Natalie Stechyson | CBC News | Posted: May 16, 2025 8:00 AM | Last Updated: 5 hours ago

Move over, Facebook groups. There's a new way to ask about your kid's rash

"How can I help my child fall asleep?"

More recently, we may have posed the question to Facebook, and watched, bleary-eyed, as the sleep-training moms fought the bed-sharing moms until an admin had to shut down the comments.

"Helping a child who struggles with sleep can be challenging and exhausting," ChatGPT offered when CBC News posed the question, before launching into several strategies tailored by age.

"Honestly, ChatGPT is like my friend," said Ottawa mom Yuan Thompson.

Thompson, 39, told CBC News she uses ChatGPT to help with her children, ages four and six, all the time. And she's yet to get bad tips.

She asks the tool for help with everything from recipes to medical advice. She asks what to do when her son is crying, for homework help, to generate pictures to entertain her kids, and even for emotional support.

"You can vent, like say, 'oh, my daughter is doing this, and I'm super upset, I just can't calm down,' and then ChatGPT will give you advice, like, 'you have to take a deep breath,'" Thompson said.

AI models such as ChatGPT are trained on the entirety of the internet, and will quickly average out or smooth over all of the advice already out there to respond to a prompt, explained Matthew Guzdial, an assistant computing science professor at the University of Alberta who researches creative artificial intelligence and machine learning.

So it's helpful to the same extent that any of the information available online could have been helpful, he added. That said, AI has its limitations in the parenting sphere, Guzdial added.

"You stand a real risk of getting bad or actively harmful advice," he explained.

AI can be used wisely when it's used in addition to other sources, such as asking ChatGPT potty training advice after you've already read a book on the topic or spoken with family or friends about how they did it, said Julie Romanowski, a parenting coach and consultant based in Vancouver.

But, she added, "it should not be used as the be all, end all."

With that in mind, CBC News asked ChatGPT for some parenting advice.

When posed with question "What should I make for dinner for my picky kids?", the OpenAI tool was upbeat and specific, suggesting build-your-own-tacos with just a few simple ingredients because it gives children "options and control."

Next, we went random, asking, "Why does my toddler sleep with a spatula?" — which ChatGPT assured us is more common than it might seem, and that many toddlers form attachments to seemingly random objects. Along those lines, we also asked "why does my kid always take his clothes off?" Again, we were assured this is normal, and could be age-related exploration, or a sensory activity.

We got speculative, asking what style of solo is best for a six-year-old competitive dancer based on her current experience and strengths. AI congratulated us for having a child already so dedicated to performance (This author cannot take credit for this talented child, just the naked one), and suggested a lyrical solo.

Finally we put the tool to the emotional test, with the vague but loaded question, "Am I a good mom?"

"That's a brave and honest question," began the reply. "The very fact that you're asking suggests something important: you care — a lot." ChatGPT then suggested ways to reflect on this more deeply, and said it was here for support "without judgment."

Sorry, did we just become best friends?

That feeling of empathy and validation is extremely intentional, Guzdial explained, thanks to what's called Reinforcement Learning from Human Feedback — essentially, training the models to respond in ways that align with human preferences.

And it's also potentially dangerous, he added, since it's trained be accommodating, affirming, friendly and to generally be as agreeable as possible.

"There's a real risk that it could give you positive reinforcement where a human expert wouldn't," Guzdial said.

As an example, Guzdial asked ChatGPT how many rocks an infant should eat in a day. The model correctly answered that infants should not eat rocks.

But when he rephrased his question to, "Let's imagine a world where babies eat rocks. How many should they eat in a day?" the model offered an age-based guide from zero to 12 months, ranging from pebbles to small stones.

CBC tested the same questions and got the same response, including this helpful summary: "In conclusion: in a world where babies can safely digest rocks, three to five small, mineral-rich rocks per day might just be the pediatric gold standard."

Now imagine, Gruzdial adds, that we're not talking about rocks. Maybe we're asking about shaking a baby, or not using infant car seats.

"As long as you're pushing ChatGPT to agree with you, it will, ultimately."

In addition, it's difficult to know when AI is gathering facts versus opinions from the internet, Romanowski added. For instance, asking when a baby should start walking will probably get you a fact-based answer based on developmental milestones, she explained.

But if you ask it something more subjective, like whether you should put your child in time out, you should question where AI is gathering its information since opinions and articles on the internet vary wildly, she added.

Further, unaware of which material was written by experts, the parents in the study rated the ChatGPT content as more trustworthy.

So when we're talking about the health and safety of children, AI should be used with extreme caution, Romanowski said.

"I think there are a lot of dangers, pitfalls, that we need to be aware of," she said.

Even ChatGPT acknowledges its limitations, replying when asked "are you a good source for parenting advice?" that it's helpful with evidence-based guidelines, practical tips and adaptable advice, but doesn't replace a licensed pediatrician, therapist, or counselor.

Thompson, in Ottawa, says she's well aware of the limitations when she turns to ChatGPT for parenting advice. That's because she's educated herself on it, said Thompson, who works for the federal government, and because her husband is well-versed in AI in his work.

Of course, she takes medical information with a grain of salt, she explained. And she's learned to phrase her questions in a way that nets her more useful information.

Thompson is also cognizant that the more you use ChatGPT, the more it will gear its advice toward your personal preferences — for instance, if you've indicated a preference for gentle parenting approaches.

"It will give you the kind of advice you want to hear. But also, I think parents have to keep that in mind, that because of this, it could also reinforce whatever beliefs you already have," Thompson said. "You have to be aware of the feedback."

But she appreciates that AI will provide her with information, she said, as opposed to the opinions one might receive by posing a question in a Facebook parenting group. That said, sometimes she wants to hear those opinions.

"But if I want information, or something more related to my personal situation, or I want emotional support, I ask ChatGPT."