Five million Chinese doctors are being competed for by major Internet companies

From building the knowledge system to testing and adjusting the conversation methods, the entire training process took nearly half a year. At first, technicians were overly cautious. For example, if a pregnant woman asks “What should I do if I have a headache now?” the AI ​​clone will continue to ask about age, gestational age, and past medical history for five or six rounds, but still no advice is given. Duan Tao told the other party that patients are willing to cooperate with doctors in the clinic for step-by-step inspections, but they will lack patience online. Therefore, the AI ​​clone cannot copy the offline consultation process. After running in, Duan Tao required that the system must provide preliminary suggestions within three rounds of dialogue, and then add additional questions.

Xu Hongjun, deputy chief physician of the Dermatology Department of Beijing Friendship Hospital affiliated to Capital Medical University, also tried to create his AI clone in Ant Afu earlier. Now, the patient can first describe his medical history and medication usage to his AI clone, and the AI ​​clone will organize the information into a structured medical record and hand it to Xu Hongjun. In the past, he needed to inquire one by one and sort out the information manually, but now he handed it over to AI for preliminary processing.

Huo Jian, head of the Ant Health AI health management and medical services department, said that doctors have limited energy and cannot take care of more patients. AI can provide assistance in pre-consultation, medical history sorting, post-diagnosis follow-up and other aspects, helping doctors broaden their service radius, which has also attracted many famous doctors to join. Currently, more than 2,000 doctors have opened AI clones on the Ant Afu Doctor Edition, including 8 academicians.

Huo Jian said that when polishing the AI ​​avatar, doctors will ask the team for requirements, such as how to make the answers more professional and the operation easier? To this end, the team will regularly polish the AI ​​model with the doctor team to make the answers closer to the doctor’s judgment standards.

Follow-up service is also one of the key functions of Ant Afu Doctor Edition. In the past, many people lost contact with their doctor once they walked out of the office. Doctors can now understand the conditions of chronically ill and elderly patients through AI follow-up. Once a patient has an abnormality, the doctor can receive an AI reminder. In the primary medical care scenario, Ant Afu Doctor Edition has explored another path: helping family doctors improve service efficiency and depth, and providing rural doctors with diagnosis and treatment assistance.

Huo Jian said that in China, a family doctor needs to be responsible for 2,000 to 3,000 residents, and many people are too busy. Ant Afu tries to help family doctors with online contract signing, health education, medication guidance and other tasks. Residents consult basic health issues through AI on a daily basis, and contact their family doctors when necessary. Now, relevant functions have been implemented in more than 40 cities across the country, covering tens of thousands of family doctors.

In the countryside, Ant Group discovered that the trouble of village doctors is that they are not trusted by the villagers. No matter whether they have a serious illness or a minor illness, everyone goes to the county hospital for treatment. In areas such as Pu’an in Guizhou and Meixian in Guangdong, Ant Afu regularly organizes AI skills training for village doctors and also helps them teach villagers to use AI. Now, when encountering complex reports and other situations, village doctors can take photos and upload them, and let AI assist them in making judgments.

Huo Jian said that in addition to direct cooperation with doctors, Ant Group also cooperates with some hospitals to develop intelligent agents, explore pre-consultation functions, and directly connect with the hospital’s information system. Patients can upload their basic personal information before going to the hospital.

Let AI enter clinical decision-making

One day in May 2026, Shao Ming, director of Xiamen Hongai Rehabilitation Hospital, searched for a paper on the Internet and accidentally clicked on the medical AI hydrogen ion link in the search results. He discovered that hydrogen ions can retrieve a large number of original documents, and at the same time complete intelligent disassembly, key points extraction and research context sorting, and also support Chinese and English translation.

Shao Ming was not confident enough about AI from the beginning. In the past two years, he encountered a pitfall when using Doubao to search for medical information: the content output by the AI ​​seemed to be logically complete, and references were listed at the end of the article. However, after checking one by one, he found that some references and discussions did not match. “It’s a matter of human life, and even the references can be wrong, how can I dare to believe it?” Shao Ming said. At first, he also checked every citation of Hydrogen Ion one by one. After many verifications, it was confirmed that the documents retrieved by the tool could be truly retrieved, and the citation content and the output views were basically consistent. Only then did he gradually let go of his worries and become an in-depth user of Hydrogen Ion.

Xu Hongjun was also an early user of hydrogen ions. In early 2026, before this medical AI tool was officially released, Alibaba Health asked him to try it out and make suggestions. He told the Economic Observer that currently, Hydrogen Ion charges doctors an annual VIP service fee of 360 yuan.

