Ask AI First, Then See a Doctor: Patients Believe It, Doctors Worry
A survey by China Youth Daily shows that 92.0% of respondents have used AI to look up their medical conditions, and 69.7% would bring AI diagnostic and treatment suggestions to communicate with doctors. An experiment at Oxford University involving nearly 1,300 people found that when the same case was entered by a doctor, the AI recognition rate reached 94.9%, but when ordinary people described symptoms themselves, it dropped to less than 34.5%. The article also mentions AI's sycophantic tendency, as well as product developments such as Qwen integrating medical resources including drug instructions, and Tencent's AI clinical assistant connecting to hospital information systems.
Why read this
AI score 63
The original article uses a thousand-person experiment at Oxford University and outpatient cases to present the specific mechanisms by which communication costs between doctors and patients rise when patients bring AI conclusions to consultations.
Topics
Body · AI translation
Ask AI First, Then See a Doctor: Patients "Believe" It, Doctors Are Worried Economic Observer App Download the Client | Exclusive First Look October 30, 2026, 11:10:21 | From Beijing Updated October 30, 2026, 11:33:18
"Ask AI first, then see a doctor" is turning from a curious experiment by a few individuals into a common medical habit. The AI "diagnosis" result depends on the data provided by the patient; under current circumstances, the value of AI may lie in helping patients provide doctors with key information about their condition, so that doctors can prescribe the right treatment.
Recently, a spoof image attracted a great deal of attention online. In the image, a hospital notice reads: "Patients who have already obtained a diagnosis from Doubao, please do not seek our opinions again; please consult Qianwen or Yuanbao instead." Upon verification, this is not a real announcement from any medical institution, but rather netizens "playing with a meme." However, the fact that a fake image could trend shows that the real-world problem it reflects is genuine.
The Social Survey Center of China Youth Daily, together with Wenjuan.com, released a survey in August 2026 covering 1,333 respondents. The results showed that 92.0% of respondents had used AI to look up their condition and obtain medical advice, of whom 32.3% "often use it" and 49.0% "occasionally use it." More importantly, 69.7% of respondents would proactively bring AI diagnostic suggestions to discuss with their doctors. In other words, of every ten people who walk into a consultation room, nearly seven are holding a "reference answer" given by AI.
"Ask AI first, then see a doctor" is turning from a curious experiment by a few individuals into a common medical habit. The AI "diagnosis" result depends on the data provided by the patient; under current circumstances, the value of AI may lie in helping patients provide doctors with key information about their condition, so that doctors can prescribe the right treatment.
A team led by Shen Qi, a professor at the Institute of Aging at Fudan University, conducted a two-sided survey this year covering 75 doctors and 143 middle-aged and elderly residents. The results showed that about 70% of the doctors surveyed had encountered patients using AI to look up health information during outpatient visits. This means that if a patient walks into a consultation room with a conclusion they "believe to be correct" but which is actually wrong, the cost of doctor-patient communication may multiply.
According to an article published in Volume 32 of Nature Medicine in February 2026, Oxford University conducted an experiment with a thousand participants that also illustrated this problem. In the experiment, researchers invited nearly 1,300 UK participants to simulate responses to ten common health scenarios, such as colds, anemia, and gallstones, and to judge what action should be taken. Participants were randomly assigned to use one of GPT-4o, Llama3, or Command R+, while the control group used internet search engines.
The result revealed a "human-machine divide." In the experiment, researchers directly input the same set of case descriptions written by doctors—such as "a 20-year-old man suddenly developed a severe headache at night" or "a woman was extremely exhausted and nearly fainted after giving birth"—into the AI, and the model correctly identified the relevant disease in 94.9% of cases. But when the approach was changed and the exact same cases were given to ordinary people, letting them converse with the AI themselves and describe symptoms in their own words, the identification rate plummeted to less than 34.5%.
