When Health Starts in a Chat Window

A new study published in Nature Health analysing over 617,000 health-related conversations with Microsoft Copilot offers one of the clearest signals yet: AI is already embedded in real health behaviour.

It shows when people turn to AI, what they ask, who they ask on behalf of, and how the device they use changes the nature of the interaction. In doing so, it points to a profound shift in the health information journey: AI is part of the first response layer.

The question for pharma media is therefore no longer whether conversational AI will influence patient and HCP journeys. Of course, it already is. The more important question is whether brands, agencies, platforms and publishers understand how to responsibly operate in a world where health intent is increasingly expressed through dialogue rather than search.

From Search Behaviour to Conversational Need

For more than two decades, digital health behaviour has been shaped by search. Conversational AI changes that dynamic.

Unlike search, chat allows users to add context, ask follow-up questions, describe uncertainty and receive responses that feel more tailored to their situation. That does not make AI a replacement for professional care, but it does make it a more emotionally and practically intimate channel than traditional search.

The study found that the largest category of Copilot health conversations was “Health Information and Education”, representing 40.8% of conversations. On the surface, this suggests that general information seeking remains dominant. But the researchers rightly note that many apparently general queries are likely to sit close to personal decision-making. A question about how a medicine works, the side effects of a treatment, or the causes of a condition may be framed generally, while still being deeply personal.

This matters for pharma media because intent is becoming harder to infer from the surface of the query alone. The old distinction between “awareness”, “education” and “action” is becoming less clear. A user may begin with general curiosity and, within a few turns, move into symptom assessment, treatment comparison or emotional reassurance.

Media strategy must therefore evolve from targeting static moments to understanding fluid journeys.

Personal Health Intent Is Larger Than It Looks

One of the most important findings is that nearly one in five conversations involved personal symptom assessment, test result interpretation or condition discussion. This is a striking figure, particularly because the researchers suggest it may be a lower bound. Many conversations labelled as general education may still reflect underlying personal health concerns.

For pharma media, this should challenge how we think about consumer health engagement.

A user asking about a condition is not always simply looking for content. They may be trying to decide whether to seek care. They may be preparing for a doctor’s appointment. They may be trying to understand a diagnosis they have just received. They may be weighing whether a treatment feels manageable, affordable or appropriate.

The implications go beyond content optimisation. They touch audience strategy, journey design, safety, compliance and measurement.

If conversational AI is becoming an early stop in the personal health journey, pharma media has to think more carefully about the role it plays before diagnosis, around diagnosis, and between clinical encounters. The opportunity is not to force brand messages into AI-shaped journeys. It is more to understand where those journeys reveal unmet needs and how media can support patients and HCPs with greater relevance and responsibility.

The Device Is the Context

Perhaps one of the most useful findings for media planners is the sharp difference between mobile and desktop behaviour.

On mobile, “Symptom Questions and Health Concerns” accounted for 15.9% of conversations, compared with just 6.9% on desktop. “Emotional Well-being” was also higher on mobile, at 5.1% versus 3.0% on desktop.

Desktop, by contrast, leaned heavily toward professional and academic use. “Research and Academic Support” represented 16.9% of desktop conversations versus 5.3% on mobile. “Medical Paperwork” represented 15.7% on desktop versus only 2.7% on mobile.

Mobile appears to be the device of immediate personal concern. It is where users ask about symptoms, conditions and emotional well-being, often outside formal healthcare settings. Desktop appears to be the device of structured work: research, documentation, administration and deeper information processing.

For pharma media, this has direct planning implications. Mobile health engagement may require more concise, supportive and empathetic experiences. Desktop may support more detailed education, evidence, resources and professional workflows. The same message cannot simply be resized across devices and expected to perform equally well.

Device should now be treated not just as a delivery variable, but as a proxy for mindset.

The Night-Time Health Journey

The study also shows that personal health queries increase in the evening and nighttime hours, precisely when traditional healthcare access is often most limited.

