What Dating App Algorithms Actually Do With Your Swipes

Most people imagine the algorithm as a well-meaning librarian: somewhere in the machine, their preferences are being studied, and candidates are being fetched accordingly. The reality, documented in patents, leaked internals, and the published research of the companies themselves, is closer to a market maker. Understanding what actually happens to your swipes explains most of what feels wrong about the apps, and none of it requires a conspiracy theory. It only requires reading the incentives.

You are being scored, not understood

The foundational mechanism in swipe matching is desirability ranking. Early versions were literally Elo, the chess rating system: when someone highly swiped-on likes you, your score rises more, exactly like beating a grandmaster. Modern systems are subtler but do the same job: every swipe on you is a vote, and your aggregate becomes a rank. Then comes the consequential part: ranks are matched assortatively. You are shown, mostly, people whose scores resemble yours, whatever your stated preferences say. The librarian was never reading your list of values. The market maker is pairing you by photographic market price, which is why the people you see feel weirdly interchangeable: they are not selected for you, they are selected for your tier.

What can the system actually learn from a swipe? A half-second judgment on a photograph. Not kindness, not direction, not conflict style, not whether they want children. The input channel physically cannot carry the information that decides relationships, so no downstream cleverness can recover it. The algorithm is not bad at finding your person. It was never looking.

The engagement thermostat

The second thing your swipes feed is pacing. These are engagement businesses: the metric that pays is time in app, and match delivery is the reward lever. So matches arrive on a schedule tuned to keep you playing, a dry spell, then a small flurry, exactly the variable-ratio pattern that maximizes compulsive checking. Many apps openly rate-limit visibility and sell the bypass back to you: boosts, super likes, priority placement. Read that structure plainly: scarcity is manufactured, then priced. Your frustration is not a malfunction of the product. In a real sense it is the product.

None of this requires malice from the people building it. Optimize any system for engagement and it will discover, on its own, that unresolved longing retains better than satisfied users who leave in couples.

What an honest algorithm looks like

An algorithm aligned with you would have to invert every choice above. Input: the material that actually predicts, stated life direction, values, family intent, conflict style, a voice, not a face ranked in half a second. Objective: not engagement but outcome, did two people meet, did it hold, with the system graded on that answer. Delivery: few and deliberate rather than many and addictive, because a real candidate deserves attention that a stream destroys.

That inversion is, concretely, what 30andme runs. Matching scores trait alignment on the deciding dimensions, introductions arrive at most twice a day, one connection holds at a time, and five days into every connection we ask both people the only question that grades us: did you meet? The answers tune the engine. We win when you leave, which is the only incentive structure under which you should trust anyone's algorithm with your love life.

The takeaway is not that algorithms cannot help you find someone. It is that an algorithm helps whoever sets its objective. Before trusting one with your evenings, ask the only question that matters: what does this thing get graded on? If the answer is your attention, you are the crop. If the answer is your absence, you might be the customer.