# AI-Native Dating: Product Exploration > **Status:** Workshopping. Preparing for conversation with Known. ## Theses Under Consideration | Thesis | Core Question | |--------|---------------| | **Our tools know us better than we know ourselves** | When stated preferences diverge from revealed behavior, who's right? | | **Dating apps optimized for the wrong metric** | Swipes and matches are proxies. Can AI optimize for relationship outcomes instead? | | **The cold start problem is a self-knowledge problem** | New users can't describe what they want. Can AI learn faster than users can articulate? | **Exploring:** Thesis 1. The others may weave in. --- # Our Tools Know Us Better Than We Know Ourselves ## The Preference Gap Every dating app collects two datasets: 1. **Stated preferences.** The filters you set. Age range, distance, education, height, interests. What you say you want. 2. **Revealed preferences.** The profiles you linger on. The messages you actually send. The matches that become conversations that become dates. What you actually respond to. These datasets diverge. Sometimes dramatically. The user who says "I want someone ambitious and career-driven" spends 3x longer on profiles mentioning hiking and dogs. The user who filters for 25-30 swipes right on 34-year-olds. The user who says "no smokers" matches with someone whose third photo is clearly at a hookah bar. Traditional apps mostly ignore this. They respect the stated preferences because that's what the user asked for. The algorithm serves what you said you wanted, even when your behavior suggests otherwise. AI-native means you can close this gap. The question is whether you should. ## Three Stances **Stance 1: Stated preferences are sacred.** The user set those filters for a reason. Maybe they're aspirational. Maybe they reflect values that haven't shown up in behavior yet. Maybe the revealed preferences are noise and the stated ones are signal. Respect what people say they want. - *Product implication:* AI assists within user-defined constraints. Smarter ranking, but never breaks filters. **Stance 2: Revealed preferences are truth.** People don't know what they want until they see it. The algorithm has seen more of your behavior than you've consciously registered. If it learns you consistently engage with people outside your stated type, it should serve you more of those. - *Product implication:* AI overrides filters (or makes them optional defaults). The algorithm is the authority. **Stance 3: The gap itself is the product opportunity.** Don't hide the divergence. Surface it. "You said X, but you seem drawn to Y. Want to talk about that?" Use the AI not to silently optimize, but to help users understand themselves better. The app becomes a mirror, not just a marketplace. - *Product implication:* AI as self-knowledge tool. Matching is almost secondary to reflection. ## Where It Gets Interesting Stance 3 is the hardest to build and the most defensible if you pull it off. Most dating apps are marketplaces with recommendation engines. They optimize for engagement metrics: swipes, matches, messages. These are proxies for what users actually want (relationships that work), but the feedback loop is broken. The app doesn't know if the match led to a good date, let alone a good relationship. It only knows if you kept swiping. AI-native unlocks a different possibility: an app that helps you understand your own patterns, surfaces the gap between what you say and what you do, and treats that gap as information rather than error. This requires trust. The user has to believe the AI is working *for* them, not *on* them. The difference between "creepy" and "clarifying" is whether the user feels agency in the process. ## The Self-Knowledge Stack The technical stack is the easy part. The hard part is the self-knowledge stack: | Layer | Question | |-------|----------| | **Data collection** | What signals predict compatibility? Swipes are weak. Messages better. Dates best. Relationships gold but rare. | | **Pattern recognition** | Can you identify revealed preferences without explicit feedback? Dwell time, message tone, re-engagement. | | **Surfacing** | How do you show users what you've learned without making them defensive? | | **Agency** | How do you let users correct or override the AI's model of them? | | **Trust** | How do you prove you're optimizing for their outcomes, not your engagement metrics? | The last one is existential. Dating apps have a misaligned incentive: a successful match that leads to a lasting relationship removes two users from the platform. The business model optimizes for churn dressed up as matching. AI-native could break this if you're willing to optimize for outcomes instead of activity. That probably means a different business model (subscription, not ad-supported swipes). It definitely means different metrics. ## Questions I'm Sitting With 1. **How much should an AI infer vs. ask?** There's a spectrum from "we figured out you like X" to "hey, we noticed this pattern, does it resonate?" Where on that spectrum does trust live? 2. **What does "compatibility" even mean?** Traditional matching optimizes for similarity. Some research suggests complementarity matters more for long-term success. Does the AI know which frame to apply? 3. **Can you build a dating app that wants you to leave?** The honest version of "we help you find your person" is an app that celebrates when users delete it. Is that a business? 4. **What's the onboarding for self-knowledge?** If the app's value is helping you understand your patterns, how do you demonstrate that before you have enough data to surface insights? ## The Bet AI-native dating isn't about better matching algorithms. It's about changing the relationship between user and app from "serve me options" to "help me understand what I'm actually looking for." That's a harder product to build. It requires trust, transparency, and a business model that doesn't punish success. But if someone pulls it off, they won't just win the dating market. They'll have built something closer to a self-knowledge tool that happens to help you find a partner. The tools know us better than we know ourselves. The question is whether they'll use that knowledge *for* us or *on* us. --- **See also:** - [[Thesis - Current tools are lossy membranes]] - Interpretation layers as constraints - [[02.10.26 - RingConn and the Case for Dumb Hardware]] - Who owns the interpretation of your data - [[Thesis - Agents are for understanding systems; Personas are for understanding people]]