Sphere Match Quality Report

Sphere publishes this report quarterly. It exists because no one else does this. Dating and connection apps collect enormous amounts of behavioral data — and they publish almost none of it. The metrics they do share ("500 million matches," "X% of couples met on our app") are marketing, not measurement.

This report is measurement. It covers the first 1,000+ AI-generated matches on Sphere, tracked from creation through real-world outcome. The methodology is described in full so it can be replicated, criticized, and improved. Other companies in the space are welcome to publish equivalent data — we'd welcome the comparison.

1. Methodology

This section describes exactly how we defined, collected, and analyzed the data. Transparency here is not optional — match quality claims without methodology are marketing.

What we measured

We tracked three outcome events for each match:

Event 1 — First message sent: Did either matched user send a message within 48 hours? We chose 48 hours because this is when signal is clearest — a message sent 3 weeks after a match reflects different dynamics than one sent the same day.

Event 2 — Reply received: Did the first message receive a response? This is the clearest signal of genuine mutual interest — someone opened the message, read it, and chose to engage.

Event 3 — In-person meeting: Did the pair report meeting in person? Sphere has an optional post-match check-in feature. Reporting is voluntary — this metric understates actual meeting rates. We note this in the interpretation.

What we did not measure: Content of conversations (Sphere does not read messages — privacy model explicitly excludes this). Match satisfaction ratings. Relationship outcomes beyond initial contact. These are on the roadmap for future reports.

Comparison methodology

The ~2% industry match-to-conversation rate cited throughout this report is derived from academic research and investigative journalism on major swipe apps, including analysis published in peer-reviewed journals and third-party user behavior studies. We acknowledge this is an approximation — the major swipe apps do not publish this data. We have cited the most conservative available estimate.

2. Primary findings

Finding A: Match-to-conversation rate

67% of Sphere matches resulted in a first message within 48 hours. This is the headline number. It reflects whether a match prompted action — the most basic measure of whether the match was real rather than performative.

Note: Swipe app figures are estimates based on available third-party research. Major apps do not publish match-to-conversation data.

Finding B: Signal type and conversion

We analyzed which input signals, when shared between matched users, most strongly predicted a first message. We categorized signals into five types. Results:

Behavioral signals outperform stated interests by 4.3×. This finding has direct implications for how AI matching systems should be designed: optimizing on what users say they like (the input most apps capture) is dramatically less effective than capturing what they actually do.

Finding C: Match quality by connection type

Sphere matches across four connection types. Contrary to our initial hypothesis, romantic matching is not the highest-converting type.

Activity-partner matching performs highest across all metrics. The most likely explanation: activity-partner matches have a built-in, concrete first-message. "Want to play tennis on Saturday?" requires no social risk assessment. Romantic matches carry more ambiguity about intent — even with an AI explanation, the emotional stakes of the first message are higher.

This finding suggests that AI matching systems built primarily for romantic use cases may be leaving significant match quality on the table. The social infrastructure of human connection is much broader than dating.

Finding D: Match explanation quality and reply rate

We ran an internal experiment comparing two explanation formats: "rich" (specific, behavioral, referenced concrete overlap between the two profiles) and "generic" (broad statements like "you share similar values" or "you have compatible goals").

Rich explanation format example: "You both go climbing on Tuesday mornings. You're both building B2B startups in fintech. You're both available Thursday evenings based on your schedules."

Generic explanation format example: "You have compatible interests and complementary professional goals."

The specific version gives both people something to respond to. It reduces the activation energy for a first message — you don't have to generate a conversation starter, because the match explanation already contains one.

Finding E: Match cadence and quality degradation

We compared outcome metrics across tier segments (Basic: 4 matches/month vs Elite: up to 36 matches/month).

Higher match volume correlates with lower quality outcomes per match. We attribute this to cognitive load and attention degradation: when you receive 4 matches per month, you engage deeply with each one. At 36, the same psychological shortcuts that make swipe apps feel like catalogues begin to appear.

This finding is actively informing how we redesign the Elite tier — the goal is not to give more matches, but to give more meaningful ones.

3. Limitations

This section exists because the data has real limitations and they should be stated directly.

Sample size: 1,000+ matches is a meaningful starting dataset. It is not large enough for statistical confidence in subgroup analyses (e.g., geographic or demographic breakdowns). We have not broken out results by city or age group for this reason.

Self-selection: Sphere's current user base is early-adopter, tech-adjacent, and predominantly in a handful of European and US cities. Results may not generalize to broader populations.

In-person meeting undercount: The check-in feature is optional. We estimate actual in-person meeting rate is 15–25% higher than the reported figures.

Comparison benchmarks: The ~2% swipe-app figure is an approximation. Major apps do not publish this data, which is itself a meaningful observation. We have cited the most conservative available estimate.

Time horizon: Five months of data captures initial outcomes. We cannot yet report on long-term connection quality, sustained relationships, or multi-month engagement. This is on the roadmap.

4. What this means

Three conclusions that hold even with the limitations above:

Behavioral signals are substantially more predictive than stated interests. This is the most actionable finding for anyone building in the space. Apps that rely primarily on user-stated preferences (interests, hobbies, values) as inputs to matching are starting from worse data than apps that capture behavior.

Match explanation quality is a product decision, not a nice-to-have. The 2.1× difference between rich and generic explanations is not marginal. If you're building an AI matching system and your explanation is "you seem compatible," you're losing half your conversion.

Volume and quality trade off in predictable ways. The industry assumption that more matches = more connections is not supported by this data. Attention is finite. Scarcity of matches increases attention per match, which increases engagement quality.

5. Next edition

The Q4 2026 report (targeting October 2026) will include:

— 3,000+ match dataset with geographic and demographic breakdowns
— 90-day and 6-month outcome tracking (sustained connection, not just first message)
— Match explanation A/B test results at scale
— First data on AI vibe check impact on reply rates

To be notified when the Q4 report publishes, join the waitlist or follow us on Telegram.