A new Chinese dating app is betting that AI can do more than keep users chatting — it wagers it can see them married in three years or less.

If it fails, the team behind Liangpei — roughly “good match” in Chinese — promises to refund the user’s one-time fee of 2,000 yuan ($275). The guarantee comes as marriage rates fall in China, with only 6.76 million couples tying the knot in 2025, about half the number a decade earlier.

Launched this past July, Liangpei begins with a roughly 20-minute voice conversation with an AI matchmaker. Users answer 25 questions about their relationship history, marriage plans, and even whether they intend to buy property. The system then uses their answers to build a detailed profile, calculate compatibility scores, and recommend potential partners.

Once two people match, they can talk only until one decides to move on, a restriction the app’s developers say is intended to keep them focused on one relationship rather than constantly searching for alternatives.

“We want to get the users off the platform as soon as possible,” founder Zeng Xinxun told Sixth Tone.

Before launching Liangpei, the 32-year-old worked on AI search technology at Kimi, one of China’s best-known generative AI platforms. His 18-person team now uses similar technology to draw out information that basic dating-app tags such as age, occupation, and location cannot capture.

Approximately 91% of users who start Liangpei’s interview finish it, Zeng said, going on to create profiles averaging 516 Chinese characters, about four times as long as those on several major Chinese dating apps.

About one in 10 users has reached an exclusive one-on-one match, according to Zeng. Those conversations last more than 10 days on average and run to more than 200 messages.

“One-on-one allows both parties to invest more focus and effort into building a relationship,” Zeng said. “Even if it doesn’t work out, it’s a conclusion reached after genuine communication.”

So far, he said, Liangpei has more than 9,300 active users across 300 cities. Men account for 57% of users, and women 43%. The two largest age groups are 26 to 28 and 29 to 31.

China’s established dating and social platforms operate at a much larger scale. Tantan, one of the country’s oldest dating apps, had 10.7 million monthly active users in March 2025, while Soul draws large audiences with broader offerings that combine dating, social discovery, entertainment, and shared interests.

Liangpei also requires real-name registration, a marital status check, and live photo verification. The company has blocked 487 people found to be married, Zeng said. It also uses AI to watch for signs of fraud once conversations begin.

Wang, a Shanghai-based user who asked to be identified only by her surname for privacy reasons, has tried more than 10 dating apps. Liangpei was the first on which she was willing to answer so many questions.

“It has helped me understand myself better and clarified exactly what I’m looking for,” she said.

Within a week, Wang entered an exclusive one-on-one match. She said the format made online dating less exhausting and allowed her to focus on getting to know her counterpart. But with no other matches to turn to, the arrangement depended on that person keeping the conversation alive.

“Once matched, your slot is locked,” Wang said. “If the other person isn’t proactive, you’re left waiting with nowhere for the conversation to go.”

On Xiaohongshu, the lifestyle app also known as RedNote, Liangpei has drawn sharply mixed reactions. Some users described it as a welcome option for people anxious about finding a spouse, while others questioned whether treating marriage as a fixed target made dating too rigid.

Free users can send only one “like” a day, according to posts reviewed by Sixth Tone. The app also prominently displays compatibility scores, which some users said made them more reluctant to start conversations.

Despite the compatibility score, several users admitted they still ended up judging potential partners by appearance.

Three-year itch

Six years ago, Zeng said he became frustrated trying to meet someone online — a period that lasted a year. He wrote a lengthy profile, but few people seemed to read it, and even fewer replied.

“Profiles were too bare to spark interest,” he said. Zeng eventually expanded his own to around 2,000 Chinese characters, then came across a woman whose profile was just as detailed.

“We quickly went from strangers to a close connection,” Zeng said. By the time they met in person, he thought the relationship might go somewhere.

They are now married. And Liangpei uses that same outcome as its measure of success.

His team had considered a first date or the start of a relationship instead. Marriage, Zeng said, was different: “It’s legally verifiable and undisputed.”

The three-year cutoff came after his team interviewed 150 couples obtaining marriage licenses at a civil affairs bureau in the southern city of Shenzhen. More than 70% said they had married within three years of meeting, Zeng said.

Liangpei has been operating for only a few months, so none of its users have yet to reach that deadline. For now, the company is reviewing the data it does have, including whether higher compatibility scores are associated with more likes, replies, and longer conversations.

“For example, among 1,500 users in Shenzhen, the AI can directly rank the top 10 people most likely to succeed with you,” Zeng said. “You don’t have to browse everyone — just start with the highest scores.”

To address relationship anxiety, Liangpei offers an “icebreaker” feature that suggests questions when a conversation stalls. An “AI wingman” can also analyze an exchange and offer advice.

But the “AI wingman” will not compose replies. Zeng said someone who struggles with online banter might still make a good partner in person, but AI should only carry early exchanges far enough for both people to find out.

Du Shichao, an assistant professor of sociology at Shanghai’s Fudan University, said Liangpei’s focus on marriage could appeal to people who already know what they are looking for. He also saw value in matching people using more than surface-level information.

“It may not address everyone’s pain points, but it can at least solve problems for some people and offer them more choice,” Du said.

But he was more cautious about AI taking on a larger role in relationships.

“This ‘enhancement’ can easily slide into ‘replacement,’” Du said. “When this kind of use is taken to an extreme, people need to be wary of ceding their capacity for emotional connection to technology.”

Whatever help an AI assistant can provide online, he said, people still have to learn about each other in person. “Ultimately, people must rediscover each other and themselves through real-world interaction.”

Du said technology may help with some of the difficulties individuals face in meeting and communicating with partners, but it cannot address the forces shaping marriage more broadly, including demographics, economic pressure, and family expectations.

“It may benefit some, but won’t significantly change China’s marriage patterns,” he said.

Zeng is making a more modest claim. “We don’t need to serve everyone,” he said. “We can’t help those with particularly poor prospects, or those with unrealistic expectations. If the people in between can find success in Liangpei, that’s good enough.”

Whether they do will become clear in 2029, when Liangpei’s first users reach the end of its three-year guarantee.

Additional reporting: Xu Ruishuo; editor: Apurva.

(Header image: Visuals from VCG, reedited by Sixth Tone)