Synthetic Personas: Rethinking Access to Hard-to-Reach Target Groups

Recently, I sat down with a mid-sized client who was considering launching a user study. The focus was on a new software solution for procurement managers in the manufacturing industry. "But we just can't find these people," he said. "And when we do, each interview costs a fortune." A familiar issue: people with limited time, niche target groups, industries with access barriers. Traditional qualitative research often hits its limits here.

In this situation, I asked: "Why don't we simulate one of these procurement managers first?" And that was the entry point into the world of synthetic personas.

A Practical Tool - Not Science Fiction

We're currently experiencing a noticeable shift in the research world. Following innovation waves around mobile research and remote UX, AI technologies and synthetic data are now bringing fresh momentum. Not only are tools like ChatGPT or Claude being integrated, but research methods themselves are under pressure to justify their value more quickly and cost-effectively. This is where synthetic personas can help: AI-generated user profiles based on real sources, but fictionally assembled. What might sound like a gimmick is, in practice, often a surprisingly productive tool.

Back to our procurement manager. With a few well-researched sources - including publicly available expert interviews in trade journals, LinkedIn profiles and posts from procurement managers in similar companies, industry studies and whitepapers, as well as discussions from professional forums - and a precisely worded prompt, we succeeded in creating a synthetic persona: "Martin, 52, head of procurement at a supplier company." We had Martin react to a new product idea, asked him questions about his concerns, his KPI pressures, his view on supplier relationships.

The result isn't a substitute for a real interview.

But the AI can learn from sources and generate statements that provide additional insights: What arguments are convincing? What language works? Where are the mental barriers?

Once such synthetic B2B representatives are modeled, they can be "interviewed" - much like real experts. These simulations can provide valuable insights to sales and marketing teams regarding what needs to be considered in innovation and product design - before investing effort in real interviews. In UX research, synthetic users could be used to predict usability issues or derive necessary features by letting the AI describe a usage scenario.

Faster Access to Insights for Small Teams

Especially for smaller companies without dedicated UX or research teams, this access can be a game-changer. Synthetic personas can be created quickly, require no incentives, no data privacy agreements, no lengthy field times. They provide initial impulses that help formulate hypotheses or sharpen argumentation lines for sales and product development.

But - and this is key - the benefit stands or falls with the quality of the data, the clarity of the setup, and the methodological approach. Those who "generate a target group" without content depth will at best get generic marketing fluff. However, those who carefully select which sources to include, what contexts matter, and use good prompting will get an astonishingly credible reflection of a target group. One that stimulates thinking - not replaces it.

Synthetic personas are especially useful when real people are hard to reach. Think of healthcare workers, individuals under psychological stress, people in public institutions, or highly occupied executives. Often, a few publicly available sources are enough to build a well-founded perspective for an AI-supported dialogue. What's important is to treat them as a hypothesis space. Not as truth, but as an invitation for validation: "Would this be plausible?"

Acknowledging the Limits - Ensuring Quality

However, working with synthetic personas also comes with limitations. AI-generated individuals have no real experiences - they only aggregate what's found in textual sources. This means they tend to offer generic or average answers. Truly innovative ideas or unconventional solutions may be missing. There's also the risk of stereotyping: if training data frequently states "engineers focus only on technical details," the AI may overemphasize this cliché and ignore other perspectives. Real people are more diverse and must be understood in their context.

I see three typical risks:

1. Confusing plausibility with validity: Just because something "feels real" doesn't make it empirically correct. Synthetic simulations must be treated with great caution. They can generate hypotheses but not predict real decisions.
2. Overinterpretation: Teams might be tempted to use individual AI-generated statements as "proof" for their hypotheses.
3. Hallucinations: Synthetic personas often carry biases from their training data - such as stereotypical patterns or exaggerated group images. The AI may also invent experiences that never happened. These effects can cloud analysis if taken at face value.

Still a Valuable Tool - If Used Wisely

Despite these limitations, there are compelling reasons to use synthetic personas strategically. In early research phases - for hypothesis generation or testing questionnaires - AI makes qualitative research more agile and creative. Thinking options can be explored faster, and potential misunderstandings in question phrasing identified. Mistakes are less critical here, as they serve exploration.

AI also shows its strengths under tight budgets or time constraints: it provides rapid feedback, simulates reactions to campaigns or product ideas, and allows initial assessments across different countries.

But human expertise remains indispensable.

Synthetic personas are not autonomous insight machines.

They require human validation. This can occur on multiple levels: by triangulating with real people - for example, by conducting some actual interviews and comparing them with synthetic responses - or through plausibility checks by experienced researchers. Consistency checks within AI-generated texts (e.g., does the persona make contradictory statements?) can also indicate quality. Finally, established findings can serve as a reference: if statements deviate significantly from what's commonly known, that's a red flag.

Human research competence, contextual knowledge, and expert interpretation are still needed: What was surprising? What's missing? What should be validated in the real world?

Synthetic personas are not a revolution. But they are a serious tool to gain new perspectives on hard-to-reach target groups. Especially when used responsibly and interpreted critically. For me, they serve as an early-phase conversation starter - as a stimulus - not a replacement.