How it works

The short version, without the statistics.

1. You bring your questions

Type or paste the questions you plan to ask, with their answer options: agree–disagree statements, yes/no questions, "select all that apply" lists, number answers such as hours worked. You do not need a scoring key or any data.

2. You say who should answer

Pick the people you want to hear from: an age range, sex, whether they work, education, states, and, if you like, a plain-English description such as "people who work in healthcare". You choose how many rows you want.

3. We start from real Census records

Every synthetic respondent is based on a real, anonymous record from the US Census Bureau's American Community Survey: millions of US adults, with their age, sex, education, state and job situation. We draw people from those records in the right proportions, so your file looks like the population you asked for, not like an AI's idea of it.

4. AI reads your questions

An AI model reads each question and answer scale and predicts two things: how people like these usually answer it, and which of your questions tend to be answered alike (for example, someone who agrees with one anxiety question usually agrees with the others). The second part comes from our own model, trained on how hundreds of thousands of real people answered real surveys. More in the FAQ.

5. A statistical model writes the answers

A statistical model turns those predictions into individual answers, person by person. People differ from each other, they are a little inconsistent, and some use the ends of a scale more than others, as real respondents do. The AI does not role-play each person; that approach makes answers too tidy.

6. You download your file

You get a spreadsheet (one row per respondent), a codebook, and a short read-me that says honestly what the file is worth. Most studies take 5–15 minutes.

What it is for, and what it is not for

• Good for: trying out a questionnaire before you pay to field it, spotting questions that don't hang together, planning your analysis, teaching, and practice data.
• Not a replacement for real data when a decision rests on the answer. A file made from questions alone is about as informative as a small real survey (see what works best), and its rows are not real people: don't run significance tests on it as if they were.