Online dating is as old as the internet. It has exploded in popularity in recent years: roughly 40% of straight couples now meet online. It has given women more safety and choice as they participate in the dating market, and brought down divorce rates as people date more widely and make better-informed decisions than in the past.

However, these benefits are effects of online dating becoming more common, not better. The recipe for online dating itself has been static since the 90s: some pictures, a quick description of yourself and what you’re looking for in a partner, and endless looking through other profiles. No matter whether you’re on Hinge in 2020 or OKCupid in 2005, the structure is the same. Only the volume is different.

Disappointingly, innovation in online dating has mostly been marginal. There are hundreds of dating apps, but differences are slim. Some limit the number of daily matches, others force someone to chat first, others yet differ in their question prompts to the users, etc. These are micro-optimizations, but of minimal value because the existing model is generally suboptimal.

What’s broken in the current model is that it encourages preference falsification. Normally, preference falsification is very overt – for example, when someone lies to their friends about which political candidate they’re going to vote for. In online dating, it’s subtle – people are usually not actively lying, but they are (often unknowingly) misrepresenting, to themselves and to others, who they are and what they’re looking for.


People struggle to represent themselves accurately for two core reasons:

  1. Self-knowledge is hard
  2. People try to make themselves appear as appealing as possible.

The first point is obvious, though unpleasant: lots of people go through life doing things they hate, or simply killing time. Few manage to ultimately find their passions. I’m sure you’ve met people who were all about Hobby X, until they dropped it one day because it didn’t bring them joy and they were doing it for some other reason. The second point is twisted game theory: people are insecure in themselves, so they try to imagine what the opposite gender might most like to see, and then to ape that.1

Consequently, dating app profiles are full of wildly wishful descriptions – surfer, investment banker, loves reading, passionate about sunsets. When you meet the real person, it turns out they don’t care about sunsets and put that in purely as a romantic signal, and they’ve been surfing just once a year ago but they’d sure love to try it again. They actually spend 14 hours a day working and the remaining 2 watching King of the Hill, but they can’t write that – what would the other people on the app think? Surely they’re only interested in surfing, sunset-loving, well-read investment bankers.

These dating app bios are not just unrealistic to appeal to potential matches, but also as constructions for the users themselves: no-one wants to admit to themselves that they are not living quite the life they aspire to. Thus, when you give them a paintbrush, they paint an idealized version of themselves. Reading is the dead giveaway: barely anyone actually reads books, but virtually everyone says they love reading. What they love is the idea of themselves being well-read. Consequently, a love of reading is an extremely common bio line, even when most people spend less than 1% of the year reading.

False Preferences about Partners

People also struggle to accurately identify what they’re really looking for in a partner. Again, there are two main reasons:

  1. Filters are necessary on dating apps
  2. Knowing who is a good fit for you is another, even harder form of self-knowledge

Because of the enormous number of people on dating apps, having filters is the only way to manage the quantity of potential matches. The trouble is that people usually have only weak priors to base those filters on. This is especially bad for younger people, who haven’t dated very much, and who might even fall to peer pressure dictating who they should date. They filter hard, and consequently never end up dating broadly, which would be the only way to figure out what kinds of people are good fits or not.

This leads into the second reason – you try to filter for people who will be good fits, but that’s extremely difficult to know a priori. I’ve been personally surprised by this many times over. Every time I thought I knew exactly what kind of woman I wanted to date, life threw me a curveball and I fell for someone totally outside that box. By the time I was in my mid/late-20s, I learned to filter less: romantic attraction and compatibility was actually very difficult to predict without really getting to know the other person. I learned that common interests didn’t mean much, whereas common attitudes seemed to be the strongest predictor of a fit.

But many people in the dating world aren’t having those experiences now. They filter so specifically that they never learn who else might be out there. I’m reminded of an acquaintance of mine – she spent the better part of three years aggressively dating only people more successful than her – and not a single one worked out. I was baffled that at no point did she ever consider whether her “I am only dating CEOs” mantra was the wrong filter.

Outcomes and Solutions

The dating pool is large and liquid, but the core issue is that individual participants are hard to match because (1) they falsify who they are and (2) they falsify what they want in their partners. This makes it profoundly difficult to generate matches at a better-than-random rate. Both of these issues are fundamentally due to the user having too much choice, both in how they describe themselves and how they choose others. Importantly, these are not really blamable faults of the users, and it’s clear that the current generation of dating apps will not remedy these problems.

I am hopeful that the next generation of dating apps will solve this issue by relying more on large-scale, latent data: I am imagining an app where an amalgamation of your Reddit, Spotify, Twitter, Facebook, Google Search, etc. histories know you better than you know yourself – and that can be used to match you perfectly with a partner. I’ve been surprised that I haven’t seen such technology in the wild just yet. Probably the greatest obstacle to building a matching engine leveraging such data is that there isn’t any good training data to use, but I would’ve expected some of the big incumbents to use their datasets for this purpose by this point. We will see.

  1. The great tragedy in distorting the self is that it doesn’t work at all. The real you has to come out eventually, at which point the ruse is up. There is a great irony in users misrepresenting themselves: they are selecting against people who might like their true selves. Suppose they do match with the person that loves well-read, sunset-walking surfers, and then their date eventually finds out – regardless of what they claim to be passionate about – that all they really do is work, eat croissants, and watch TV. It’ll have been a waste of time.