Consumer AI

The Future of Consumer AI

For twenty years, the industry’s shorthand for the informed patient was Dr. Google. The joke carried an assumption: the patient did the searching, saw a list of sources, and assembled a picture from links. That assumption is gone. In a DHC Group roundtable hosted by Mark Bard, six executives spanning pharma, agency, procurement, technology, and AI transformation worked through what replaces it — and what pharma has to do differently, starting now.

Health doesn’t behave like other AI-mediated categories

Christine Maguire (Fullspan Health) opened by drawing the line that shapes everything after it. Across shopping and general consumption, people turn to AI for convenience and speed, and AI delivers both. Health sits differently. Speed is necessary but nowhere near sufficient — the information has to be trusted, accurate, unbiased, objective, and medically validated. That difference is not a nuance to manage. It is the entire strategic problem.

Which makes the second finding harder. Stephanie Schulman (Publicis Health) pointed to what actually comes back when patients ask. She cited a recent BMJ Open study that stress-tested five major models across ten everyday health questions in five therapeutic areas — 250 responses in total. Roughly half had problems. The models refused to answer twice. And every response, correct or not, arrived in the same calm, confident voice.

Models will improve. But Schulman’s point was that the improvement curve doesn’t change the job description. Marketers can no longer think only about the campaign and the content going into market. They have to think upstream — about the entire information environment these models draw from when they compose an answer.

The patient is already activated. That was always the goal.

Barbara Salami (Decyd) reframed the situation in a way worth sitting with: patients are walking into appointments more prepared than they have ever been. Brands spent years trying to make that happen. It’s happening. The instinct to “fix” it is the wrong instinct.

The real problem is downstream. A patient arrives with a hypothesis about a product, sourced from an answer that may be right or may be confidently wrong. Meanwhile provider time per visit keeps compressing. Salami’s argument: providers are consumers of AI too, and they need help understanding what patients are hearing about a specific brand and how to redirect the conversation constructively. Consumer AI is the headline, but the provider-side gap may be the bigger unaddressed exposure.

Estafania García Zapata (Takeda) described the shift in question sophistication that creates that pressure. Patients aren’t asking what their symptoms mean anymore. They’re asking what the treatments are, which brands exist, whether a particular brand is right for them — and, increasingly, they’re asking after the physician visit, testing what they were told. She views AI as a new audience and asks: Are we preparing content for it? Are we preparing strategy for it?

The traffic has already moved

Dan Haller (Heartbeat) supplied the structural context. Look at the top 100 websites in the world: broadly, they’re losing traffic, and that traffic is being diverted into consumer LLMs. His view is that pretending otherwise isn’t a strategy. The job — right information, right moment — hasn’t changed; the mediation layer has.

And the mediation layer is thin in exactly the places that matter most. In disease states with a shallow content base, AI does the best it can with what exists, which often means confident answers to detailed questions that aren’t actually correct. The role Haller described for pharma is populating that ecosystem with content rich, valuable, trustworthy, and authoritative enough that both the models and the humans benefit.

Should pharma put its name on this? And in what order?

Mark Bard pushed the panel on participation: given that some platforms are experimenting with ad models, how does pharma become part of the experience rather than a bystander to it?

Schulman’s answer: Patients cannot distinguish a general-purpose model from a health-regulated one — both look like a chat box. What differentiates them is entirely underneath: the structure, the governance, the rules about what the system can pull from, what it says when it doesn’t know, and whether it routes the patient toward an appropriate physician conversation. There is real opportunity for trusted brands here, she argued, but only in the right order. The patient-facing interface is the last mile, not the first. Build the context, knowledge base, and internal governance first — otherwise you’ve launched a very competent chatbot that doesn’t know how to regulate itself.

