their answer.

​

Interview depth

The Interview Depth slider in the sidebar controls how often your agent follows up on each item in your guide: You also get an estimated interview length below the slider, based on the depth you pick and the number of items in your guide.

​

Outreach

Click Outreach in the sidebar to set your opening message. This is the first message participants see:

Hi, I’m {{agent_name}} from {{org_public_name}}. We’re trying to make checkout smoother, and I’d love to hear about your last purchase. How did it go?

When you open the Outreach panel, Pillow drafts an opening message from your interview guide and agent. Click Regenerate for a new version, or edit it directly. You can also add translations for multi-language studies.

​

Persona Optional

Click Persona in the sidebar to describe your target audience. Attributes are freeform key-value pairs that give the agent context about who it’s interviewing. For a checkout study, you might add: Define your persona →

​

Reward Optional

Offer participants a reward for completing the interview, and collect a way to reach them. Choose whether every participant is rewarded or a few winners are drawn at random, then pick what identifier the agent collects (email, phone, username, or a custom field). Offer rewards →

​

Editing via chat

You can also modify any of these settings by chatting with the Study Builder. Ask it to change the agent’s name, update the outreach message, adjust persona attributes, or switch the interview depth, and the builder updates the configuration inline.

​

Step 4: Test your study

Once your Agent and Outreach are configured, click Preview to try the interview and get a shareable test link. The sidebar shows:

• Test link: Share internally for QA before going live
• Embedded preview: Test the interview directly in the builder

​

What to test

• Do your scripted questions read naturally where they land in the conversation?
• Are follow-ups appropriate based on responses?
• Does the interview feel natural?
• Is the tone appropriate for your audience?

Use the chat in the study builder to refine your interview guide and agent behavior based on what you observe in the preview.

​

Step 5: Go live

When you’re satisfied with the preview:

1. Review your configuration one final time
2. Click Go live in the study builder
3. Your production link is ready to share

​

What’s next