Questions about
synthetic research

A digital respondent is an AI persona with a complete profile — age, city, income, education, household, habits and life context — that answers your questionnaire in character. It is not a generic chatbot given a costume: the profile is built from the segment parameters you set, and the answers are generated to stay consistent with that profile across every question in the test.

A chatbot gives you one answer from one averaged point of view. FeedbackEngine runs a whole panel: each respondent holds a distinct, internally consistent profile, answers independently, and the results come back as a distribution — means, Top-2-Box shares, and grouped open-ended answers — rather than a single opinion you have to trust.

It replaces the part of traditional research that was too slow or too expensive to run at all — the pretests and screenings that get skipped because they take weeks. For high-stakes final decisions, the sensible pattern is to use synthetic testing as a filter and live research to validate the finalist.

Digital respondents reproduce reasoning and stated preference well, and sensory experience poorly. They will tell you whether a film idea is clear and distinctive; they will not tell you how a fabric feels or how a drink tastes. Group dynamics — the thing a focus group is actually for — are also outside the model by design, which is a feature when you want independent answers and a limitation when you want a discussion.

Six test types cover the common cases: static creatives and key visuals, names and slogans, ideas and concepts, video ideas from a script or storyboard, open audience surveys with no material at all, and side-by-side comparison of a longer short list.

Minutes from launch to a finished report. Setting the test up — writing the segment and uploading the material — usually takes longer than running it.

Through structured parameters — gender, age, cities and regions, income level, education, marital status and children — plus a free-text description of behaviour, which is where most of the precision comes from. Segments are saved, so you can go back to the same audience for a follow-up test.

JPG, PNG, GIF, PDF and DOC, up to 8 files per idea.

No. Each test type ships with a questionnaire built for it, following the standard methodology for that kind of pretest. The Insights test is the exception, and there by design: you write up to five of your own questions, open-ended or scaled.

A summary of the findings, scaled results per metric, an overall ranking of the variants tested, and the open-ended answers grouped by sentiment — positive, neutral, negative — with verbatim quotes.

Top-2-Box is the share of respondents who gave a 4 or a 5 on the 1–5 scale. It is the standard headline metric in pretest research because it tracks the people who actually reacted positively, rather than a mean that a mass of indifferent 3s can quietly prop up. Reports also carry the mean and Top-1-Box.

Read the open-ended answers. A uniformly low ranking is a finding: it usually means the problem is upstream — in the idea or the message rather than in the executions — and the verbatims tell you which.

Yes. Reports are available in the app and can be exported for sharing and for the decision deck.

Yes — the first test is free as a demo, with a reduced number of ideas and questions. It is a real test on a real panel, not a mock-up.

Create an account, pick a test type, define your segment and launch. There is nothing to install and no sales call in the way.

Plans and panel sizes are set out on the pricing page. Larger panels, deeper segmentation and API access come with the higher tiers.