“On the matchmaking app, if we ask you a question and your tonality changes in the response, it cues to us that you may not be telling us the full truth. And so we’ll ask you that same question in two or three different ways throughout your experience,” Cohen-Aslatei says. “We built this to mimic what a matchmaker would do for a client. The LLM is tracking pitch and tone change in your voice because we want to make sure that we have an accurate understanding of who you are and what you’re looking for.”

After answering dozens more questions about lifestyle, future goals, boundaries, family, attraction, hobbies, and more over the course of a few days, Tai told me it’d take the information provided and get back to me. Two days later, I received my first two potential matches.

I Love You, Alive Girl

As a 31-year-old woman, I put my ideal age range at a healthy 26 to 40 years old. My first two matches were 23 and 47. One was not alive when 9/11 happened, and the other had already graduated from college at that time. Off to a rocky start.

When a potential match is found, the person’s picture is blurred, and Tai gives you a synopsis of what makes you a potential good match. (You need to provide selfie verification to confirm identity, and no one unverified will ever be matched.) After that, you can click to see a bit more about them, like profession, age, income, and a short bio that the AI creates.

At this stage of AI adoption, there is still a strong statistical bias toward, let’s say, men who wear wraparound sunglasses and think driving a Cybertruck is a sign of virility. Nearly every one of the 16 matches I received during testing was Christian and wanting children ASAP, which Tai flagged each time as a potential issue. Many were also flagged initially by Tai because they only wanted to date a certain race or valued traditional gender roles, both of which I made clear that I wasn’t aligned with.

Out of journalistic duty, I accepted every match I received; even a MMA-loving body builder that enjoys grilling meat (I’m vegan) and going to the gun range (I’m generally anti-gun). Matches ranged from Portland, Oregon, to DC, to New York City (where I live, although most matches were outside NYC). Overall, not a single person I was matched with would be someone I’d swipe right on if I saw them on a traditional dating app.

If you accept, you’ll either need to wait for the other person to accept or pass on the match, or they will have already accepted, and you can begin chatting. Here, your AI dating coach steps in to play wingman, providing prompts based on the other person’s profile, highlighting similarities you have, and giving conversation questions based on answers from the match’s profile. Not only does the coach provide potential ice breakers (and responses), you can also chat and ask for pointers.

I asked it to give me tips on how to break the ice with new matches, and it gave me advice, with each point having an explanatory paragraph below. Advice included giving compliments, asking open-ended questions, using humor, referencing current events, sharing about yourself, and mentioning mutual interests. The advice was basic but solid, and mirrored what the coach was doing with the provided conversation prompts.

This is all a great idea in theory, and could be very helpful with people who have a tough time communicating with strangers. But it could also lead to a bigger problem. You don’t really know who you’ve been talking to if AI has been doing all of the chatting for you. And if you meet in person, you don’t know much about your date’s actual personality. You can tell so much from how people type, what questions they ask, and their sense of humor. That was all missing here.