“Symptom Questions and Health Concerns” rose from 10.6% of conversations in the morning to 13.4% at night. “Emotional Well-being” increased even more sharply, from 3.3% in the morning to 5.2% at night; a rise of more than half.

Health anxiety, symptom concern and emotional vulnerability do not operate neatly within office hours. They often intensify when people are alone, when clinics are closed, when caregivers are unavailable, or when uncertainty has had time to build. Conversational AI is increasingly being used in those moments.

For pharma media, this raises an important question: are our media strategies designed around when people actually need support, or around when we are used to delivering campaigns?

Dayparting has long been a tactical media consideration. In the AI-enabled health journey, it may reemerge as a key strategic one. A nighttime symptom query carries a different emotional weight from a lunchtime educational search. The tone, content, safety routing and next-best action should reflect that reality.

This is where emotional intelligence becomes operational. It is not enough to know that a user is in-market or condition-relevant. We must understand the emotional context in which the interaction is taking place.

AI as a Caregiving Tool

Another significant finding is that one in seven symptom and condition queries are asked on behalf of someone else. For “Symptom Questions and Health Concerns”, 14.5% of conversations were about a dependent. For “Condition Information and Care Questions”, the figure was 14.9%. Even in “Emotional Well-being”, 7.6% concerned someone other than the user.

As is often the case in the reality of care, the person typing is not always the patient. They may be a parent asking about a child, an adult child asking about an ageing parent, or a partner trying to interpret symptoms on behalf of someone they care about. That introduces layers of complexity which are incredibly difficult to decipher, but crucial to try and approach because caregiver journeys are often under-served or treated as secondary. Yet this research suggests that conversational AI may be surfacing caregiver need at scale.

Caregivers need different forms of support. They need clarity, reassurance, practical next steps and guidance on when professional care is required. They may also need content that helps them ask better questions, prepare for appointments or understand treatment pathways on behalf of someone else.

Brands that understand this will move beyond patient-only journey models and recognise the wider network of people influencing health decisions.

The Administrative Burden Is Showing Up in AI

The study also found a meaningful share of queries focused on navigating healthcare systems: finding providers, understanding insurance, booking appointments and completing paperwork.

This is important because it shows that people are not only using AI to understand health. They are using it to navigate healthcare, and that distinction matters.

In principle, finding a doctor, understanding coverage or completing forms should be straightforward. In practice, these tasks create friction, confusion and delay. The fact that users are bringing these needs to AI should be seen as a signal of system complexity.

For pharma media, this connects directly to access, affordability and adherence. A campaign may succeed in generating awareness, but if the next step is unclear, administratively burdensome or financially opaque, impact is lost. Media cannot solve every structural barrier, but it can help reduce confusion, signpost support, and make the path to care easier to follow.

What This Means for Pharma Media

First, intent strategy must become more sophisticated. We know health conversations are not linear, and AI makes them even less so. A general information query may quickly become a personal concern. A symptom question may become a treatment question. A caregiver query may become an access challenge. Media strategies need to reflect this fluidity.

Second, platform planning must account for mindset, not just reach. Mobile and desktop are serving different health behaviours. Mobile is more personal, immediate and emotionally charged. Desktop is more research-led, administrative and professional. Each requires different content depth, tone and design.

Third, timing matters. Evening and nighttime health behaviour should not be treated as leftover inventory. These may be some of the most important windows for supportive, responsible engagement.

Fourth, caregiver needs must be more deliberately built into media planning. If one in seven personal health queries concerns someone else, then care networks are not peripheral to the journey, they are often central to it.

Finally, measurement must evolve. Clicks, impressions and completion rates will not be enough to understand success in AI-influenced health journeys. The more relevant questions will be: did the interaction improve understanding? Did it reduce friction? Did it encourage appropriate care-seeking? Did it support access, affordability or adherence? Did it build trust?

To access the full study click here.