How We Use AI

Most dating apps are coy about this. We'd rather just tell you: there is exactly one place where 50/50 uses automation on anything you give us, and it is a safety check on profile photos. Everything else — who you see, when matches are revealed, what your response rate says, whether your verification selfie passes — is either a fixed rule or a person. This page explains both halves: the one thing that can be automated, and the things we have deliberately chosen not to automate.

The one thing that can be automated

Profile photos go through an automated explicit-content check. When that check is switched on, a copy of the photo you just added is sent to Microsoft Azure AI Content Safety, a third-party service that processes it on our behalf. It rates the picture for four things — sexual content, violence, self-harm, and hateful content — and returns a severity score for each. Those four scores are all we receive back. We do not receive, and it does not send us, any description of who is in the photo or what they look like. While the check is switched off, no photo leaves our own storage for it.

What we do with those scores:

• A high score hides the photo until one of our admins looks at it. You still see it on your own profile, marked as under review, so it never silently disappears on you.
• A middling score leaves the photo visible and adds it to our review queue for a person to check. This is deliberate: swimwear and beach photos are normal, and we would rather a human make that call than hide something over a borderline number.
• Anything below that, nothing happens and no one is notified.

The check runs after your upload finishes, so it never blocks or slows down adding a photo, and a problem on their end never prevents you from posting one. The automated check can hide a photo pending review, but it can never permanently remove a photo, and it can never suspend or ban an account. Only a person can do either of those.

It is not face recognition

This is the distinction that matters most, so we want to be exact about it. The check rates content, not people. It does not detect, measure, match, or remember a face. It cannot tell whether two photos are of the same person, it is not comparing you to any database, and it produces nothing that could identify you.

We do not create a faceprint, face scan, face template, or any other biometric identifier — of anyone, ever. This is a deliberate choice rather than a feature we haven't gotten to yet, and we do not intend to change it.

Your verification selfie is never run through this check, or through any automated system at all. Verification is a person comparing your selfie to your profile photos by eye. There is more about how that selfie is handled — who sees it, how long we keep it — in our Privacy Policy, and we show you the same explanation in the app before you take it.

What we don't automate

These are not oversights or things on a roadmap. Each one is a decision, and if any of them changes we will update this page and ask you again before the change affects you.

• Your messages are not scanned or read by any automated system. The photo check is wired to images only. Messages are encrypted where we store them, and a person only ever reads a conversation in response to a report — every one of those reads is logged.
• Nothing in 50/50 is written by AI. There are no AI profiles, no AI chat partners, and no AI-generated replies or openers. Every profile in the pool is a person, and every message you receive was typed by one.
• Matching and reveals are fixed rules, not a learning model. Your deck, the three daily reveal slots, and your response rate all follow rules we could write on a napkin. There is no model ranking you, scoring your desirability, or deciding who deserves to see you.
• Verification is human. A person compares your selfie to your photos. No automated system approves, rejects, or scores that comparison.
• Moderation decisions are human. Automation can flag a photo for review; only an admin removes content, suspends an account, or bans someone.

We don't train AI on you

We do not use your photos, prompt answers, or messages to train AI models, and we do not sell or licence your content to anyone else for that purpose. Our Terms of Service commit us to this in writing. Your verification selfie is used only for verification, as described in the Privacy Policy.

Why this isn't something you can switch off

We ask you to acknowledge this before you use 50/50, and we record that you did. What we do not do is dress it up as a choice it isn't: the photo check cannot be turned off for an individual account.

The reason is straightforward. It exists to keep explicit images out of a pool that other people browse, including images sent to people who never asked to see them. An opt-out would be used first and most by exactly the accounts the check is there to catch, which would leave everyone else less safe while looking like a courtesy. We'd rather have one honest safety floor than an optional one.

You are of course free to decline it — by not using 50/50, or by deleting your account, which you can do at any time from your profile and which removes your photos with it.

If this changes

If we materially change what is automated — for example, if we ever build the practice feature we've discussed, which would involve generative AI and a new provider — this page is revised and the date above changes. We will also ask you to read and accept the new version in the app before the change affects your account. An acknowledgement you gave for this version does not carry over to a different one.