> ## Documentation Index
> Fetch the complete documentation index at: https://mr.verifyyou.com/llms.txt
> Use this file to discover all available pages before exploring further.

# How the check works

> Liveness and uniqueness in a few seconds, no government ID, and the honest line on what it does not do.

A human check is the step you put in front of an action that confirms a real, live, unique human in seconds, and returns a simple verified or denied. It proves humanness and uniqueness. It does not prove a name or a legal identity. Here is the whole of it.

## Two questions, a few seconds

Real and unique are really two questions, and the check answers both in the same moment in the browser.

<Steps>
  <Step title="Liveness">
    Is a live person present right now, and not a photo, a recording, or a deepfake. This is what stops a bot or an AI-generated respondent from clearing the door.
  </Step>

  <Step title="Uniqueness">
    Have you already seen and paid this person before. The new respondent is compared against the ones you have already verified, which is what catches duplicates and survey farms.
  </Step>
</Steps>

A duplicate fails on uniqueness, a bot fails on liveness, and a survey farm fails both the moment the same face appears again.

## You decide what counts as unique

Uniqueness is only useful if it matches the way you run studies, so it is yours to scope. Ask for uniqueness per survey, per study, or per panel, whatever bucket fits the question you are answering. Need one response per person within a single survey? Scope it there. Protecting a standing panel from the same human enrolling twice? Scope it to the panel.

## Genuine repeat respondents are not punished

A real worry with any verification step is that you will add friction for your good participants. You do not. A respondent who passes carries a portable credential across your surveys, so a genuine returning participant is recognized rather than asked to prove themselves from scratch. The work happens on the first visit. After that it is continuity, not a fresh hurdle.

## Put the check right before the incentive

Where you place the check matters as much as the check itself. For conversion, the placement to reach for is right before the incentive or the payout. By the time a respondent gets there they have done the survey work and want to finish, so real people push through, while the accounts that balk at putting a live, unique face on the line at that moment are usually the ones you were about to pay for nothing. Placement turns the check into a filter on who collects, not a hurdle early in the flow.

## The honest boundary

<Note>
  A human check is additive to your quality stack, not a replacement for it. It sits alongside your attention checks, speeding detection, and geo and device signals, and answers the question most of those cannot: whether there is a real, unique, live human behind the response at all. It is not KYC and does not prove anyone's name or legal identity, on purpose. For survey work you almost never need to know who someone legally is, only that they are a real person you have not already paid.
</Note>

## What we keep

To recognize a returning respondent, we keep a face vector, a one-way numerical representation of facial geometry, an array of numbers rather than an image, checked against the ones already held. That number is not a picture and, on its own, is just a string with no name attached. What your dashboard sees is a verdict, verified or denied, never a face. The full detail on what is collected and how long it is kept lives in our privacy policy, which is the document to rely on for retention specifics.

## The easiest way to see it

The best way to know whether the numbers hold for your audience is a free pilot with a branded share link, a check you push to a batch of your own respondents with no engineering work, so you can watch real people run through it before writing integration code. If that is useful, [book a chat with us](/meet-us).