Conversation with Ma Min of Yuanfudao: Hand over the extremely lonely act of reading to AI

Ma Min talks about her seven-year-old son.

One day, the boy's father casually asked, "Who is your favorite character in *The Romance of the Three Kingdoms*?" She had expected the child to respond with his usual question, "Who do you think is great?" Kids at that age were always used to gauging adults' reactions first. But this time, he blurted out, "I like Cao Cao." His father was taken aback: "Why would you like Cao Cao?" Without the slightest hesitation, the child replied, "Yeah, he's very ambitious, and he made it happen." Pausing for a moment, he added, "But I can only give him a score of over 80 points. The dozen or so points deducted are because he really was too insidious and cunning."

In a story where Liu Bei is portrayed as the protagonist and Cao Cao as the villain, a seven-year-old child dared to voice his unconventional preference while still keeping a sense of propriety about it. Ma Min said that at that moment, she realized her son's entire attitude toward learning had transformed—from passive reception to "I will take the initiative to think, and even if my views differ from everyone else's, so what?"

This is exactly what "Yuanfudao AI Advanced Reading" aims to achieve.

In July 2026, Yuanfudao AI Advanced Reading was launched. It is an AI one-on-one reading companion product designed for students in Grades 2 to 6 of primary school. Priced at 1,499 yuan per year, it guides children to intensively read one good book each month, providing personalized follow-up questions and feedback throughout the reading process. It brings the heuristic guidance that was once exclusive to high-priced private tutors to every ordinary family through AI technology, at an average cost of less than 15 yuan per session.

Behind the product, Yuanfudao collaborated with the Research Center for Children's Reading and Children's Book Evaluation of Beijing Normal University to carry out the "Practical Research on AI Technology Empowering Personalized Reading Guidance for Children," using professional teaching and research to consolidate the educational value of technological implementation.

In fact, according to Ma Min, Vice President of Yuanfudao, reading has been the loneliest activity for thousands of years: you read on your own, you either get into the book or you don't; after finishing, you can't express what you've read, and parents can't help—they can't read the book for you. The saying "If you read a book a hundred times, its meaning will reveal itself" ultimately comes down to the fact that you have to read it over and over again by yourself. What she wants to solve is this "real problem" that few people have addressed head-on: what to do when a child can't get into a book, and what to do when they can express nothing after finishing it.

Over the past decade of the internet education boom, Yuanfudao has been regarded as a leader. Now, as the AI wave surges forward, almost everyone is asking the same question: Will traditional education giants be disrupted, and are they truly delivering value to users?

At the 2026 China Internet Conference on July 9, Ma Min gave a straightforward answer to this question: "Ten years ago, Yuanfudao defined online education with mobile internet; today, we hope to redefine education itself with AI."

Therefore, Yuanfudao AI Advanced Reading is the first step Yuanfudao has taken, and in Ma Min's eyes, it is also a greater stepping stone: making parents believe for the first time that AI education products are practical and valuable.

After the speech, 36Kr conducted an in-depth interview with Ma Min. They discussed the "long-term companionship" of AI Advanced Reading, how an education company views AI, and the bottom line she firmly believes in—no matter how powerful a general large language model is, it can never grow a "soul" on its own.

Because behind every choice in education—whether to "correct, guide, or directly give the answer"—lies an educational philosophy, which a machine cannot make on its own. The following is an edited transcript of the interview:

Reading Has Been a Lonely Activity for Thousands of Years

36Kr: In your speech, you talked about "long-term companionship" and mentioned that parents' expectations have changed. When did this change start?

Ma Min: Probably after the pandemic. Internally, we say that in the past, parents only cared about test scores, but now they want both scores, abilities and literacy, and ideally good habits and a broad knowledge base as well.

This reflects that parents have become more rational. Many of today's parents were high-score achievers themselves, but after entering society, they realized that being good at exams and getting high scores cannot solve all problems in work and life. Therefore, they increasingly value the abilities that will benefit their children for a lifetime, and reading is one of them.

36Kr: So, what is the "real problem" you have identified with AI Advanced Reading?

Ma Min: Reading has been a solitary activity for thousands of years. If a child can't get into a book, parents can't do anything about it; if they can't say anything after finishing it, what else can parents do? Maybe they haven't even read that book themselves.

I'm a mother myself. When summer vacation comes, all the mom groups are saying, "Don't let your kid hold the tablet all the time, get them to read a book." But in reality, most children have no interest in books, can't get into them, and find short videos more engaging.

