Recently, a story about new parents blindly following feeding advice from the AI assistant “Doubao” (also known as Coze) made headlines—their one-month-old baby’s weight didn’t budge for a whole week. This sparked a heated debate about how far we should trust AI tools when it comes to parenting and health advice.
On May 28, Doubao’s official team released a detailed statement addressing the reports, calling them “inaccurate” and urging everyone to take AI-generated content with a grain of salt. According to Doubao, they reached out to the hospital and the doctor mentioned in the reports right after the news broke. The doctor said that a family had brought in their baby who had jaundice, and during the consultation, they casually mentioned that “Doubao suggested feeding 60 ml each time.”
But here’s the thing: the family never showed the full conversation log with Doubao, nor did they clarify whether Doubao had also provided a recommended daily total or any other context. The attending doctor also admitted that he had no idea what exactly the family had asked the AI or how the whole interaction played out. Doubao believes that without the full picture, it’s possible the parents misunderstood or cherry-picked part of the AI’s response.
Doubao further explained that they’ve run multiple tests on the exact feeding advice mentioned in the case. Under normal circumstances, Doubao would never just say “feed a one-month-old only 60 ml per meal.” Instead, when answering such questions, it typically offers a reference range for the daily total milk intake and clearly reminds parents to watch the baby’s reactions—like if the baby keeps crying, they should increase the amount or consult a doctor. Other major AI models give similar standard advice for the same kind of questions. Doubao also said they hope the family will get in touch so they can look into the issue and improve their service.
As reported earlier by Nanguo Morning Post and other outlets, the couple’s baby had been hospitalized for jaundice after birth, so they were extra careful. When the baby turned one month old, he kept crying a lot. Lacking parenting experience, they didn’t turn to a real doctor or a trusted parenting book—they asked Doubao instead. Following what they thought was the AI’s advice, they strictly limited each feeding to 60 ml.
A week later, they took the baby back to Nanning Maternal and Child Health Hospital for a jaundice check. During a routine chat, the neonatologist was shocked to learn that the baby was only getting 60 ml per feeding. When pressed, the parents finally spilled the beans. The doctor pointed out that a healthy full-term one-month-old normally needs 80 to 100 ml per feeding. That 60 ml was way too low to support proper growth. Because of this long-term underfeeding, the baby’s weight hadn’t budged in a week, and he was constantly crying from hunger.
Actually, there are clear official guidelines for infant feeding. The National Health Commission’s Infant Nutrition and Feeding Assessment Service Guide (Trial) stresses that babies under three months old should be fed on demand. If a baby shows hunger cues, you respond—no rigidly sticking to a fixed amount per meal. It’s okay if the amount varies each time, as long as the daily total is adequate. The Chinese Nutrition Society’s Infant Feeding Guide for China (2022) also says that for babies 0 to 6 months old, a single feeding can range from 60 to 120 ml, and the exact amount should be adjusted based on the baby’s weight, appetite, and growth. There’s no one-size-fits-all number.
Now that generative AI is everywhere, more people are turning to large language models for health, medical, and parenting info. But these models have a built-in flaw—hallucinations—and in high-stakes areas like this, that can be really dangerous. According to the latest research from the top AI conference AAAI, even the best general-purpose models still hallucinate about 29.1% of the time on complex clinical tasks, and some open-source models go as high as 57%. A study in BMJ Open also found that half of the responses from five major AI chatbots to evidence-based health questions had serious flaws, with some models hitting a 58% error rate.
Insiders point out that many general AI models do add a disclaimer at the end—like “for reference only, please consult a professional”—but it’s usually at the very bottom, in regular font, and easy to miss. And since people asking these questions often lack the expertise to judge the accuracy of the answers, they tend to follow blindly.
Here’s the full statement from Doubao:
