Keeper AI charges up to $50,000 if you get married: well-aligned incentives, but with metrics only the company knows

By Pasquale Pillitteri · November 1, 2026.

Keeper AI is a US matchmaking service that introduces one candidate at a time. Candidates are selected through long questionnaires and psychological tests and ranked by AI models. Human matchmakers, the traditional kind, then review each pairing. The bill arrives only if the couple reaches a milestone such as a wedding. This "marriage bounty" averages about $50,000, according to founder Jake Kozloski in an interview with AlleyWatch.

The company, Keeper Dating Inc., is incorporated in Delaware, operates from New York and was founded in May 2022, according to its FAQ. The service runs in the browser, in English, for people over 18 and, for now, only for heterosexual couples. There are no profiles to browse and no chat until both people say yes. The woman sees the candidate's profile first, and if she is interested, it goes to the man.

According to the sources cited, the process includes more than a hundred open-ended questions, plus tests of IQ, personality type, attachment style and the Big Five traits. The FAQ says the service uses its own models and OpenAI's GPT through the API. The sources disagree on how the work is split between classic algorithms and language models. In December 2025, Kozloski said AI handled 95% of the filtering and that he planned to remove the human step in the third quarter of 2026. The official pages still mention human review.

Pricing is the least transparent part. The FAQ describes a free plan and a "marriage bounty" plan with a fee for each accepted match that falls by 20% with each introduction. Five milestones trigger payment: a civil or religious wedding, a child together, adopting the other person's child, living together for at least a year, and still being together 18 months after the introduction. The company does not publish the amounts. The founder calls $50,000 an average, while the press calls it a cap. A "per date" payment of $5,000, deductible from the final fee, appears in the pitch deck and in Global Dating Insights but not in the FAQ.

The funding figures also need context. Keeper closed a $4 million pre-seed round led by Lightbank and Lakehouse Ventures, completed in October 2024 and announced in December 2025. The "$14 million in commitments" in the pitch deck are promises that depend on someone getting married, not cash in the bank.

The success figures come from Keeper itself. It claims that one in ten first dates ends in marriage, compared with one in a thousand on the most effective apps (one in ten thousand, according to the founder's interview). According to the FAQ, 63% of proposed matches are accepted by both people, and 37% of those lead to a first date. Keeper says it has more than 1.5 million sign-ups, though Global Dating Insights reports about 300,000 completed accounts. There is no public methodology and no total number of dates, so the 10% could be based on a tiny sample.

The main controversy is the attractiveness score. The official pages do not mention any such score in matching. They say the AI "evaluates photos" to check appearance criteria. Dating Industry Insights and a summary of the pitch deck say that looks are in fact judged, by a model trained on "millions" of human ratings. No one has published how that judgment is calculated.

**Our take.** Charging only when there is a result is the most defensible part of the idea. It aligns incentives and forces the company to make the product work. But the same design has a side effect. Free-plan users are raw material that makes the service cheaper for paying customers, and the system has every incentive to push toward the milestones that trigger the fee. A milestone like "living together for a year" is not the same as a happy relationship.

More broadly, AI for relationships is splitting into two paths. One is the virtual companion, where the other party is a model. The other is Keeper's approach, where AI works behind the scenes and the other party is still a real person. In my view, the second is the healthier path: it uses AI's ability to process data at scale without replacing the human bond. That ties into our core thesis: in the long run, AI can help us spend less time searching and more time living what matters to us.

The short-term risks are concrete. A model trained on the average preferences of millions of strangers could reproduce uniform beauty standards, though this is a hypothesis from the author at Dating Industry Insights, not a proven fact. The amount of data requested is also large: according to HyperAI, it can include SAT scores, salary and net worth. That makes it essential for each question to explain how the answer is used, as the FAQ says it does. Language models also tend to agree with their users, so human oversight of such a personal decision matters far more than a footnote would suggest.

For now, Keeper is an interesting bet, not a proven one. Until it publishes its methodology and its total number of dates, its "one in ten" is a marketing claim, not a verified figure.