June 14, 2023

AI-generated respondents: super-charged research or just synthetic insights?

By: Joe Marshall, Associate Director, Strategy at Evoke Mind+Matter

It’s likely we’ll look back on 2023 as the year that generative AI went mainstream. The relative merits and risks of ChatGPT have been endlessly debated. The EU are exploring stricter standards to regulate the technology. A disconcertingly convincing computer-generated Drake song went viral.

Much like other sectors, healthcare is only just beginning to scratch the surface of the potential uses for generative AI – as well as the potential implications. For the healthcare marketing and communications industries, one of the most immediately valuable possibilities of AI is the role it could play in market research, particularly predictive analytics. 

In its current forms AI can rapidly and efficiently support across a research project, from generating skeleton discussion guides to summarising open-ended data.

But it also has the potential to go far beyond this. Perhaps the most notable example is the creation of synthetic or virtual respondents. By combining multiple data points from a host of previous projects or studies, AI should theoretically be able to predict how particular audiences respond to certain questions or stimuli. While the idea of conducting research with computer-generated respondents might seem strange, there are potential benefits. It could certainly help to make research more cost effective and time efficient.

Market research within healthcare communications is almost always a longer and more expensive process than within consumer sectors. A major reason for this is that respondents are often more difficult to recruit – whether that’s due to their professional specialism or the prevalence of the disease they’re living with. Could generative AI help to address this challenge through the creation of synthetic respondents?

The prospect of gathering insights from larger samples is certainly attractive. We should always aspire to develop campaigns that are grounded in robust data and insight. But of course, it’s not that simple. How do we judge the point at which we have enough data to reliably model synthetic respondents? The prevalence of many rare diseases is so low that any attempt to predict responses based on existing data would not only be challenging but would arguably distort the realities of living with these conditions. And what might we overlook as part of this process?

At Evoke Mind+Matter we aim to make health more human™. To that end, some of our most successful campaigns have been grounded in qualitative insights. Whether the focus is rare disease or a broader public health initiative, these human truths can often form the basis for powerful behaviour change interventions. Generative AI will also invariably predict the most likely response to questions. That means research risks becoming very generalised, losing the individual viewpoints and nuances that are captured through speaking with real people.

AI is clearly here to stay and it’s important that we embrace its potential. It can help us conduct research more efficiently and optimise the way we use data to plan our campaigns. But we should also be aware of its limitations. In chasing the holy grail of larger samples and more robust quantitative data, there’s a risk that we lose sight of the humanity that sits at the heart of effective healthcare communications.