Why Primary Research is the Power Source for AI That Works  - FleishmanHillard

Article

August 11, 2025

By Marina Stein Lundahl

Generative AI isn’t a promise anymore. It’s here.  

In the momentum of this modern gold rush though, it’s easy to forget a critical truth: the power behind these tools is still human. The quality of generative AI outputs depends on the inputs we feed them, and that begins with the rigor of primary research.  

Since 2023, the use of generative AI by organizations has more than doubled, with 71% of companies leveraging it by 2025. One standout application? Synthetic audiences, a powerful new way for communicators to gain insight into their audiences’ attitudes, perceptions and behaviors. But just like it’s easy to get swept away by the wave of generative AI, it’s easy to think that synthetic audiences are rendering traditional primary research obsolete. Nothing could be further from the truth.  

Synthetic audiences can’t outrun the human source 

Primary research and AI aren’t in competition. They’re codependent. 

The best synthetic audiences are built on the back of great human data. On the other hand, primary research can be made more focused and agile when layered with synthetic audience outputs. Synthetic audiences can extend the life of primary research when we incorporate real-time news or cultural data, keeping the insights fresh and up to date. Understanding the complexities of this relationship enables researchers to maximize benefits of both methods.  

As the old saying goes garbage in, garbage out. 

That’s never been truer than it is today. 

The Human Edge: What AI Still Can’t Simulate 

AI’s emergence has elevated the importance of research design and data quality vigilance, as MRS chief Jane Frost highlights in her article covering the Global Data Quality Initiative. Now more than ever, poorly designed studies don’t just lead to flawed short-term insights; they embed those flaws into synthetic audiences that rely on these studies as crucial training datasets. When applied carelessly, this flawed insight can lead to misinformed decisions that create business or reputational risk. 

This new reality demands that we approach primary research with heightened rigor and foresight. The questions we ask, the participants we recruit and the methodologies we employ must all be optimized not just for their immediate results but for their value as training inputs for AI models that expand the radius of these data.  

The equation is simple: better human data lead to better AI models. Human insights provide texture and nuance that synthetic models currently fail to accurately simulate. 

Cultural Context: AI models struggle to understand deep-rooted and implicit cultural knowledge that humans navigate effortlessly through lived experiences 

Emotional Nuance: The richness and range of human emotional responses remains difficult to synthesize  

Emerging Behaviors: Primary research captures to-the-moment changes or evolutions in human behaviors before they become widespread enough to appear in secondary sources 

Contradictions and Complexity: Humans often hold conflicting views simultaneously; a complexity that enriches our understanding but challenges AI models 

These qualities aren’t “nice to haves.” They’re ingredients for insight that inspire action. The kind of action clients, policymakers and customers can trust.  

The ‘garbage in, garbage out’ dynamic shouldn’t be viewed as loose guidance for fine-tuning virtual audience models; there are real risks involved when primary sources are undervalued (e.g., algorithmic bias, insight homogenization and missed innovation opportunities). 

Reimagining Primary Research for the AI Age 

While critical to the relevance and credibility of AI-driven audience research, traditional primary research isn’t immune to the pressure to adapt and evolve in the age of advancing generative AI. Today’s research must be crafted with dual purposes: 

Delivering precise and actionable insights 

Creating high-quality, scalable inputs for AI systems and synthetic audiences 

This evolution means considering: 

Data Structure: How will this data need to be formatted to serve as effective model inputs? 

Comprehensive Capture: Are we collecting the contextual information AI needs for proper interpretation? 

Longitudinal Value: How will this data remain relevant as behavioral patterns evolve? 

Ethical Considerations: What guardrails ensure our data fuels responsible AI development? 

Forward-thinking organizations recognize that the competitive advantage is not in choosing between primary research and synthetic audiences, but in their purposeful integration. Investing in the quality, design and implementation of primary research is no longer optional. It’s a requirement to fuel the next generation of insights, both human and artificial.  

As we navigate this rapidly evolving landscape, we’re firmly planting the flag:  

Primary research isn’t just still relevant, it’s more important than ever and will improve synthetic audiences.