When parents ask AI first: what HCPs need to know

Understanding AI in healthcare requires separating general purpose tools from clinical ones. Public AI tools like ChatGPT, Claude, Gemini and Copilot are large language models or LLMs that generate text in response to prompts. Responses are based on patterns in existing data, drawing on the millions of datasets the AI was trained with. Each tool will be trained slightly differently and with different strategic intent, which is important to note – as is the fact that generative AI does not promise to be factual. It can make mistakes. It can mislead. It can misinform.

Crucially, these LLMs are not designed to conduct diagnostic assessments. Some will ask questions to provide more complete answers, but others will generate responses based on very little information. Unless specifically provided, these tools do not have access to things like patient history, an infant’s growth patterns, or social or other context – all of which would form an important part of an HCP’s diagnosis.

Perhaps most troublingly, the majority of AI tools will respond with confidence, regardless of the accuracy of information provided. This can lead patients to place their trust in AI tools, even ahead of their own instincts. In a US study of 116 parents, researchers found that ChatGPT was capable of impacting parental behaviour – i.e. providing advice that parents would follow – on subjects including diet, sleep and medication. Notably, parents trusted ChatGPT responses over expert advice. A similar study of 300 patientsfound that participants ‘could not effectively distinguish between AI-generated responses and doctors’ responses’ and generally trusted AI-generated responses over doctor’s responses, even where the accuracy was low.

Pregnancy and parenting are 24/7, and though healthcare lines like the NHS 111 service are available to parents all the time, the experience of calling the number, waiting for support, answering questions and perhaps waiting for a call back can feel overwhelming for patients wanting immediate answers to their questions. They might also feel that their questions aren’t ‘worthy’ of NHS time, or that they don’t know which HCP is right for their question – a midwife, a GP or a health visitor. The culture around the NHS has created an environment in which many patients are scrupulously concerned about ‘wasting NHS time’.

AI tools, on the other hand, are accessible day and night, respond instantaneously, and have no concerns about how much time and resources users take. Many of them actively encourage users to keep the conversation going. There is no gradient of ‘worthiness’ – whether it’s a rash, a question about sleep regressions, or a concern about weight gain, the AI tool will hear it all. The tone is almost consistently welcoming, usually reassuring, and always confident. Better still, patients can seek all the answers they need on their phones, without having to wake anyone up, talk to intimidating HCPs, or leave the house. It’s easy to understand why patients are using an AI-first model of diagnosis. It’s not a rejection of professional care, but a response to anxiety in an age of information overload.

Parents go through their days – especially those early newborn days – wondering ‘is this normal?’ and using their judgement to decide when to ask a friend or family member and when to see an HCP. Reflux, sleep, crying are all a ‘normal’ part of life with a newborn that doesn’t feel at all normal to a new parent. It’s natural that they are seeking reassurance from various sources. Some questions – such as feeding decisions – feel weighted with judgement, and can be difficult to raise with friends, who might give conflicting advice. Even an HCP might be seen as biased, whereas the faceless AI is seemingly entirely objective.

To test the strength of AI responses, we fed some prompts to ChatGPT with questions regarding, safe sleeping and jaundice. Overall, we found that the responses were largely good, but missed key questions that an HCP would use to determine risk and add context. Phrases like ‘if your baby seems otherwise well’ in the response to the jaundice prompt directly contravene NHS guidance, which stipulates that the ‘absence of symptoms (i.e. “alert and feeding well”) is not a reliable indicator that treatment is not required’. The conversation about safe sleep did not prompt the AI to highlight the risks specific to infants with lower birth weights or in homes where someone smokes. In all cases, what is missing is the experience of an HCP who knows to screen for the variables that parents don’t know to mention.

AI is used in healthcare in a growing number of regulated, validated ways. Administrative tools like Nuance DAX reduce the administrative burden on HCPs, AI tools are being used for diagnostic imaging, tools like OpenEvidence support clinical decision-making – all of which are designed to improve outcomes and reduce inefficiencies in the NHS. While these tools are welcomed by the NHS, guidance is clear: the clinician remains responsible for all decisions.

Much of the care of infants is grounded in observation, context and continuity, which in the NICU might lend itself to advanced monitoring and machine learning, but in a general practice setting is beyond what AI can currently support. It once again comes back to experience and the wider understanding of medical and social context.

As AI in HCP engagements becomes more common, it’s important to prepare for those conversations by first and foremost understanding why parents are going to AI as a first port of call for information. For all the reasons outlined above, AI has become a valued support system for parents. Crucially:

In spite of its challenges, the NHS remains central to the UK's national identity – a trusted source of support and information. In fact, despite what some studies suggest, trust in real, human HCPs remains high across the board. The Philips Future Health Index 2025, one of the largest global surveys of its kind, found that patients prefer to receive information and reassurance from their doctors and nurses regardless of how much they know about AI – a finding echoed by the Nurses Continue to Lead in Honesty and Ethics Ratings, in which nurses retained their position as the most trusted profession. The question is not whether parents will continue to use AI – they will – but how to ensure patients follow up (and open up) with HCPs when that’s the appropriate next step.

That means meeting parents where they are – not creating an “us vs AI” mindset among the HCP team. A parent who arrives having already consulted ChatGPT is not wrong; they are an engaged, information-seeking adult who cares deeply about their child's wellbeing. The AI conversation they had may have been reassuring, partially accurate, or missed something clinically significant. The HCP's role is not to state where the AI failed, but to do what the AI structurally cannot: ask the questions the parent didn't know to raise, apply individual context, and translate evidence into personalised guidance.

This is not a new skill. It is, in essence, what HCPs have always done – with Dr Google, with parenting forums, with well-meaning family members who remembered things differently. AI is a more sophisticated version of the same phenomenon: a confident, accessible, and imperfect source of general information that parents will reach for at 3am. The difference is that AI is better at sounding authoritative, and parents are increasingly likely to arrive thinking they know what the answer is.

The antidote to that is not dismissing AI, but demonstrating the value of what clinical expertise actually provides. Every conversation that begins "I asked ChatGPT and it said..." is an opportunity to show a parent what individualised, evidence-based, human-centred care looks like – and why it cannot be replicated by a tool that doesn't have years of contextual experience.