Finding 04 · Identity and profiles

A 26-minute AI interview gets dates but is too long a gate

Known's voice AI interview averages 26 minutes and the company claims 80% of its introductions led to in-person dates; personality similarity, by contrast, explains very little of how satisfied couples are.

As of September 2026 · Peoplarium is a Digi117 venture in development.

ConfirmedInterpretationResearch wave AChecked 2026-09-28

  • AI

The finding

Deep intake can work. Known, an AI matchmaker, interviews new users by voice. The average session was 26 minutes, and one ran over an hour and a half. In its San Francisco beta the company said 80% of its introductions led to in-person dates. That figure is the company's own claim.

Questionnaires about personality are a weaker basis. A 2023 study of couples found that how similar partners are on personality traits explains very little of their relationship or life satisfaction. Each person's own traits matter far more.

The research read both points the same way: long interviews are a quality ceiling for keen users, not a sign-up step for everyone, and what people do over weeks says more than what they declare on day one.

Why it mattered to the design

Onboarding became a quick card. A new member gets a first suggestion within about 3 minutes.

The interview continues as a drip: at most one contextual question a day, by voice or text, always skippable.

What the system believes is shown in the Mirror. Each card can be confirmed, corrected or deleted. Mutual "meet again" after a meetup is logged as the main signal.

Plan v2 made behavioral learning opt-in rather than on by default, because patterns in behavior can reveal sensitive traits. Personality typing as matching science was dropped.

How it was checked

AI research agents produced the first answer; it was then checked against primary sources. The TechCrunch article on Known was opened and matched the 26-minute average, the 80% claim and the San Francisco beta. The research also quoted a 0–4% range for how much similarity explains. The cited study was opened: it confirms that similarity explains very little but gives no such range, so the figure was left out.

Sources

Ask about this, or request more analysis

Your message goes to Vadym Chernega, founder of Digi117, not a bot. Reply within one business day.

We use your email only to answer. No list, no newsletter. Want this kind of work on your own product? See what we make.