Human context in an age of synthetic research

Most discussions around synthetic data tend to start in the wrong place, with people asking whether synthetic respondents are good enough to replace traditional research or whether they represent some kind of existential threat to the industry. Those are natural questions when you're facing a major shift in how work gets done, but they miss the bigger point: what's changing here is the operating model around research.  

Most discussions around synthetic data tend to start in the wrong place, with people asking whether synthetic respondents are good enough to replace traditional research or whether they represent some kind of existential threat to the industry. Those are natural questions when you're facing a major shift in how work gets done, but they miss the bigger point: what's changing here is the operating model around research, with synthetic tools becoming part of the work that happens before organizations ever speak to customers.  

How the economics of creation have flipped 

For a long time, the biggest constraint in innovation was production. Building concepts, creating prototypes and developing variants all took significant time and money. Research sat downstream from that work, helping organizations evaluate a relatively small set of ideas before committing resources to them. 

That equation has changed. I recently heard a quote that it costs less to actually build a prototype and show it to customers than it does to hold a meeting about whether you should build it in the first place. Whether or not that's literally true, it captures what's happening. We can generate hundreds of concepts, messages and product variations in an afternoon. Then we’re faced with deciding which possibilities deserve attention. That's a human question requiring a deep understanding of what customers need, what they value and what's changing in their lives. This is where the synthetic layer starts to earn its keep by giving researchers a way to explore a much wider set of possibilities before they go to real people, so the human research they do is better and more focused. 

Synthetic tools work upstream 

When we talk about synthetic data in market research, we generally mean AI-generated personas or personas designed to simulate how customer segments might respond to something. The word approximate carries real weight here, since these simulations represent neither lived experience nor emotional truth. They function as tools to help us explore and pressure-test ideas before we invest in real human research. 

My team, which leads innovation and design at a market research technology company, has been experimenting with systems that start with synthetic personas for different segments, then generate multiple synthetic respondents based on each one. We control for variability and memory as we explore how these systems can help surface edge cases, pressure-test concepts and expand our thinking about a problem. 

The opportunity is to use synthetic tools as part of a larger workflow. Before you spend real respondent time, you can explore possible reactions, surface tensions and improve the quality of what you take to humans. The whole point is to narrow the field so the conversations you have with real people are more focused and capable of answering the questions that matter most.  

Fluency isn't evidence 

One of my biggest concerns surrounding this whole subject is plausibility. These models are very good at producing outputs that sound reasonable. A synthetic respondent can give you a polished, confident answer, but fluency is not evidence. These models can sound true before they are true, and that is where teams need to be careful. 

There’s also something I think of as context rot. Most organizations experimenting with synthetic research quickly discover that the quality of the output depends heavily on the quality of the customer understanding behind it. Off-the-shelf language models know very little about your customers, so teams often begin adding research findings, community conversations, support interactions and other customer knowledge to give those systems more context. 

That creates a new responsibility. Those knowledge layers need maintenance and oversight. If they are updated automatically without enough human review, small inaccuracies can accumulate over time. Assumptions get reinforced, nuances disappear and the picture of the customer can gradually drift away from reality. The synthetic outputs may still sound convincing, but they are now drawing from a version of the customer that is less accurate than teams realize. 

Different decisions require different levels of evidence. Using synthetic personas to explore possible objections to an early concept carries much lower risk than using synthetic output to support a major investment decision. The higher the stakes, the more you need real human signal involved in the process. 

Connecting the layers 

One practical direction is building research systems where multiple layers work together rather than operating independently. Picture an organization with an active insight community, a knowledge base of customer understanding and a synthetic layer all working together. Ideas come in, and synthetic tools help explore possible reactions and narrow the field. The most promising directions go back to the community for real validation. That human feedback feeds the knowledge base, which helps make the next cycle smarter and more informed. 

It also changes the role of insights teams. Historically, research often functioned as a gatekeeper. The business had ideas, research helped evaluate them, and that process sometimes felt like it slowed things down. In an environment where AI can generate endless concepts, that role starts to shift. Organizations still need people who understand methodology, uncertainty and the limits of simulated responses. As synthetic systems become more common, evaluating where simulated feedback is useful and where direct customer input is required becomes part of the research function.  

I see two roles becoming more important. The first is stewardship. Insights teams will be responsible for the quality, freshness and interpretation of the human context that AI systems rely on. The second is translation. AI can summarize patterns, but researchers help organizations understand what those patterns mean and what to do next. That's a role focused on helping organizations make decisions, not simply running studies. 

Maintaining human context 

AI can automate more of the mechanics of research than ever before, which changes where researchers create value. Communities, support conversations, behavioral data and research findings become part of the customer knowledge these systems depend on. 

As teams add more of that information to synthetic research workflows, the work doesn't end once the knowledge base is built. Research findings need to be updated. Assumptions need to be revisited. Customer priorities, behaviors and expectations change over time. 

Synthetic systems draw from whatever context they are given. As that understanding changes, the systems need to change with it. Maintaining that human context is becoming one of the core responsibilities of modern insights teams.