Yin Jie, head of Alibaba Health’s hydrogen ion industry ecology and business strategy, told the Economic Observer that before the establishment of the hydrogen ion project, the team conducted in-depth interviews with a large number of doctors, covering grassroots doctors, doctors in tertiary hospitals, clinical scientists and hospital managers, and found that doctors have many scattered needs, such as making PPT, checking instructions, and checking guides. The team did not simply combine these requirements into a set of functions, but set a core goal for hydrogen ions: to provide doctors with more reliable decision-making suggestions in the process of diagnosis, treatment and scientific research. Now, in some top hospitals, the proportion of doctors using hydrogen ions has exceeded 15%. In addition, Hydrogen Ion has a group of “extremely high-frequency users” who ask questions more than 100 times a month.

There are more than 10 billion diagnosis and treatment operations in China every year, and doctors need to make judgments based on knowledge and experience. In the past, when doctors encountered difficult problems, they often consulted instructors and colleagues, or deduced on their own. Now, they are starting to try to hand over some problems to AI.

This also determines the main form of hydrogen ion conversational AI. Yin Jie said that retrieval is not the purpose. The key is whether recommendations based on medical evidence can be given that comply with the principles of evidence-based medicine. To this end, Hydronion regards medical evidence, evidence-based medicine and AI capabilities as three pillars.

In May 2026, Hydrogen Ion announced its cooperation with the BMJ Group British Medical Journal Group. Since then, it has successively cooperated with medical publishing organizations such as NEJM (New England Journal of Medicine) and JAMA (Journal of the American Medical Association). Yin Jie said that cooperating with top medical journals to provide full-text content instead of just abstracts is an important difference between Hydrogen Ion and some competing products. Hydrogen Ion’s answers will go into key details in the full text, such as statistically significant subgroup data.

Cooperation with top medical journals does not come naturally, and both parties need to establish understanding and consensus from scratch. Yin Jie said that before the cooperation, Hydrogen Ion held dozens of meetings with several top journals. They repeatedly asked Hydrogen Ion how it views medicine, what value AI can bring to medicine, and the application boundaries of the product.

After the product is launched, the team regularly conducts research and feedback from doctors. In continuous investigations, Hydrogen Ion found that the focus of doctors’ questions was changing from “check it for me” to “how to deal with this patient” and “what are the latest diagnosis and treatment opinions for this type of patient”. Some doctors even uploaded four to five thousand words of desensitization medical records at one time. In Yin Jie’s view, this change means that doctors are beginning to try to involve AI in more complex clinical thinking.

Duan Tao will use hydrogen ions to search for literature and guides, and sometimes ask his AI clone. Medication during pregnancy alone involves thousands of drugs, and relevant knowledge is constantly being updated. It is unrealistic to store all the knowledge in the human brain.

However, Duan Tao believes that there is still a gap between vertical medical models and real clinical applications. Patients in reality often suffer from multiple diseases at the same time, and the situation is far more complicated than in clinical studies. In addition, although vertical models are less prone to hallucinations, their ability to dismantle complex problems may not be as good as general-purpose large models such as ChatGPT and Claude. The latter’s stronger reasoning and context analysis capabilities can make up for the shortcomings of literature retrieval tools to a certain extent.

From answering questions to completing tasks

Compared with other domestic doctor AI tools, MedWork places more emphasis on task execution. Hu Haoyuan, head of JD Health Intelligent Algorithm, used an analogy to say that some tools provide ingredients and some are responsible for seasoning. MedWork hopes to connect all links and finally deliver a plate of food. Hu Haoyuan gave the example of writing a paper. In the past, doctors had to repeatedly ask AI questions, find research progress, review literature, and then sort out their own ideas and design experiments. MedWork connects these links: first discuss the direction with the doctor and determine the entry point, then outline and design experiments. After obtaining the data, continue statistical analysis, build models, draw charts, and finally form the first draft of the paper. “We are doctors’ assisted driving, not fully autonomous driving. The doctor is always the one who makes the decision, and the doctor can stop at any time,” Hu Haoyuan said.

Hu Haoyuan introduced that before, JD Health had formed a product matrix for ordinary consumers and doctors around health management, serious medical consultation and other scenarios, but it lacked an AI workbench dedicated to serving doctors, and MedWork was designed to make up for this link. Currently, MedWork has been released for more than a month and has been open for access, but it has not yet been promoted on a large scale. It is mainly used by seed users such as partner hospitals and doctors.

On October 21, Shao Ming experienced MedWork for the first time and felt that it was similar to Tencent’s AI office intelligence WorkBuddy, but it was more oriented to medical professional scenarios.

