Face verification for rewards and offerwall sites: one face, one account

Short answer

Ask for a short video selfie only where money is about to leave: before a cashout on a rewards site, or when a user clicks an offer on an offerwall network. Check that one live person turns their head in a random order, then compare the face with your other accounts. Treat a match as a reason to review, never as an automatic ban.

Where to ask

Before a cashout on a rewards site; at the offer click on an offerwall network, while the user is on the wall.

What it proves

One face in every frame, the same person throughout, and a head turn in the order asked.

What it finds

The same face behind other accounts of your site, or of other sites, which stay unnamed.

What it keeps

An encrypted face print for 12 months. Photos are never stored.

Why a face, when device checks already exist

Most multi-accounting is caught by the device: the same phone, the same browser profile, the same payout address. Farms know this. They reset phones, rent cloud phones by the hour, rotate residential proxies and buy cheap second-hand devices, so each account looks like a different person on a different device.

A face does not reset. One person running twenty accounts has one face, and asking for it at the moment money moves turns twenty clean-looking accounts into one person again. It costs the honest user twenty seconds, once.

When to ask for it

Asking everyone at signup loses users who were never going to cheat. Ask where the money is:

On a network, narrow it down by publisher, offer and country: the publishers whose traffic already reverses, the high-payout offers, the countries where farms work. On a rewards site, start with first cashouts and widen it if the matches show a ring.

What the check proves

1. One live face. Every frame must show exactly one face. A second person or a photo held next to the camera fails.
2. A head turn in a random order. The user looks straight, then turns left and right in an order chosen for that session. A video recorded for another session does not fit.
3. The same person throughout. Each frame is compared with the straight-on frames, so a face swapped halfway through fails.
4. A real camera. Virtual camera programs (OBS, ManyCam and similar), which feed a recording into the browser, never pass.
5. The same face elsewhere. A face print is computed from the straight-on frames and compared with the prints of your other accounts, and optionally with other sites, whose names you never see.

What it cannot catch, and how to use it fairly

• Look-alikes. Twins and close family can produce similar prints. A match is a reason to look at the accounts, not a verdict.
• Advanced injection. A determined attacker can feed a manipulated video into the browser. The random head-turn order and device checks raise the cost, but a check run in the browser is not a certified liveness test.
• People without a camera. Some users have no camera or refuse. Give them another route, such as a slower manual review, instead of a dead end.

Privacy: what users agree to and what is kept

A face print is biometric data in most privacy laws, even when it is only numbers. Treat it that way:

• Ask first. A consent screen before the camera opens, saying what is compared, with whom, and for how long.
• Keep the print, never the photo. Frames are used during the check and dropped; only an encrypted face print is kept.
• Delete on time. Twelve months is a sensible limit for fraud prevention; delete earlier when the account is closed.
• Respect local rules. Some US states have strict biometric laws. Until your lawyer has read your consent text, do not offer the check there.

Rolling it out without losing honest users

The first weeks decide whether a face check helps or hurts. Roll it out in steps, and let the numbers tell you when to make it stricter:

1. Record only. Ask on first cashouts (or offer clicks on a network) and change nothing in the decision. Watch how many users pass on the first try, how long it takes and how many give up.
2. Read the matches. Open the accounts that share a face. If most of them also share devices, payout addresses or a signup burst, the check is finding rings your other signals already suspected, and you can trust it more.
3. Hold, then block. Move "failed three tries" and "same face on other accounts" to review first. Block only once the reviews keep confirming them.
4. Explain it. Tell users in one sentence why you ask and what you keep. Support questions drop, and the people who refuse to show a face at the cashout are rarely the ones you want to pay.

Measure the false rejections by device too: old phones with weak cameras and poor light fail more often. Give those users a second route, such as a manual review, so a bad camera never costs an honest user their balance.

Face ID in CashoutGuard

CashoutGuard runs the check on its own servers, with no third party involved. You turn it on in the dashboard (Face ID page) and choose the events, how often, for whom and what the result changes: nothing, review or block.

• When an event matches your rules, the answer of /v1/evaluate carries face_check with the link to send the user to (docs).
• The face.completed webhook tells your server when a check ends, with the accounts of your site that share the face (webhooks).
• Each account page shows its check and the same-face accounts, next to the device links and the cloud phone signals.

It is a pilot, off by default. Start by only recording the results, look at the matches for a week, then decide what they should change.