Study types

When you click New study you pick one of four kinds. You can switch kind while the study is a draft; the draft below changes with it.

Kind	What happens	Price
AI interview	The AI interviews people by voice, asking your guide's questions and following up on their answers.	From $10 per interview
Human interview	You or a teammate run the call. We handle screening, booking and the call room.	From $10 per call
Text survey	People answer in writing, like a form, and the AI asks follow-ups.	Free with your people, $3 a response from ours
Website survey	You write a brief. When something happens on your site, a small chat opens and the AI asks visitors about it, following up on each answer.	Free

Prices depend on who takes part. See Launch & cost.

AI interview

The default kind. Set-up steps: Interview → Guide → People.

• Format: Voice only (the default; works on any phone) or Video + screen share. With video, the participant's camera and one window or screen they share are recorded alongside the call. The AI interviewer only hears them: it doesn't see the camera or the screen, and it asks people to say what they see and do.
• Camera and screen required (video only): off by default, so someone who can't share (phones can't share a screen) or says no carries on by voice and you get the audio. Turn it on and only people who share both can take part, which means a laptop or desktop.
• Length: 15, 30 (default) or 45 minutes.
• Moderator: Adaptive follows what people say and changes the order when it helps; Structured asks the same questions in the same order for everyone.
• Probe depth: Light (no follow-ups), Standard (one) or Deep (two).
• Language: the language the interview runs in.
• What they'll see (optional): images, video, a website, a Figma prototype or a YouTube video, shown at a chosen question. You can shuffle the order so the first one doesn't colour the rest. In a voice-only interview a site or prototype opens as a link; to record them using it, pick Video + screen share.
• On-screen questions (optional): a rating from 1 to 5, NPS from 0 to 10, single or multiple choice. The AI pauses, the question appears on screen, and the answer shows in the study report as a chart.

The default target is 12 interviews; you can pick 3 to 40.

Human interview

For calls you want to run yourself. Set-up steps: Calls → Guide → People.

• Who runs the calls: one or more hosts from your team.
• Availability: the days and times people can book, your time zone, slot length (30, 45 or 60 minutes) and a buffer between calls.
• Video call: calls run in our call room, a browser link with nothing to install.
• Booking link: sent to everyone who passes the screener. You can share it yourself too.

The discussion guide sits beside the call so nothing gets skipped. See Live interviews.

Text survey

People answer in writing and the AI asks follow-ups. Set-up steps: Setup → Questions → People. The default target is 120 responses (20 to 500).

Send to 5 people first

To try the survey on a small group before everyone gets it, click Send to 5 people first on the Summary card instead of launching.

1. A pop-up shows what the 5 responses cost: free with your own people, $3 each from our people. Click the button, for example Send to 5 people · $15. It's paid like a launch (see Launch & cost).
2. The survey goes out, but only the first 5 people can start it. With our people, only a few are invited. After 5 have started, anyone else sees that it's full for now.
3. The study's Overview shows a Pilot card with how many have answered. Click Read the answers to see them, and Edit questions to change anything.
4. Click Approve the rest. The pop-up shows the price of the remaining responses (your target minus 5). Once it's paid, the survey opens to everyone it's meant for.

The pilot's 5 answers are real responses: they count toward your target and the study report. If you never approve the rest, you keep what the pilot got; nothing is refunded.

Website survey

A small chat on your website. You write a brief: what you want to learn from visitors. When something happens on the page, the chat opens and the AI talks with the visitor about your brief. There are no questions to write: the AI writes its first message from the brief, follows up on each answer, and thanks the visitor after 3 to 5 messages. Set-up is one page:

• Brief: what you want to learn, in your own words (up to 500 characters). For example: Why do visitors leave the checkout before they pay? What stops them, and what would get them to finish? Change it any time from the study's Widget tab (Edit brief). The AI writes a new first message from it within a few seconds.
• Where and when: your site's URL and what opens the chat: an event, time on the page, exit intent, scroll depth, or your own code.
• More options (closed by default):
  • Pages: the pages it can show on (all pages by default). Use * as a wildcard.
  • Only on these sites: the domains allowed to show it (acmeflow.com covers its subdomains too). Leave it empty and any site with the snippet can show it.
  • Sample: the share of matching visitors who see it (25% by default), and whether that's all, new or returning visitors.
  • At most once every 30 days: how often one visitor can be asked (off: once a week).

