Know what customers would say before you -
Build.Set Price.Launch.

Simulated respondents are directional: they show which way opinion is likely to lean and where segments differ, not a measurement of real opinion. They are useful for spotting likely reactions, differences between segments and confusing questions quickly and cheaply. They can reflect biases of the underlying AI models, so validate important decisions with real respondents. Every HumanPanel summary includes these caveats.

In three steps. First, an AI model maps your target population into worlds and segments, each with weighted attributes such as age, education, occupation, income, politics, media habits and a few traits specific to your topic. Second, code, not the model, seats exactly the number of respondents you asked for in proportion to those weights, from a seed, so the same seed gives the same people. Third, each seat is expanded into a full person with a name, a life story, beliefs, blind spots and an honest picture of what they know about your topic. That person then answers the survey.

No. HumanPanel is for piloting and exploring: testing wording, comparing ideas and forming hypotheses before you spend on research with real people. For decisions that matter, confirm what you learn with real people.

A synthetic survey panel is a group of AI-simulated respondents that answer a survey the way people with specific backgrounds plausibly would. HumanPanel builds up to 1,000 of them for each run, spread across population segments you describe, and each one explains their answer in their own words.

The first answers usually appear within seconds, and you can watch respondents join and answer live. Small panels typically finish in minutes; larger panels take longer. HumanPanel emails you when the results are ready.

Sign up for free with just your email address: no password and no credit card needed. You can write your first survey right away, describe who should answer it, and start a run when you're ready.

Signing up and writing surveys is free, and every new account gets $1 of credit to get started. Runs are pay as you go: each one is charged what its AI calls actually cost, plus a margin, so a small panel costs less than a large one: most runs cost $2 to $5, and asking the same people a new survey costs cents. The most a run can cost is shown before you start and reserved from a prepaid wallet you top up by card; only the actual cost is charged when it completes. If a run fails, nothing is charged.

Yes. Give HumanPanel your website address, or paste your homepage text, and say what you want to learn. It writes the title, describes who should answer and drafts the questions, all of which you can edit before running. Writing a survey this way is charged from your wallet at what the AI call costs plus a small margin, usually a few cents.

Up to 20 questions per survey, in five types: single choice, multiple choice (with an optional maximum), yes/no, true/false and free text. Free-text answers are summarized into themes with verbatim quotes.

Download every answer and reason as a CSV file from the Results tab of a finished run, with each respondent's world, segment and traits. Or copy the full results with a ready-made prompt for ChatGPT, Claude or any AI assistant.

Yes. You describe the target population, for example "working adults in United States across income levels", and choose how many respondents (up to 1,000) and how many worlds (1 to 10). HumanPanel designs segments within that population and reports results for each one.

No. HumanPanel does not use your surveys or results to train AI models, and does not sell your data. Survey content is sent to AI model providers only to generate your run.