Artificial Intelligence
Medical Communications

We have been talking about patient insight for 13 years. What is holding us back?

September 11, 2026

By Andrew Nelson, Senior Director, Integrated Intelligence, Medical Communications

In 2013, I stood on stage at the Pharma Market Research Conference in New York and argued that pharmaceutical marketers needed to move beyond physician-centric models. Patients were already using social media to share experiences, influence one another and make health decisions. Pharma needed to listen.

Thirteen years later, I keep hearing the same call: we need patient insight earlier. The industry agrees. The challenge is still finding the budget to act on it.

But the economics have changed. We can now gather and interpret patient insight faster, at greater scale and through more sources than we could in 2013. Early patient insight has become a critical part of asset development. The challenge is ensuring it’s integrated early enough to influence the decisions that matter most.

The simplest reminder is still the most useful

Marketing professor Mark Ritson has a line that has stayed with me throughout my career: “You are not the customer.”

It sounds obvious. Yet organizations routinely make assumptions about what customers want, what patients need and what will matter to healthcare professionals, then test those assumptions later. By then, the research may be confirming decisions rather than shaping them.

Patient insight creates more value when it arrives early enough to influence clinical development, evidence generation, medical strategy and communications. That means treating it as a phased input, not a one-off project.

Patient insight should evolve with the product lifecycle

Preclinical to Phase 1, three to five years before launch
Listen to what patients are already saying about unmet needs, caregiver experiences, treatment burden, daily life and the outcomes that matter most.

Phase 2, two to three years before launch
Use that understanding to inform endpoint development, trial optimization, treatment preferences, segmentation and early narrative development.

Prelaunch, 12 to 24 months before launch
Focus on adoption barriers, the patient journey, support needs, influential patient groups and the content patients and healthcare professionals will need.

The questions change, but the principle does not: listen before the decisions are locked.

The financial case is stronger than many teams assume

A model led by the Clinical Trials Transformation Initiative examined the financial value of patient engagement in a typical oncology development program entering Phase 2 or Phase 3. In a pre-Phase 2 scenario, engagement that prevented one protocol amendment and improved enrollment, adherence and retention increased modeled net present value by $62 million and expected net present value by $35 million. Those results were based on an assumed $100,000 investment.

This does not mean every $100,000 spent on patient engagement produces a $35 million return. It’s a financial model, not a guarantee, but it challenges the idea that patient insight is something organizations can only afford later. The cost of listening early may be small compared with the cost of correcting a trial, evidence or communication strategy after the fact.

There’s a human case, too. Understanding lived experience can change the questions Medical Affairs and other teams ask:

What does the disease stop people from doing?
Which symptoms are most disruptive?
What would meaningful improvement look like?
Where do diagnosis and treatment break down?
Are we measuring what matters to patients?

The real challenge is getting credible answers quickly and cost-effectively. Three shifts make that more achievable.

1. Start with what patients are already telling us

Social media was a rich source of patient insight in 2013, and it remains one today. What has changed is our ability to find the signal in a vast amount of unstructured information.

The natural language processing tools I used 13 years ago returned a great deal of data, but also a great deal of noise. Researchers still had to filter, code and interpret individual conversations manually. The output was useful, but it worked best as an addition to traditional research rather than a robust source on its own.

AI can now process far more information, far faster. That does not mean letting a model loose on patient data and accepting whatever comes back. Teams still need control of the data, the sources, and, most importantly, the interpretation.

A published Sjögren’s study shows what that balance can look like. Real Chemistry worked with Novartis and the British Sjögren’s Syndrome Association to analyze 26,950 relevant social media posts across nine countries using AI-powered natural language processing and qualitative analysis. The research looked beyond the prevalence of symptoms to examine how bothersome they were and how they affected different areas of patients’ lives. Human analysts remained essential to understanding what the conversations meant.

AI gives us scale. Humans give us context. Together, they can surface insight that would have been far more difficult and time-consuming to reach in 2013.

Digital listening has limitations. Social media does not represent every person living with a disease. That is not a reason to dismiss it, but it is a reason to understand what it can tell us, where the gaps are and which additional sources we need.

2. Understand what AI is telling patients

Patients now have another influential source of health information: generative AI and AI-powered search. They’re using these tools to ask about diseases, symptoms, treatments and clinical information.

A March 2026 Real Chemistry survey of 525 consumers across seven markets found that 28% use AI search for health-related information. Among frequent users, 58% viewed these tools as reliable as traditional health sources, while 48% viewed them as reliable as their own healthcare provider.

That creates a new patient insight question for pharma: What are AI systems telling patients about your disease and your science, and which sources are shaping those answers?

The goal is to understand whether the information being surfaced is accurate, complete and current, and to make credible scientific information easier to find and understand. Real Chemistry’s HealthGEO is designed to examine that emerging information environment. If patients are asking AI about a disease, organizations should know what it is telling them.

3. Use primary research where digital evidence falls short

In rare and ultra-rare diseases, there may not be enough social conversation, forum activity, audio or video to generate the depth of insight teams need. Primary market research still has an essential role.

The opportunity is to modernize how it is used. Digital recruitment and smartphone-based participation can make it easier for patients to respond in their own time and share richer accounts of their experiences, including video. These approaches can extend reach without defaulting to the largest possible research program.

We should also test new ways to gather collective insight. A study published in npj Digital Medicine examined a real-time “swarm” approach in which groups of radiologists diagnosed pneumonia from chest radiographs. The approach outperformed individual radiologists and traditional crowd-based methods in the task studied, and performance improved further when the swarm was paired with deep-learning AI.

That evidence comes from healthcare professionals, not patients, so it should not be treated as validation for patient research. But it raises a worthwhile question: could a carefully recruited group of patients or caregivers deliberate together to reveal priorities that individual interviews might miss? The method needs exploration, but the underlying idea is compelling. Patients do not experience disease in the same way, and collective intelligence may help teams understand where those experiences converge and diverge.

We don’t need to wait until launch to listen

In 2013, patients were using digital channels to share experiences and influence healthcare decisions. In 2026, we have more data, better technology and more ways to listen. Yet early patient insight is still too often treated as important but optional.

It is time to rethink what patient insight can look like:

Listen to existing patient conversations before commissioning new research.
Match the questions to the stage of development.
Use AI to find patterns at scale and human expertise to interpret them.
Examine what AI systems are telling patients and which sources shape the answers.
Use focused primary research when digital evidence is limited.
Test new, responsible approaches to collective patient intelligence.

The investment can be modest compared with the cost of getting a protocol, evidence plan, support program or narrative wrong later. The potential return is commercial and strategic, but the most important outcome is better decisions for patients.

Five years from now, I do not want to hear, “We know earlier patient insight is important, but we cannot find the budget.” I would rather hear, “We had the insight we needed, early enough to make a difference.”