The number doing the heaviest lifting here is a share of workload, not a count of scammers. Ninety-eight percent [1] describes where Tinder's moderation actions landed. It does not say how many fake accounts reached a first message, and those are different measurements. The residue is the part product teams should read closely: everything moderators do about real people mistreating other real people has to fit inside the remaining two percent of actions [18].
The harm underneath is real and has a bill attached. The FBI figure WIRED cites, more than $4 billion in US romance scam losses between 2015 and 2025 [6], averages out to at least $400 million a year across that decade [19], and nearly half of current US online daters have been targeted by a dating scam [7]. That is why the check spread to more than a dozen major dating apps and sites, using liveness checks, 3D authentication, video selfies or biometric ID at the door [5].
What the check proves is narrower than the badge implies. A liveness check establishes that a person was in front of the camera at sign-up, and not that they were not altering their appearance with AI in real time; an ID scan establishes that an ID was presented, not whose it is [14]. Cybersecurity researcher Katie Moussouris told WIRED that normalizing a face scan as assurance of identity or age is security theater [13], and that a verified profile can still belong to a scammer who took the account over later [15].
Here is what users actually do: at sign-up they hold the phone up and turn their head, because Face Check has been mandatory for new Tinder members since October 2025 and there is no non-biometric door [3][20]. Here is what teams tell themselves users do: read the badge as information about the other person. Match's own framing reaches for the second reading, with trust and safety head Yoel Roth calling it "a statement about what human connection online should require" [8]. The rollout to existing users in the US and UK [4] is where that statement meets people who signed up under different terms.
Anyone inheriting this expectation has to answer a design question the badge hides: where the artifact lives and how long. Match says Face Check stores an encrypted map of the shape of the face rather than the photo [9]. Tools for Humanity, whose optional World ID has been on Tinder since an April partnership [10], says its orb sends the coded image to the phone and then deletes it [11], and chief product officer Tiago Sada calls the handing of personal data to every individual app "super creepy" [12]. Both are vendor claims about deletion, and dating apps do get breached [17]. A leaked password gets replaced, which is what makes a leaked face map a different class of asset [16].
So the forcing function is two written answers, and they make a usable 2x2. First axis: does the check move a number you already track, such as accounts removed before a first match, or reports per thousand conversations. Second axis: can you state what is stored and when it is deleted in one sentence you would accept seeing quoted in a breach notice. Measurable number plus short retention is a control. Measurable number plus indefinite store is a liability you took on with your eyes open. No number plus short retention is a badge, which is cheap and mostly harmless. No number plus indefinite store is the quadrant that reads badly on Friday, and it is the one an industry default pulls teams into by inertia [21].
