What Sphere Found After 1,000 AI Matches

After crossing 1,000 AI-generated matches on Sphere, we did something most dating apps never do: we looked at the data honestly. Not the vanity metrics. Not DAU or time-in-app. We looked at which matches led to real conversations, real meetings, and real connections — and which didn't.

What we found challenges almost everything the mainstream dating app industry claims about what makes matching work.

The data set

This analysis covers 1,000+ matches made on Sphere between January and May 2026. We tracked three outcomes: whether the match sent a first message, whether the match received a reply, and whether the pair reported meeting in person (via our optional check-in feature).

We don't track conversations or read messages — Sphere's privacy model explicitly rules that out. What we do track is aggregate signal: did the match result in meaningful contact?

Finding 1: Activity overlap outperforms interest overlap

We expected shared interests to be the strongest predictor of a productive match. "Both love hiking" should produce better outcomes than "both list movies as an interest," right?

Wrong. Shared activities — specific things people do regularly, not just list as interests — were 4.3× more predictive of first-message rate than shared interests.

The difference is precision. Saying you "like hiking" is a marketing statement. Having gone hiking 14 times in the past 6 months is a behavioral signal. Sphere's onboarding captures the difference, and the matching engine weights behavioral signals heavily. The result: when we match two people who both run early mornings, they actually have something concrete to talk about — not a shared checkbox.

Finding 2: The highest-converting match type isn't romantic

Sphere matches across four connection types: dating, friendship, business networking, and activity partners. We expected dating matches to drive the highest engagement — that's the use case with the most emotional stakes.

The data says otherwise. Activity-partner matches have a 79% first-message rate, higher than any other type. Business networking is second at 71%. Dating comes third at 64%. Friendship is fourth at 58%.

This makes intuitive sense once you see it. Activity-partner matching has a clear, low-stakes opener built in: "Want to play tennis Saturday?" There's no ambiguity about intent, no performance pressure, no fear of rejection beyond a simple scheduling question. The match has a natural next action.

Finding 3: Match explanation quality predicts reply rate

Every Sphere match comes with an explanation — the top 3 reasons you were matched. We ran an internal experiment comparing "rich" explanations (specific, behavioral, referenced concrete overlap) vs. "generic" explanations (broad statements like "you share similar values").

Rich explanations produced a 2.1× higher reply rate. When someone reads "You both go climbing on weekday mornings. You're both building companies in the same vertical. You're both free Thursday evenings based on your schedules" — they have something real to respond to.

This finding reshaped how we generate match explanations. Vague explanations aren't just useless — they actively reduce the match quality signal that a person uses to decide whether to engage.

Finding 4: Volume is the enemy of quality

We looked at what happens when users receive more matches per month vs. fewer. Sphere's Basic tier provides 4 matches per month; Elite provides up to 36. You'd expect more matches to produce better absolute outcomes — more shots on goal, more connections.

The data is more nuanced. Basic-tier users (4 matches/month) have a 74% first-message rate and a 31% in-person meeting rate. Elite users have a 61% first-message rate and a 22% in-person meeting rate.

Higher volume degrades the signal. When you have 4 matches, you read each profile carefully. You treat each potential connection as meaningful. At 36 matches, the same cognitive shortcuts that make Tinder feel like a catalogue kick in — and the quality of individual attention collapses.

We're currently rethinking how Elite is structured because of this finding. The goal isn't more matches. The goal is more connections.

Finding 5: Morning users connect differently

This one surprised us. Users who completed Sphere onboarding between 6am and 9am local time had a 19% higher in-person meeting rate than users who onboarded in the evening. Same matching quality. Same match explanations. Different outcome.

We don't have a clean causal explanation. One hypothesis: morning onboarding signals a higher baseline of intentionality — someone who fills out a thoughtful profile before 9am is probably taking the matching seriously. Another: morning users tend to describe outdoor and active routines, which produce higher-converting activity-partner matches.

We're still looking at this. It might be noise. But it's been stable across 4 months of data now.

What this means for AI matching

These findings confirm something we believed when building Sphere, but it's different to see it in data: the job of an AI matching system is not to optimize for engagement. It's to reduce the gap between a match and a real-world connection.

Mainstream apps optimize for time-in-app. Their matching exists to keep you swiping, not to get you off the app. Every design decision — infinite scroll, push notifications for new likes, boosted profiles — is in service of that metric.

Sphere's matching is explicitly optimized for the opposite outcome: get both people off the app and into a real interaction as quickly as possible. The data suggests this alignment is starting to produce results.

We'll publish a more detailed breakdown of the full 1,000-match dataset later this year. In the meantime, if you're building in this space — or writing about AI dating — we're happy to talk.