Frequently asked questions

It runs a short survey wherever you embed it — a thank-you page, a product screen, a documentation site, anywhere you can paste a script tag. One tap answers a closed question, then one short written follow-up. Written answers are grouped into repeated insights, each named with a sentence describing the problem, with the exact answers it was counted from underneath. It works on any page where you can add a script tag, and where you cannot, a link to the survey works in an e-mail or a message.

Grouping answers into insights is table stakes now; several survey and feedback tools do it, some at no cost. The difference is what happens around the insight: you can open the exact answers behind it, the app refuses to name any insight under ten written answers, and it tells you when a recalculation failed instead of showing you a stale result.

No. The closed question can be a rating, but there is no NPS dashboard and no score to watch. The product is built around what people write and what repeats in it.

No. Nothing written here is published, shown on your site, or shared with anyone. The answers go to you.

No. It does not try to answer "where did you hear about us"; other tools do that well and this one stays out of it.

No. It never sees your sales or traffic, so it will not claim to have moved a number it cannot see.

No. The default survey is one tap and one short follow-up, not a panel study with screening, quotas and multiple waves.

Open it. Every insight lists the answers it was counted from, word for word, and the list is exactly as long as the number next to the insight. The only text in that view no respondent wrote is the insight's name and description, and both are labelled as generated.

Because there are fewer than ten written answers in the range you are looking at. In that state the app shows three numbers — how many you have, how many are needed, how many are missing — and a link to the raw answers, instead of grouping a handful of replies into something that looks convincing and means nothing. Narrowing a filter can drop you back below ten, and the same thing happens.

The panel says the run failed, shows the last successful result together with the date it was calculated, and offers to try again. It never shows insights without a calculation date and never presents an old result as a new one. Your answer list keeps working.

A language model, and they are labelled as generated. Every quoted answer under an insight was written by one of your respondents. Grouping is the only step a model is involved in — counting, matching answers, and the arithmetic behind every number are ordinary code. Nothing is filtered out along the way: every answer you collected stays in the list and in the counters.

It depends on how many people see the survey and how many of them write something. Ten written answers are needed before anything is counted. A store running the survey on its thank-you page is one example: published response-rate benchmarks suggest a store doing a couple of hundred orders a month reaches that within the first month — but that is someone else's benchmark, not our measurement, and your panel shows you the real counters from day one.

No. Answers are sent to our AI provider, Anthropic, for grouping only. Anthropic does not train on this data and deletes prompts and outputs within 30 days, unless the law requires otherwise or its safety systems flag the content. The provider is named in the privacy policy, together with everything else that leaves your account.

The text of the written answers, and nothing else. Order numbers, respondent identifiers, response identifiers, your account name and email address are removed before anything is sent; the model receives a list of sentences and returns groups, mapped back by position, not by identifier. On our own server, before an answer reaches our AI provider, we also mask categories such as email addresses, phone numbers, IBANs, card numbers, national IDs and tax/VAT numbers, and postal addresses inside the sentence itself — masking is best-effort and doesn't catch everything. Full detail and examples: see the security page.

No, and we will not call it that. The widget never asks a respondent for personal data, but each answer is tied to the visit it came from, which is what lets you open an answer in context. Calling it anonymous would be easier and untrue.

For the data-retention period of your plan, listed on the pricing page. Deleting your account removes it, and everything under it, immediately.

No. There is no industry benchmark and no cross-account comparison anywhere in the product, and there will not be one. We count how many accounts use a feature; your answers are never put next to somebody else's.

No. The script loads asynchronously and never blocks your page from rendering. If it fails to load, or our servers are unreachable, it simply does not appear — no broken layout, no error message, nothing for a visitor to notice.

Minutes. You pick a template, the wizard has at most three steps, and you can finish it without typing a single character, because the default questions are ready to publish. You can rewrite them afterwards and change them later without starting over.

By default: one closed question answered with a single tap, then one short written follow-up chosen by what they tapped. Nothing else in that default template: no email field, no discount, no newsletter sign-up. Someone who taps and closes the tab has still answered — the tap is saved the moment it happens.

Any page where you can add one script tag: your own site, a landing page, a web app, a documentation site. Some hosted checkouts do not allow scripts on their order confirmation page; there, put the survey on your other pages or send its link in the order e-mail.

By limits, not by features you have to unlock one at a time. Each plan includes everything in the plan below it: Free runs 1 survey with 100 responses a month, Basic 5 surveys with 5,000, Pro 20 surveys with 100,000. Seats, how many insights we recalculate per survey, how often they refresh, and how long we keep your data all move up the same way. Recomputing insights by hand is on or off instead: Free does not include it, Basic and Pro do, and each run uses analyzed answers from the plan's monthly pool. Yearly billing is charged as ten months instead of twelve, so two months are free.