More than 5% of all ChatGPT messages globally are healthcare related, according to a January report from OpenAI. Anthropic cited healthcare and wellness as the largest guidance domain in an April study.

According to Bain’s June 2026 Frontline of Healthcare survey, nearly one-third of US consumers report using AI for healthcare questions. Many consumers research conditions and treatment options before they even schedule an appointment, then bring what they found, accurate or not, into the exam room, expecting to help shape their care.

Doctors are increasingly turning to AI, too. In our December 2024 Next-Gen Pharma Customer survey, US doctors reported using AI tools, but just 3% ranked AI among their top three sources for informing prescribing decisions. As of June 2026, 20% do. And more than one-third say AI-generated recommendations are important to their treatment decisions.

Doctors who ranked AI among the top three sources
for prescribing decisions

One-third of doctors also report that their organization uses AI-enabled clinical decision support (AI-CDS) tools, including UpToDate and OpenEvidence. More than half of AI-CDS users tap into these resources multiple times a day. Roughly two-thirds say AI-CDS has improved their diagnostic and treatment decisions, and nearly three-quarters say it has improved medication safety and interaction alerts (see Figure 1).   

Figure 1

Adoption goes beyond purpose-built clinical tools. Nearly half of doctors use general-purpose large language models (LLMs), such as ChatGPT or Claude, in clinical practice, making them the third most widely adopted AI tool in our latest survey, behind UpToDate and OpenEvidence.

Pharma has spent decades competing to influence the conversation in the exam room. The next frontier is how to shape the information patients and doctors consume before that conversation even begins. AI is now both the commercial channel and the marketing customer, and it needs to be managed as such.

Doctors still write the script. But AI and patients increasingly shape the decision-making process.

AI as customer and channel

AI at the point of care feels complicated, and rightfully so. It spans general-purpose LLMs, EHR platforms, in-house solutions like Mayo Clinic’s Nurse Virtual Assistant (NVA), and disease-specific tools. In such a fragmented landscape, it remains to be seen where influence will concentrate. Adding further complexity is the fact that AI tools occupy two roles at once: customer and channel.

As a customer, AI can be influenced. It’s no longer enough to articulate and substantiate differentiation through traditional channels. Pharma companies have to determine what AI systems know about a product and shape how AI represents it in generative responses. In crowded disease areas, a therapy has always needed to reach the right treatment pathways and workflows, but now that therapy also needs to be represented in AI tools or it may never be considered.

As a go-to-market channel, AI relays messaging to physicians and patients. Although AI doesn’t directly sell or distribute, it speaks to on-label features and benefits, much like a sales rep. It’s well aligned with the goal of getting accurate information to the right person at the right moment, much like non-personal promotion. But it also competes, pulling control from the company’s own sales and marketing efforts.      

Many pharma companies are unsure about how to handle a stakeholder that is a customer, channel, partner, and rival at once. The issue lacks a natural owner because it cuts across medical affairs, evidence generation, commercial functions, and PR. Without an integrated strategy, each function focuses on its own part and risks losing the holistic view of how AI is shaping perception of the product. 

Although this is daunting, it’s not entirely unfamiliar. Leading pharma companies already have the muscles they need from managing influential third parties. Companies that apply the same approach to the AI ecosystem now can lock in meaningful, lasting advantage, namely, an accumulated learning curve that late movers will struggle to replicate.

How pharma can make the first move

None of this requires completely upending the commercial model. It calls for evolving the capabilities that pharma companies already have: sensing how content and evidence are being used, influencing how therapies are represented across tools, and coordinating cross-functionally to go to market.

Sensing

• Understand how evidence is used at the point of care. Map how doctors and patients use AI tools to make decisions and what they need at different stages of the care journey. This will clarify the right information to highlight.
  
  
• Track awareness, trial, and usage (ATU) for AI. Expand traditional ATU research beyond doctors to see how therapies appear across AI-CDS platforms, general LLMs, and patient-facing tools, such as symptom checkers and chatbots. Leaders will continuously monitor AI-driven queries, recommendations, and positioning, adapting evidence generation and marketing strategy accordingly. The ATU of the future will query systems, not just doctors.

Influencing

• Evolve the integrated evidence plan for an AI stakeholder. Since AI can synthesize far more evidence than any one doctor, comprehensive evidence portfolios that address a broader range of clinical questions, patient subtypes, and real-world scenarios become more valuable. The plan now has to consider not only which publications the evidence belongs in but also whether it’s readable and bite-sized enough for a model to use and whether there’s enough sub-cohort analysis to answer specific queries.
  
  
• Optimize for retrieval. Making evidence retrievable and accessible across AI tools, including an effective generative engine optimization (GEO) strategy, is now essential. When multimodal data and real-world evidence are presented in structured, machine-readable formats, it’s far easier for AI-CDS tools and general-purpose LLMs to retrieve, interpret, and apply the information.
  
  
• Move upstream in how clinical decisions are shaped. Leaders will partner with tool vendors, sharing data and expertise to influence how evidence is represented and surfaced. This will likely work best in rare disease areas where pharma companies have proprietary data that vendors value and can’t get elsewhere.  

Coordinating

• Designate product managers for AI tools. An important new role in pharma will be the product manager for AI-CDS platforms and general-purpose LLMs. The product manager will understand the brand strategy, run the ATUs of the LLMs, and have the technical fluency to understand LLM behavior, such as where they’re reliable, where they break, and why outputs drift. They’ll work closely with vendors to improve how the brand shows up. And they’ll help assess if the company has the right data to support that brand strategy, including data coverage, labeling, quality bias, and recency.
  
  
• Evolve the field role. Even though doctors are increasingly open to AI-assisted decision making, 66% cite a lack of trust as a top barrier to realizing value from the tools. That insight indicates how the field team can evolve, across medical science liaisons (MSLs), sales reps, thought leader liaisons, and key account managers. When a doctor hears something from a pharma rep and sees it corroborated by an AI tool, both gain credibility. Additionally, field teams can help healthcare providers and patients correctly contextualize the information that AI surfaces and bring questions and needs back to the organization. That will require upskilling sales reps, key account managers, and patient-facing roles to become more clinically fluent, as well as enabling MSLs to speak dynamically to a much broader body of evidence.
  

AI is already reshaping how evidence is surfaced and how decisions are made at key moments of care. The technology itself is revolutionary, but the changes that pharma companies need to make are more evolutionary, building on capabilities most already have. That makes speed the differentiator. New standards are emerging, and the companies that engage early will help form them.

The authors would like to acknowledge the contributions of Brittany Rodriguez, Kristin Moneyron, Monica Pinto Basto, Max Cuda, Fernando Rosas, Sudhanshu Sharma, Pushpal Tayal, Rushikesh Dharurkar, and Harshet Pushoula.