Hu Haoyuan does not shy away from the similarities between MedWork and general AI tools such as Codex and WorkBuddy. He said that what JD Health needs to do is to adapt this type of tool calling and task execution capabilities to medical scenarios, and strengthen the retrieval, citation and verification of medical evidence outside of the model to reduce hallucinations as much as possible. In other fields, AI may require “high emotional intelligence”, but the medical field requires more accuracy. He gave an example. When some general models lack necessary information such as height and weight, they may assume “China’s average height and weight” and continue to answer questions. This is what MedWork wants to avoid.

誰來買單

At present, Alibaba Health, Ant Group, and JD Health have not disclosed the user scale of their doctor-side AI applications, nor have they disclosed specific data to the Economic Observer.

As of October 21, the Android application market showed that Ant Afu Doctor Edition, Hydrogen Ion, and JD Doctor had been installed 19.94 million times, 4.89 million times, and 2.38 million times respectively.

Yin Jie told the Economic Observer that after sorting out the medical industry chain and value chain, the team believes that doctors are one of the groups that AI should serve most. There are about 5 million doctors in China. Compared with Internet products for all consumers, the scale of doctors is not large, but their value is very high. Alibaba Health does not simply measure this business from the perspective of Internet business, but also values ​​the natural match between doctors’ strong productivity attributes and AI as a productivity revolution, as well as its extremely high social value.

Before the establishment of the Hydrogen Ion project, Alibaba Health was thinking about how to better serve doctors. The team had tried to build a medical knowledge graph, but this type of infrastructure was difficult to transform into tools commonly used by doctors. Yin Jie said that the emergence of large language models has allowed the team to see new possibilities.

Ant Group is also planning how to serve the doctor community in the long term. Huo Jian introduced that the group has positioned Ant Afu as one of the most important products of Ant Group in the next ten years, and hopes to serve doctors in different scenarios such as tertiary hospitals, grassroots medical institutions and rural areas in the future.

Xu Hongjun found that although major Internet companies invested a lot of advertising, there were still many doctors around him who did not understand these products.

Duan Tao uses four criteria to measure AI applications: “available, easy to use, willing to use, and inseparable from”. He believes that many AI products for doctors are still stuck in the first two stages.

When talking about the business model of AI products for doctors, many interviewees mentioned OpenEvidence, an American medical AI platform founded in 2022, which is one of the most watched AI medical companies in the world. OpenEvidence specializes in providing evidence-based medical retrieval and clinical decision support for practicing physicians, covering more than 40% of the U.S. physician community, with a valuation of approximately US$12 billion. Its business model is to open it to doctors for free and sell advertising space to pharmaceutical companies.

Duan Tao believes that China’s medical environment and policies are different from those in the United States, and it is difficult to directly copy this model because China does not allow prescription drugs to be advertised. And if you only charge doctors, your income will be limited.

JD Health has internally discussed MedWork’s business model. Hu Haoyuan said that if it only helps doctors search for information with higher quality and easier, there will indeed be challenges in commercialization, because the search engine is free; but if it can help doctors solve complex problems such as processing complex materials, writing papers, and improving the quality of diagnosis and treatment, the situation will be different.

“We don’t shy away from asking doctors to pay. As long as the value provided is strong enough, doctors don’t necessarily have to pay. But from the perspective of inclusiveness, Hydrogen Ion is also evaluating the launch of a free basic version to meet the basic needs of doctors.” Yin Jie revealed that Alibaba Health’s research shows that in China, about 3% of doctors have paid to use some international models to solve daily clinical problems. In some top-level hospitals, this proportion even exceeds 10%.

Another possible business path is to provide services to pharmaceutical companies. Yin Jie said that in the future, medical AI platforms can help pharmaceutical companies communicate the latest drug clinical data to doctors and meet the needs of doctors and companies while complying with domestic compliance requirements. He said that commercial cooperation between pharmaceutical companies should not interfere with the generation of medical AI answers. Hydrogen Ion still organizes answers based on timeliness, authoritativeness, comprehensiveness and relevance. A way to explore is to provide extended reading in addition to the answers, so that doctors can learn about the company’s latest research data on their own.

Related Reading

• 17 years old, annual salary of 7 million2026-10-25
• I went to the Yushu Conference and Wang Xingxing made an explosive speech.2026-10-24
• Is it possible for the entire population to “quit” red fruits?2026-10-24
• Liang Wenfeng reveals the secret to making money2026-10-24
• Musk opens the door to a SpaceX-Tesla merger as shareholders weigh a deal2026-10-21