Take *Andersen's Fairy Tales* as an example. My son was reading it. After finishing *Thumbelina*, he followed the AI and said that the story was about Thumbelina being bullied at first and then becoming powerful. At the age of 7, he already sensed that it was a story about growth. Even while playing, he suddenly looked up and said, "I think it's a bit like *The Ugly Duckling*." He was actively thinking and reflecting. This predicament of loneliness, being unable to get into a book, and having no one to turn to for help is the real problem.

36Kr: I understand that many parents outsource reading. They spend all day watching short videos themselves, yet demand their children to read, or simply sign them up for crash courses. If I were such a parent, how would you convince me?

Ma Min: Because our product form is very different.

For this pain point, the industry has tried various solutions. In the early days, there were animated adaptations, such as turning *20,000 Leagues Under the Sea* into cartoons. While this method can spark interest, children often only focus on the plot and still struggle to express their thoughts after watching.

Then came the large live-streaming class model, where famous teachers directly taught the main theme of *Journey to the West* or the rebellious spirit of Sun Wukong. This is essentially knowledge instillation, though it was a relatively optimal solution before the advent of AI, at least providing professional guidance.

We have gone a step further now. Yuanfudao AI Advanced Reading does not accept passive instillation, nor does it ask children to memorize a single correct answer after listening. Instead, it encourages them to speak out and explore the deeper meaning of the book on their own through heuristic questioning.

Which parent wouldn't want to see their child truly immersed in reading, being able to express their views coherently and make more and more sense as they talk? Compared with traditional large live-streaming classes, this sense of gain based on independent thinking is not only deeper but also easier to visualize and verify.

36Kr: So what is missing in reading now may not be methods or content, but "guidance"?

Ma Min: It's one-on-one guidance and critical thinking.

Why Bet on a "Slow" Activity

36Kr: Reading is slow, and results may not come quickly. Why is Yuanfudao making such a heavy investment in this "slow" activity?

Ma Min: It's not about being noble, nor is it deliberately trying to tackle a difficult problem.

First of all, reading is a very "rigid" demand. After the implementation of the new curriculum standards in 2021, schools have clear reading assignments, and required reading lists are even directly included in the exam scope. Secondly, the pain point of reading is relatively hidden. When a child encounters a math problem they can't solve, they will immediately get stuck; but not reading or pretending to read will not cause a family to fall into trouble in the short term. However, this "frog in warm water" type of hidden pain point is precisely the biggest opportunity for reading in the AI era.

It's like before the invention of the car, people were used to riding horses and didn't see it as a pain point; before the popularity of smartphones, people didn't feel inconvenienced without a music player. But once a new species emerges, everyone will exclaim, "It could have been like this all along!"

The same goes for reading. In the past, apart from silently plowing through books by themselves, what better choice did children have?

But now, when a child is reading, an AI tutor can talk to them at any time to inspire their thinking. Isn't this a huge leap in experience?

36Kr: I think there is no lack of demand for reading, but a lack of supply. You have repeatedly mentioned "staying ahead of the supply curve." What does that mean?

Ma Min: Education is a highly specialized field, and it's hard for users to clearly describe "what kind of product I need." It's not like buying a pair of shoes or a camera, where the standards are concrete—you know the camera needs to take clear photos and the shoes need to look good. When it comes to education, parents can only express a vague anxiety like "I don't think the current situation is good," but they don't know how to solve the problem.

Therefore, our requirement for ourselves is to use supply to drive demand and stay ahead of user needs. We must anticipate current educational pain points and children's growth bottlenecks, and take the lead in proposing solutions of "this is how you should learn and this is how you should cultivate" to guide and motivate parents.

Take English education as an example. In the past, parents didn't pay much attention to it because it accounted for a small proportion of exams; now they not only care about grades, but also focus on whether the pronunciation is authentic, whether the child reads original picture books, and whether they have an international vision and multicultural awareness. These have gone beyond exam-oriented needs and are demands at the competence level. Of course, we are not trying to get ahead of the public education system, but closely following the national education layout.

The country's research in this area is very forward-looking, and the "Big Concept Learning Method" we advocate is essentially in line with the core literacy concept put forward by the country and moving in the same direction.

The implementation of core literacy cannot be achieved overnight in all primary and secondary schools. With our advantages in talent and technology, we can scale up the implementation of advanced concepts and take the lead in responding to and assisting the promotion and popularization of the national education policy.

