FirstDate: What Singapore's Government Matchmaking Pilot Gets Right

Where this lines up with how Vera works

FirstDate's own site describes its intake plainly: “Our questionnaire covers users' interests, habits, values and preferences.” Preferences here folds in more specific fields too, like religion, age range, and dealbreakers, per how the pilot's been reported. That's not a controversial list, and most dating apps ask for far less, usually a set of filters for what someone thinks they want rather than what a relationship actually needs. It's close to what any matchmaker, human or AI, has always known matters more than a profile photo, and it's the same territory Vera, our AI matchmaker, works through with every person she talks to.

The other place FirstDate gets it right: it's built to get two people to an actual date, not to keep them messaging indefinitely inside the app. A match, a short accept window, contact details released only once both sides agree. No endless back-and-forth before anyone commits to meeting. That's a shared instinct worth naming, because a lot of dating products these days optimize for keeping people inside the app rather than getting them out of it.

Where compatibility goes deeper than the four categories

FirstDate's four categories, interests, habits, values, and preferences, are a genuinely strong starting point, further than most dating apps go. Relationship research has pointed for years at what tends to sit one layer below that: how two people communicate, especially under stress, and attachment style, the patterns learned early on around trust and closeness that shape how someone actually shows up once the early stage wears off. Two people can share the same values on paper and still struggle if their conflict styles clash, or if one person's attachment patterns make sustained closeness hard to trust.

That layer is hard to put on a form, mostly because most people don't have ready language for their own attachment style or communication patterns without some guided reflection first. It tends to surface through conversation instead, which is part of why Vera's intake keeps building over time rather than stopping at a single questionnaire. Values and interests come first, and the deeper layers, including how someone handles trust and conflict, build in alongside the first introductions rather than gating them.

Where a ranked algorithm runs into a real limit

FirstDate's matching uses the Gale-Shapley algorithm, the same math behind stable school-placement and residency-matching systems. It's elegant for problems where preferences genuinely rank on a single scale, your first-choice school really does beat your fifth choice. Worth asking, gently, whether relationships work the same way. Two people's values, or their idea of the life they want, aren't usually a case of one ranking higher than the other. They're often just different, in a way that isn't obviously reducible to a preference order. That's a structural question about applying rank-based matching to compatibility in general, not a claim about how any specific implementation scores anyone, since that detail isn't something outsiders can see.

What a one-time questionnaire can't quite catch

A survey captures what someone says about their values on the day they filled it in. A conversation can catch more than that, because how someone actually talks about a past relationship, what they bring up unprompted, where they hesitate, often carries more signal than a form field ever will, and people can express the same thing differently depending on the day they're having. It's also two-way. Vera can follow up, ask a clarifying question, or gently push back when an answer feels more like the polished version than the real one, something a static questionnaire has no mechanism to do.

FirstDate does collect feedback after each date too, which is a good instinct. The real difference isn't whether feedback happens, it's cadence: FirstDate's matching runs in fixed cycles, a new round computed at set intervals, while Vera keeps adjusting continuously as the conversation continues, rather than waiting for the next batch to run.

None of this makes one approach right and the other wrong. It's genuinely encouraging to see the government treat this as seriously as it has. If you want to see what a continuous, conversational version of the same idea looks like, talk to Vera.