The first 3 interviews are free · Start free or see the prices

How to Find Fake and Careless Participants in AI Interviews

Paid studies attract people who pretend to qualify. These rules help you stop them at the screener, find them in the interview and replace them fairly.

Overview

A fake participant claims to be someone they are not, to get paid. A careless participant is real but gives rushed or empty answers. Both reduce the quality of a study. An AI moderator will not notice either one unless you design for it.

Prevention starts in the screener and continues with one or two check questions in the interview. After the session, read for signals across sessions, not only inside one transcript.

In User Evaluation, you can flag a session for replacement with a reason: "Not who we asked for", "Low effort answers", "Audio or video problems" or "Something else". The session leaves the study report, a new person can take the place, and the replaced participant is not paid.

Rules

  1. 01

    Do not show the qualifying answer in the screener.

    A yes/no question tells a fake participant which answer gets them in. Use a list with several real options, plus one option that does not exist.

    Do not

    Do you use project management software at work? Yes / No

    Do

    Which of these tools did you use at work in the last 30 days? (Asana, Trello, Jira, Planmore, None of these)

  2. 02

    Add one open screener question that only a real member of the audience can answer well.

    A real user gives a specific, slightly messy answer. A fake gives a generic one. In User Evaluation, an AI condition can hold such answers for a check instead of accepting them at once.

    Do not

    Why are you interested in this study?

    Do

    Describe the last task you did in that tool. What was it for?

  3. 03

    Ask one check question in the interview that repeats a screener fact in a new form.

    A fake participant can learn the right screener answers. It is harder to describe the real steps of a task in their own words, out loud.

    Do not

    Note: none. The guide trusts the screener.

    Do

    Walk me through how you add a new user to the tool. Note: if they cannot name the steps, do not push. Record it.

  4. 04

    Know the signals of a fake voice session.

    Watch for answers read aloud from a script, and for a long pause before a fluent answer. Also watch for answers with no names or dates, and facts that do not fit the screener. One signal is not proof. Look for several.

  5. 05

    Know the signals of a fake text session.

    Watch for perfect structure, lists and generic language. Long answers that appear faster than a person can type are a signal. So are the same phrases in sessions from different names.

  6. 06

    Compare sessions with each other.

    Some fraud shows only across sessions. Look for the same voice, the same unusual story or the same phrases under different names.

  7. 07

    Separate a careless participant from a weak question.

    If many real participants give short answers to the same question, the question is the problem. Flag a session only when the answers are short across most of the guide.

  8. 08

    Flag with evidence, and write it down.

    A replaced participant is not paid. Be fair: pick the correct reason and note the moment that shows the problem, with its time in the recording.

    Do not

    Flag: Something else. No note.

    Do

    Flag: Not who we asked for. At 04:10 they could not name any step in the tool they said they use daily.

Checklist before launch

  • Screener choice questions hide the qualifying answer among real options.
  • The screener has one option that does not exist, set to disqualify.
  • One open screener question needs real experience to answer.
  • The guide has one check question that repeats a screener fact in a new form.
  • You know the voice and text signals to look for.
  • You plan to compare sessions with each other, not only one by one.
  • Each flag has a reason and a note with the time in the recording.

Questions

It depends on the audience and the reward. Studies with high rewards and hard-to-reach audiences attract more attempts. Design every paid study with checks.

Not on its own. It follows the guide. Add check questions to the guide, and read the transcripts for the signals yourself.

A trap question includes an option that is not real, such as a tool that does not exist. A person who selects it is not a reliable participant.

In User Evaluation, the session leaves the study report and a new person can take the open place. The flagged participant is not paid.

Ask real people your question

The AI moderator interviews screened people by voice or text. Each claim in the study report links to its clip. The first 3 interviews are free.

Start free