The AI Dating Glossary

The AI dating industry loves sounding technical. "Semantic matching." "Neural compatibility scoring." "Deep preference capture." These terms appear in press releases and app stores, but rarely get explained.

This glossary defines the real terms — the concepts that actually matter when evaluating what an AI matching system does, and what it doesn't. It's written for people who want to understand the difference between genuine AI matching and apps that slapped "AI" on the same swipe mechanics.

Terms are organized by category. Each one includes a plain-language definition and, where useful, a real example of how it works (or fails) in practice.

Matching Fundamentals

Semantic Matching

A matching approach that compares the meaning behind what two people say, rather than the exact words they use. Instead of matching "I like hiking" with "I like hiking," semantic matching understands that "I trail run on weekends" and "I spend Saturdays in the mountains" are compatible — even if they share no keywords.

In practice: Two users both describe "spending time outdoors" in completely different words get matched because the AI understands both phrases refer to the same underlying preference.

Vector Compatibility

A technical measure of how similar two user profiles are, expressed as the distance between their mathematical representations (vectors) in a high-dimensional space. Closer vectors = more compatible profiles. This underlies most serious AI matching systems.

Profiles are converted into numerical vectors based on hundreds of attributes. The AI finds users whose vectors are close together — not just on one dimension (like age), but across the full multidimensional profile.

Behavioral Matching

Matching based on what people actually do, not what they say they like. Behavioral signals include recurring activities (goes climbing every Wednesday), schedule patterns (available mornings), and usage behavior. Behavioral signals are dramatically more predictive than stated interests.

Sphere's data shows behavioral matching produces 4.3× higher first-message rates than interest-based matching. See: 1,000 Matches Report.

Intent-Based Matching

Matching users based on what they're looking for right now — not a permanent label. Someone might want a running partner this month and a collaborator next month. Intent-based systems adapt to what a user needs at a given moment, rather than locking them into a single category.

Sphere matches across four connection types (dating, friendship, business, activity) simultaneously. A user's active intent determines which pool they're drawn from.

Collaborative Filtering

A matching method that uses patterns across all users to predict who you'll like, based on who similar users liked. The "people who liked X also liked Y" logic from Netflix and Spotify. Used in dating apps for suggesting profiles, but unreliable when applied to human connection because similarity-of-taste ≠ compatibility.

Traditional swipe apps use collaborative filtering to rank profiles. The problem: it optimizes for what keeps users engaging, not what produces real connections.

Cold Start Problem

The challenge of matching a new user who has no history in the system. Without behavioral data, the AI has to rely on stated preferences, which are less predictive. Most apps solve this poorly — they guess using demographic data or popularity. Better systems solve it with deep onboarding that captures behavioral signals upfront.

AI Architecture

Embedding

The process of converting a user's profile, preferences, or message history into a numerical vector that a machine learning model can work with. Embeddings allow an AI to understand meaning, not just keywords. A rich embedding captures nuance — the difference between "looking for something serious" and "open to whatever."

When you answer Sphere's onboarding questions, your answers are embedded into a vector representation. The matching engine compares your embedding to others to find compatible users.

RAG (Retrieval-Augmented Generation)

A technique where an AI retrieves relevant information from a database before generating a response. In dating apps, RAG allows the AI to ground its match explanations and onboarding responses in actual user data, rather than making things up. Critical for match explanation quality.

Deep Preference Capture

The process of building a detailed behavioral and preference model through extended conversation rather than a simple profile form. Instead of asking "What are your hobbies?", a deep preference capture system asks follow-up questions, explores context, and builds a multi-dimensional model of who you are.

Sphere's AI onboarding is a conversational chat, not a form. The depth of the resulting preference model is what allows the matching engine to explain exactly why two people fit.

AI Vibe Check

A pre-match AI interaction where one user's AI agent holds a brief conversation with another user's AI agent to assess compatibility before the humans are introduced. Sphere implements this — your AI talks to their AI before you meet, providing a preliminary compatibility signal.

Two users' AI agents exchange questions about schedules, goals, and communication style. The system surfaces a compatibility summary before either human sees the match.

Semantic Layer

The part of a matching system responsible for understanding meaning and context, as opposed to raw data processing. A good semantic layer is what separates "AI that understands what you said" from "AI that counts keywords."

User Profile & Signals

Behavioral Signal

Any piece of information about what a user actually does — as opposed to what they say they like. Recurring activities, availability patterns, communication timing, and in-app behavior are all behavioral signals. They are consistently more predictive than stated interests in matching research.

Life Stage Matching

Matching based on where someone is in their life — not just age. A 28-year-old building a startup has different social and professional needs than a 28-year-old who just relocated and is rebuilding a social network. Life stage matching uses context to find people at compatible points in their lives.

Sphere's matching distinguishes between "founder building a company" and "professional looking for peer community" — both might be in the same age bracket, but they want different things.

Schedule Alignment

Matching users whose availability actually overlaps — so they can realistically meet. Matching someone who works nights with a morning runner produces a match that will never happen. Schedule alignment uses stated and inferred availability to filter for realistic connection potential.

Sphere's data shows schedule alignment is the third strongest predictor of first-message rate, at 69%.

