Writing a study
The study editor is one form with a live preview beside it. Everything here maps to a field you can see on that form, in the order you meet them.
Start from an example
A new study offers four worked examples — Pricing clarity, Homepage, Feature reaction and General feedback. Picking one fills every audience-facing field with a setup that already works, which you then bend towards your own product.
If you have already edited fields by hand, picking an example asks first: replace everything with the example, or keep your edits and let the example fill in the rest. Nothing is overwritten silently.
Once an example is in the form, Clear appears beside the examples — it empties every field, after asking, so you can start from scratch. If you would rather write the study yourself, close the examples with the × in the corner.
The two ways a study opens
The first decision in the form is whether to lead with a rating or go straight to a question. These are the live launchers, in their default styling — the same code your visitors run.
The settings
- Internal only — visitors never see it. Name it after the page or the question, so a dashboard of a dozen studies still makes sense in three months.
- rating, or straight to chat
Yes, ask for a ratingshows a thumbs or 1–4 control first and opens the chat once someone taps. The tap itself is recorded, so even people who write nothing leave a signal — which is why rating-first studies collect far more total feedback.
No, start with a question opens straight into the conversation. Fewer people engage, but everyone who does gives you words.
- rating-first only
- Thumbs up / down for a two-way reaction, or a 1–4 scale when you want degrees. Four points, with no middle, means nobody can sit on the fence.
- rating-first only
- The line above the rating buttons — “How was your experience?” if you leave it blank. Keep it to one short sentence: it has to fit in a compact launcher.
- chat-first only
- The first thing the interviewer says. Ask about something the person just did rather than about your product in the abstract: “How did you find this page?” gets an answer, “Do you have any feedback?” gets silence.
- The goal every follow-up steers towards. Write it as the question you would ask a colleague — “Why do visitors hesitate to upgrade to a paid plan?” — not as a topic. The more specific it is, the sharper the follow-ups get.
- Where and when you are asking: which page, which feature, what the person has just done. This is how the interviewer tells whether a complaint is about your product, this page, or the task they just finished.
- inline or floating
- Inline component is a compact block that sits in the page where you put it. Floating button is a tab pinned to the edge of the screen. Both open the same chat panel — see embedding the widget.
- floating only
- Which edge the tab sits on: Right or Left as a vertical tab at mid-height, or Bottom right / Bottom left as a pill in the corner.
Border (with radius, width and colour), background colour and text colour for the inline component; background, text colour and a shadow (size, opacity, colour) for the floating tab. Turning background off leaves the launcher transparent, so it picks up whatever is behind it.
Reset styles at the bottom of the group puts just that group back to the defaults — your placement and content settings are untouched, and it asks before it does anything.
- 1–8, default 4
- The most follow-up questions the AI may ask after the opening. It often stops early when someone clearly has nothing more to add. Three or four is a good default; past five, drop-off usually costs you more than the extra answers are worth.
The AI cannot see your site
It never loads your page, and it knows nothing about your product beyond the goal and context you type. If interviews keep going somewhere irrelevant, the fix is almost always more context — say what the page is, who is on it, and what they just did.
Let the setup coach check it
Review my setup, in the bar at the bottom of the editor, runs an AI review of the configuration before anyone sees it. You get a verdict — Strong, Could be sharper, or Needs work — a score out of 100, and a one-paragraph summary you can expand.
Where a field is holding you back, the suggestion appears beside that field with a rewrite you can apply in one click. Review as often as you like: it judges what is on screen, saves nothing, and changes nothing until you accept a suggestion.
A field the coach is happy with says so too, so a review tells you what is already working — not just what to fix.
The live preview
The preview mirrors the settings as you type, in a phone or desktop frame, using the same code your visitors will run — what you see is what gets embedded.
Preview conversations are real responses
On a saved study, running the interview in the preview records a response like any other. Handy for a first end-to-end test; worth deleting the study and starting clean if you have been experimenting heavily before launch.
Ending, reopening and deleting
- End study stops it collecting. The widget quietly removes itself from your site — visitors see nothing broken — and the embed snippet is hidden in the app until you reopen. Every response you already have stays.
- Reopen study puts it straight back to collecting on the same key. Snippets already pasted on your site start working again; no re-copying.
- Delete study is permanent, and takes the responses and analyses with it. Download what you need first — see reading responses.