AI has made faking a respondent easy. Verso Shield checks every interview while fieldwork runs, before anything reaches your insights or gets paid.
Fake and careless respondents are the problem research buyers rarely ask about and suppliers rarely talk about. They don't look fake in a table. They look like data.
Most platforms check people once, before the interview, then trust whatever comes next. But the tells show up during the conversation: in the rhythm of the speech, the detail of the stories, and the same story told by two people who have never met.
So we check every respondent, on every answer, across every panel, and a researcher looks at every doubtful case.
Verso platform data, all client studies from August to early November 2026, test interviews excluded. Rejections cover fraud and unusable answers alike.
Most platforms screen at the door and hope for the best. Each layer catches what the others can't, and together they decide which interviews count toward your sample.
Device, network, location and identity are matched in real time against the declared profile. One person, one interview.
We see across panels, not just inside one. A respondent caught once is flagged on related studies, and their answers are checked against their own history.
Every pause, rush and shortcut tells a story. Speeders, autopilot answers and disengagement are flagged as they happen.
A real, present person on camera from start to finish, in a quiet place and alone. Doubtful cases are reviewed on video.
Real minutes spoken, not just time connected. Background noise, other voices and someone whispering the answers.
ChatGPT has a signature, and so do stories copied between respondents. We spot both, within an interview and across all of them.
Every interview gets one quality score from these six layers. Doubtful cases go to a researcher: no one is rejected by an algorithm alone.
The signals behind the six layers. All of them run on every Verso study today.
Your fieldwork view shows which interviews count and why the others don't. Nothing disappears silently, and the panel gets a reason it can't dispute.
| Participant | Status | Reason | Score |
|---|---|---|---|
| P-184 | Accepted | Specific stories, natural pace | 88 |
| P-185 | Rejected | Under 20 words per answer · finished in 2 min 41 s | 21 |
| P-187 | Rejected | Same story as P-142, word for word | 30 |
| P-190 | To review | Written-style answers, reading off-screen on video | 52 |
| P-191 | Accepted | Consistent with earlier study | 81 |
No one is rejected by an algorithm alone. Doubtful cases go to a researcher.
Written, specific and shared with the panel, so it can be checked, not argued.
Rejected interviews are replaced at no cost and never count toward your sample.
We work with several panels, so we see what a single panel can't see about its own members.
"Could you reject a real respondent by mistake?"
That's why no one is rejected by an algorithm alone. The checks raise doubtful cases; a researcher reads or watches the interview and decides. A shy person who answers briefly is not a fraudster, and we treat them differently.
"What happens to the interviews you reject?"
They are removed from your sample and the analysis, the panel receives the reason, and a replacement is recruited. You only pay for the interviews that count.
"Is fraud really that common?"
It depends a lot on the panel and the audience. Across our client studies, about one interview in nine does not hold up, and up to one in three on the hardest panels, mostly for poor quality rather than proven fraud. Without the checks, all of those answers would reach your analysis.
"Does it cost extra, or need any setup?"
No. Verso Shield runs on every study by default, with no setting to turn on and no extra line on the quote.
Ask every supplier, us included. We'll show you how Verso Shield would run on your next study.
Book a call