Synthetic Buyer Persona: A Revolution in User Research or a Risky Gamble?

A synthetic buyer persona can be defined as a large language model (LLM) trained to act like a real person. These models have learned to mimic human behavior. In fact, researchers from Peking University (Jiang et al., 2023) claim that these models develop a personality that guides their behavior, which can be induced through prompting. With AI, these models simulate behaviors, preferences, and even personalities, allowing researchers to interact with them as if they were real users.

The data used to build them includes demographic and psychographic data, purchase history, online behavior, or social media activity.

Synthetic data—data not created by humans—mimics real-world data. It’s created using algorithms and simulations based on generative AI. A synthetic dataset has the same mathematical properties as the real dataset it is modeled on. That is, it replicates the statistical features, distributions, correlations, and patterns of authentic data, but without containing the same information—allowing it to be used in analysis without compromising privacy or requiring direct user data collection.

These language models can act as “virtual” subjects, assigned with specific preferences, information, or situations (e.g., how much they value fairness, their political ideology, etc.). Researchers can then observe how these personas “respond” in various economic experiments and simulations. In different experiments, these models respond similarly to humans, although with nuances depending on the LLM version or how the simulated personality is “configured,” according to John J. Horton’s study Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus? (Cornell University, 2023).

While synthetic buyer personas offer speed and scalability, their lack of nuance and the risk of bias may compromise the authenticity of insights.

• Speed and flexibility: A 100% synthetic persona can be created quickly, without needing to conduct surveys or interviews with real users. This enables fast testing of different scenarios, attributes, or market hypotheses. AI-generated synthetic data and simulated scenarios lead to deep, fast market understanding—delivering advanced market intelligence and faster insights.
• Cost and time savings: Without the need to recruit or coordinate study groups, traditional research costs (focus groups, interviews, etc.) are greatly reduced. Time is optimized for exploring multiple profile variations. One of the biggest challenges in user research remains participant recruitment, with 97% of researchers facing difficulties. Major issues include finding participants who meet specific criteria (70%) and slow recruitment timelines (45%). Synthetic personas can help mitigate this.
• Continuous iteration and refinement: As a "living" and adjustable profile, it's easy to update with new expert input or data—enabling ongoing refinement of marketing strategies or product development. This encourages data-driven decisions and better adaptation to market changes.
• Scale and scenario variety: One or more synthetic personas with different characteristics (motivations, locations, etc.) can be generated to represent various market segments. This facilitates testing of new marketing or product strategies without additional investment in field research.

70% of researchers are already using AI in their workflows, reflecting a growing trend toward research process optimization.

Despite the benefits, synthetic buyer personas also come with important limitations:

• Lack of nuance and authentic perspectives: Real human research provides nuanced insights that AI or simulated data may miss—like specific fears, personal anecdotes, authentic language, and unconventional expressions.
• Risk of bias in generation: Any AI model may inherit biases from the data or algorithms used to train it. Interviewing real people allows researchers to cross-reference data and reduce the risk of reinforcing prejudices or unrealistic assumptions.
• Lower empirical validation: In-person research gathers direct evidence of how an audience perceives, uses, or values a product or service—offering stronger validation of marketing or design hypotheses.
• Inability to capture human spontaneity and improvisation: In traditional qualitative research, open conversations, follow-up questions, and spontaneous interaction uncover latent needs or hidden pain points that might otherwise go unnoticed.

Seeing the pros and cons of 100% synthetic personas, it’s clear that they may work well in some cases—but not all. That’s why we believe the ideal solution lies in a hybrid model.

The key is not choosing between synthetic or traditional personas, but adopting a hybrid approach that blends efficiency with depth and authenticity.

In light of all this, a promising solution is to adopt hybrid methodologies that combine the best of both worlds. This breaks the traditional boundaries between qualitative and quantitative research, creating a methodology that’s neither purely one nor the other.

While models can be trained on data, it’s also possible to explore qualitative depth and richness of experience.

This data can come from both qualitative and quantitative studies previously conducted with real people.

A hybrid approach allows teams to harness the speed and scalability of synthetic personas while preserving the depth and authenticity of traditional research.

In other words, synthetic buyer personas can be created relatively quickly based on data previously gathered from real users. These personas would offer the same flexibility and responsiveness but would behave in ways already validated by actual human research.

Synthetic buyer personas are a promising innovation, but their exclusive use could lead us down the wrong path if not combined with real user research.

100% synthetic buyer personas represent a promising innovation in user research. They offer undeniable benefits in speed, cost, and scalability. However, it’s crucial to acknowledge their limitations and the risks of relying on them exclusively. Lack of nuance, the risk of bias, and limited empirical validation are issues that cannot be ignored.

In our view, the present and near future of user research likely lies in a hybrid approach, where synthetic buyer personas complement traditional methodologies. By embracing this mixed strategy, companies can harness the power of AI to optimize research processes—without sacrificing the depth and authenticity that only human interaction can provide.

That said, we don’t rule out the possibility that generative AI may, at some point, fully replace real user research in specific contexts. We’ll be keeping a close eye on the next steps. :)