Connection Intent

What a user is actively looking for right now — not a permanent category. Dating, friendship, a business co-founder, an activity partner. Intent can change over time and should be re-captured periodically. Systems that lock users into a single intent category miss real-world behavior.

Match Quality

Match Quality

A measure of how likely a given match is to result in a meaningful real-world interaction. High match quality does not mean a match looks good on paper — it means the two people actually connect. True match quality is measured by first-message rate, reply rate, and in-person meeting rate.

Sphere defines a quality match as one that results in a first message within 48 hours. By this measure, Sphere's quality rate is 67%. The industry average is ~2%.

Match Explanation

The plain-language description of why two people were matched. A good match explanation names specific, concrete reasons — shared recurring activities, aligned schedules, compatible goals. A bad explanation says "you share similar values" (meaningless). Sphere's data shows rich explanations produce 2.1× higher reply rates than generic ones.

Ghost Rate

The percentage of matches where one or both users send no message. A measure of match quality failure — if matched people don't talk, either the match was wrong or the explanation was too weak to motivate action. High ghost rates indicate a matching system optimizing for something other than real connection.

Traditional swipe apps effectively have a ~98% ghost rate (2% match-to-conversation). Most matches are never spoken to.

Real-World Activation

The point at which a match leads to a real-world interaction — a message, a meeting, a phone call. Real-world activation is the true measure of matching success. Everything that happens on the app before this is preamble.

Match Cadence

How often a user receives matches. Sphere limits match cadence deliberately (4–36/month depending on tier). Higher cadence degrades match quality — users treat a flood of matches differently than a single curated one. Sphere's Basic tier (4/month) shows higher match quality metrics than Elite (up to 36/month).

Design Patterns

Swipeless Design

A product philosophy that removes the swipe gesture — and the mindset it creates — from matching. Swiping trains users to make instant binary judgments based on a photo. Swipeless design forces a different mode of evaluation: reading context, understanding explanation, considering compatibility rather than attraction at a glance.

Sphere delivers one match per week with a detailed explanation. There is no swipe. There is no ranking. You read, you consider, you message — or you don't.

Transparent AI

A design principle where the AI explains its decisions in plain language, rather than presenting results as if by magic. In matching, transparent AI means telling a user exactly why they were matched with someone — not just presenting the match as "a great fit." Transparency reduces anxiety, builds trust, and (per Sphere's data) increases reply rates.

Explainable Matching

A matching system that can articulate, in human-readable terms, the reasons behind every match it generates. The gold standard is top-3 specific reasons: not "you share interests" but "you both go climbing on Tuesdays, you're both early-stage founders, you're both free Thursday evenings."

AI-Washed Swipe App

A dating app that markets "AI matching" but whose core mechanic is still a photo-based swipe. The AI may do some ranking or filtering of the swipe stack, but it doesn't fundamentally change the matching model. Most major apps fall into this category. Genuine AI matching changes the matching mechanic itself — not just the order of profiles shown.

An app that uses ML to rank which profiles appear first in your swipe stack is AI-assisted, not AI-matched. The matching is still done by the human swipe. See: Why Most AI Dating Apps Are Just Tinder With a Filter.

Intentional Design

Product design that makes every interaction feel considered rather than compulsive. The opposite of engagement-optimized design. Intentional design in dating apps means fewer notifications, lower match frequency, richer individual matches, and explicit friction that slows decisions down — because good decisions need time.

Connection Types

Multi-Type Matching

A matching system that handles multiple connection types — dating, friendship, business networking, activity partnerships — within a single platform. Multi-type matching recognizes that what people need from connection changes by context and phase of life, and that the best matching system serves the full range.

Activity Partner

A person matched specifically to do a recurring activity together — tennis, running, climbing, yoga — rather than as a romantic or friendship match. Activity-partner matching is the highest-converting connection type in Sphere's data (79% first-message rate) because the purpose of the match is concrete and low-stakes.

Supermatch

Sphere's term for a top-priority, high-confidence match flagged by the AI as exceptionally strong. Supermatches are delivered at most once per week and represent the AI's highest-signal output — not just compatible, but specifically right for this user at this point in time.

Outcomes & Metrics

Match-to-Conversation Rate

The percentage of matches that result in at least one message being sent by one of the matched users. The most common measure of matching effectiveness. Industry average on swipe apps: ~2%. Sphere: 67%. The gap represents 10 years of engagement-optimized design deprioritizing actual connection.

In-Person Meeting Rate

The percentage of matches that result in the two people actually meeting in real life. The hardest and most meaningful metric in connection apps. Sphere tracks this via an optional check-in feature. Basic-tier users (4 matches/month) show a 31% in-person meeting rate.

Reply Rate

The percentage of first messages that receive a response. A high reply rate indicates that the first message was natural, context-rich, and easy to respond to. Rich match explanations are the strongest driver of reply rate — they give people something specific to write back about.

Swipe Fatigue

The psychological state of exhaustion and disengagement that results from repeated, low-stakes swiping. Characterized by increasing speed of decisions, decreasing engagement with profile content, and declining match quality over time. Not a user problem — a design problem. The product creates the behavior it then blames on users.