36Kr: Is AI Advanced Reading a short-term product or a long-term product?

Ma Min: It's definitely a very long-term product. On the one hand, the country's emphasis on national reading is a long-term strategy; on the other hand, how to keep children focused on reading and not addicted to pan-entertainment electronic devices is also a long-term pain point for families. As long as these core needs and the vast market exist, "Yuanfudao AI Advanced Reading" will have long-term value that can transcend cycles.

Don't Give the Answer, Let Them Speak Themselves

36Kr: Which hurdle does AI Advanced Reading help children get over first?

Ma Min: The first thing to solve is to make children willing to open the book and keep reading.

First, there is a guide video. We invited the famous host Li Sisi to talk about the background of the book, so that children don't just flip through it carelessly. For example, for Andersen, we will first tell the children that Andersen's fairy tales are often his deep reflections on human nature and the world, and he himself was once ridiculed a lot. With this background, when children read *The Ugly Duckling*, their sense of engagement is completely different. My son even asked after finishing it, "Was Andersen himself like the ugly duckling?" With this sense of engagement, he will also associate it with reality: "Like a classmate in our class, will everyone laugh at him too?" This empathy is instantly aroused.

Secondly, there is animation-assisted support. Some children may not be sensitive to pure text, but to images; some children are not that keen on learning, but have extremely strong empathy. Through multi-dimensional content design, we can accurately capture the interests of different types of children.

There is also a very key setting, which is the role of the "reading companion." For example, when reading *The Little Prince*, the reading companion is the pilot. In our user surveys, this role has a very high satisfaction rate. Because today's children are actually very lonely—they are not short of people around them, but there are few people who are willing to truly walk into their inner world. Adults often only care about test scores, food, clothing, housing and transportation. After my son finished reading *The Little Prince*, he kept saying, "This pilot is so nice, when can I talk to him again?" You see, when reading has an emotional sustenance, children naturally don't want to put it down.

36Kr: Can you demonstrate with a book how AI asks questions step by step?

Ma Min: Take *Journey to the West* as an example. When discussing why Sun Wukong lost to the Buddha, many children's first reaction is that the Buddha is too powerful. At this time, AI will not correct them immediately, but ask heuristic questions: "Think about it, what happened when he fought Nezha and Erlang Shen? Weren't those people powerful? How did Sun Wukong perform then?" Through such questions, guide the child to recall and compare on their own. If the child still doesn't realize it, AI will further prompt: "Go back to that specific page in the book, what's the difference between the description of fighting the Buddha and that of fighting Nezha?" Through such gradual guidance, the child can eventually piece together a complete character image and discover that it was because he underestimated the enemy.

36Kr: If a child wants to talk to AI, they can keep chatting nonstop. How do you constrain it so that it doesn't give the answer quickly or endless answers?

Ma Min: This is indeed the core point. Our principle is that we neither give the answer right away, nor keep engaging the child and center on them like a general large language model.

"Not giving the answer at the beginning, but inspiring" reflects our teaching philosophy. It aims to stimulate the child's autonomy, while "not blindly catering" reflects our professionalism. Because our AI reading companion ultimately has teaching objectives, we can't just chat for the sake of chatting and make children feel happy temporarily.

For example, when reading about Sun Wukong, the ultimate goal is to guide the child to extract the character's qualities through the description of his deeds and master this reading comprehension ability.

We have also specifically studied "shallow-level guidance." On the surface, it seems to be asking questions, but in fact it is just to lead to the pre-set sentence. It's like in daily social situations, some people ask you a question purely to segue into their own point of view. Children can easily see through this insincere guidance, and eventually feel that you are setting a trap for them to ask for the "correct answer." Therefore, AI's questions must be genuine guidance.

36Kr: Which is more important, the amount of knowledge, or the ability to ask questions and inspire?

Ma Min: The transmission of knowledge is relatively easy. Any education company, or even a random person offline, can explain the background of *Journey to the West* clearly, because the country's requirements for required reading lists are very clear. The real barrier lies in the latter—how to inspire, how to guide through several rounds of follow-up questions, and when to bring the off-topic child back to the main point. That's the core value.

36Kr: There is a challenge in designing guidance: how to make reading gradually change from a "task" to a habit that children are willing to do from the bottom of their hearts?

Ma Min: The first step is to ensure that this activity can be effectively implemented. We always have a "reading companion" by the child's side, telling them: "Read a few pages