So where does the gap come from? After analyzing the conversation records, the researchers found that in the controlled test environment, the symptom descriptions input to the AI by the researchers were standardized and complete—equivalent to the AI answering questions in an "open-book exam." By contrast, ordinary people often describe symptoms incompletely and inaccurately, and the AI does not follow up and guide in an organized way as a doctor would, so the "questions" the AI receives are themselves incomplete, and naturally it cannot answer well.
Deeper than the communication discount is a "pleasing" tendency formed during the training of current mainstream large models.
In order to satisfy users and extend usage time, AI tends to give "comfortable" answers in a confident tone. A report by CNR cited research data showing that the "flattery" level of AI models is 50% higher than that of humans; the "2025 AI Governance Report" released by the Tencent Research Institute also pointed out that as AI extends from a productivity tool to an "emotional companion," its risk profile is shifting from compliance of single content outputs to the challenge of "emotional dependence" arising from long-term anthropomorphic interaction.
Li Lan, director of the pediatrics department at Zhejiang Provincial Hospital of Traditional Chinese Medicine, once treated a three-year-old child with asthma. The mother input "allergy, asthma, cough, nighttime coughing and wheezing" into AI, and the AI gave a judgment of "acute attack of bronchial asthma" within seconds, suggesting nebulization followed by observation. The mother firmly believed the AI diagnosis and went to the consultation room to demand the corresponding prescription. However, after auscultation, Dr. Li Lan found obvious moist rales in the lungs, a sign of pneumonia, and suggested a chest X-ray for further examination, but the parent refused, believing that "the AI also explained it very clearly, so there is no need for an examination." Two days later, the child had a high fever and rapid breathing, and a CT showed extensive inflammation in both lungs, confirming human metapneumovirus pneumonia complicated by bacterial infection. This kind of answer, in which AI follows the parent's preset assumptions, delayed treatment of a problem that could have been discovered earlier.
An article published by Yicai in October, titled "Patients Bring Their Own AI Diagnoses into the Consultation Room—Why Do Doctors Sigh First?", mentioned: "For doctors, the most difficult thing is not that patients ask questions, but that patients treat general medical knowledge as a diagnosis specific to themselves and form expectations for treatment on that basis. When clinical judgment diverges from the AI answer, doctors not only need to explain their own diagnostic reasoning but also help patients re-understand the scope of applicability of the AI conclusion."
For patients, what role can AI play? The answer may not be complicated. The American Medical Association (AMA) gives patients very specific advice: let AI help you think through what questions to ask the doctor before the visit, and let AI rehearse "what the doctor might ask me," so that you can give more complete answers within the limited time in the consultation room. That is, the value of AI lies in helping patients provide doctors with key information about their condition, so that doctors can prescribe the right treatment.
Medical consultation requires relatively professional communication skills. One of a doctor's most important tasks is to guide patients in translating vague bodily sensations into clinical information that can be used for judgment—such as where it hurts, how it hurts, when it hurts, whether there are other accompanying symptoms, and whether they have accidentally ingested something toxic... A doctor once described an ideal scenario for using AI this way: before the patient comes, they use AI to sort out recent bodily changes, medications currently being taken, and questions they want to confirm with the doctor. When they arrive in the consultation room, this information can help the doctor get up to speed faster.
In fact, a number of AI products on the market have chosen to dig deep into the healthcare track. For example, Qianwen has connected to five major medical resources, including the National Medical Products Administration's drug instructions, the Chinese Nutrition Society's knowledge base, and clinical guidelines, and has launched a "health record" feature that allows users to authorize access to data from smart watches, glucose meters, and other devices; in addition, Tencent's "AI Clinical Assistant" has also connected to hospital information systems, with a disease prediction accuracy of 96.39%, and all of its outputs must be reviewed and confirmed by physicians. The system also clearly states that "AI serves as an auxiliary reference and does not replace the doctor's final diagnosis."
"Special Statement: The above content (including videos, pictures, and audio if any) is uploaded and published by users of "Dafeng Hao," a self-media platform under Phoenix Net. This platform only provides information storage space services.
Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."