Chris Lee (Pfizer), speaking from the pharma seat, characterized the moment as a transition period. Brands are learning what has to change internally; most aren’t there yet. But he was clear that compliant engagement — including helping train and build products pharma can participate in — is where this goes, and that many companies are actively working toward it. Schulman added a note that should reassure anyone feeling behind: the modular content and content automation capabilities the industry has spent years building are directly translatable to responsible AI strategy. This isn’t a from-scratch build.

Maguire located pharma’s distinct contribution precisely. Manufacturers are the experts in their own therapy — that expertise is the asset. The work is matching it to medically backed, clinically validated grounding. Her clarifying point: AI isn’t creating content or context, it’s retrieving it. So what sits underneath, and what gets cited, determines everything downstream — visibility, brand mentions, citation share.

Salami rejected wait-and-see outright. Marketers were trained to treat content creation and advertising as the measure of the craft, and in doing so ceded every other place conversations about the brand happen — including the places where people complain. Nobody will fully control these models, she said. But brands can find those source conversations and participate in them. If someone raises an issue about your product, address it. That’s part of how the models learn that a company is responsive, and what the reality of the product actually is.

García Zapata raised the urgency stakes highest. By design, once a model believes it has the right answer, it stops exploring. If pharma doesn’t have content ready to be part of that answer, it may not get a second chance at visibility. Her design criteria for AI-legible content: easy to find, easy to identify, easy to summarize — and difficult to distort.

Preparing for the visit — and what that asks of the organization

Mark Bard turned the panel toward the practical question underneath everything else: how can pharma help make the office visit itself successful?

Schulman started with listening — actually hearing what patients are asking, which is what allows a brand to populate the context landscape with the right kind of information in the first place. From there, she argued, brands can become the curators of the personalized doctor discussion guides patients are increasingly walking in with anyway. It challenges teams to be less brand-centric in some ways and considerably more patient- and customer-centric in others.

Lee picked up the discussion guide and pressed on what it has to become. In his view, the days of consumer marketing and HCP marketing operating as separate territories are ending — the two sides need to come together around cohesive engagement strategies that run all the way through to the point of care. And the artifact has to change with them. A doctor discussion guide is no longer something a patient downloads and prints; as he put it, half the people in his generation don’t even own a printer. The question is how something interactive reaches the patient on the device already in their hand — by text, by email, however it arrives — in a form that supports a live conversation with the clinician rather than a document read beforehand.

That’s a cohesion question before it’s a channel question. Both sides of the house share the same end goal, Lee noted, and engaging from those two directions in a coordinated way is what makes the patient journey manageable in the most impactful sense. It asks the organization to work as one where it has historically worked as two.

Haller extended the listening argument into method. Patients don’t receive a list of sources anymore. They receive an answer — and they carry it into the exam room. Understanding how AI is composing those answers reveals where patients are getting it right, where they’re getting it wrong, and what context is missing that a brand could supply.

He drew the parallel to long-tail search query reports, but argued prompt-level analysis is considerably more nuanced — a more direct route into audience mindset than traditional market research alone. In categories as well-trodden as diabetes and as sparse as rare disease, that analysis surfaces content gaps that matter enormously to patients and are served by no one. His advice to clients: care much more about what your audience is looking for and much less about what you’re trying to sell. Closing a gap outside your core value proposition still brings the audience and the brand closer together.

Salami added a lens most brand plans miss. Models optimize for outcomes and for the path of least resistance — so access friction, and the lived experience of staying on a therapy, feed directly into what a model ends up recommending. Brands think top-of-funnel: value proposition, efficacy. But patients are asking how to live with the product once it’s in hand. Shaping how AI surfaces the truth about a brand means looking across the entire spectrum.

Maguire closed on what makes health structurally different from every category being used as an analogy. In travel, the consumer owns the decision. In health, there is no single decision maker — it’s a team sport requiring orchestration across patients, providers, payers, and an increasingly distributed set of access points including telehealth and physician finders. The old model is being disrupted. What made it valuable — grounding, credibility, the right information at the right moment — hasn’t changed at all.