You can also ask Eva: "Run a website survey on acmeflow.com about why people leave the pricing page." She creates it from the brief and the site, and you check it before you approve.

Install the widget

Copy the snippet from the Install card (in set-up, or on the study's Widget tab) and paste it before </body> on every page:

<script src="https://app.userevaluation.com/api/embed/w.js" data-key="pk_live_…" async></script>

data-key is the study's public widget key; the snippet in the app has yours filled in. If the trigger is an event or your own code, the snippet adds a second line. Call it when the moment comes:

<script>window.ue=window.ue||function(){(ue.q=ue.q||[]).push(arguments)};ue('trigger', 'checkout_idle')</script>

For an event trigger the name must match the event name you set. For your own code, call ue('trigger') (or ue.trigger() once the script has loaded).

The widget stays asleep until the study is published, and goes back to sleep when it's paused, reaches its target or finishes.

Preview on your site

To try the real chat on your own pages before visitors see it, click Preview on my site (at the end of set-up, or on the Install card of the Widget tab). It gives you two ways in:

• Open preview: opens your page with a private ?ue_preview=… code at the end of the address. Use it once the snippet is on your site.
• Preview survey bookmark: drag the button to your browser's bookmarks bar, open any page of your site, and click the bookmark. It loads the chat on that page, so you can try it before the snippet is installed. A few sites block outside scripts; there, install the snippet and use the link instead.

A preview shows at once and ignores the trigger rules. It works while the study is a draft, live or paused, on the sites in Only on these sites if you set them, for 2 hours and up to 10 chats (one new chat every 15 seconds); after that, make a new one. Anyone you send the link to can try it too. To end every preview link at once (say one was shared further than you meant), open the ⋯ menu on the Install card of the Widget tab and choose Stop preview links: the links stop opening the survey, and preview chats under way end at their next reply. A preview chat is saved as a pilot: you see it in Sessions, but it doesn't count toward the target, the results or the study report, and it isn't counted as shown or opened.

What a visitor sees

1. The AI's first message, with a box to reply and a close button.
2. The visitor replies. The AI answers with one short follow-up about what they said, and stays on your brief.
3. After 3 to 5 messages from the AI (never more than 6), it thanks the visitor and the chat ends.

The AI only talks about your brief. If a visitor asks for something else (support, prices, other topics) or tries to give it instructions, it says it can't help with that here and goes back to the brief. It never asks for names, email addresses or payment details. A reply can be up to 600 characters.

Each chat becomes a session in the study, with the whole conversation as its transcript, plus the page, the trigger and the device. It shows in Sessions, the results group what visitors said, and the study report cites it. If a visitor closes the chat early, or stops replying for 10 minutes, the chat is saved with what they said so far. The Widget tab counts how often it was shown and opened.

Limits

To stop abuse and keep AI cost under control, the chat has limits. If a limit is reached, the visitor's words are still saved.

• New chats and replies are limited per visitor address. A visitor who sends replies too fast is asked to wait a moment.
• Each workspace has a daily limit on AI replies in website chats. Past it, chats end with a fixed thank-you instead of an AI reply until the next day.

The widget sets no cookies. It remembers, in the visitor's browser, when it last showed (for the frequency cap) and whether they're in the sample.

note

Visitors answer in writing. Voice answers aren't available on your site yet.

Study statuses

Status	Meaning
Draft	Being set up. Nothing runs and nothing is charged.
Recruiting	Launched; people are being invited and taking part.
Running	A website survey that's switched on, or a resumed study with sessions under way.
Paused	Nobody new is invited until you resume.
Study report ready	The study report is ready: it was generated automatically when the study filled or closed, or you generated it.

A launched study can go back to draft only while nobody has taken part yet, and only when its launch cost nothing. A paid launch can't go back to draft; pause it instead.