TLDR: This research explores using prompt engineering with Generative AI (GenAI) to create proto-personas, essential for early product discovery. A case study with 19 participants showed the AI-powered approach drastically reduced creation time (from days to minutes) and effort, boosting productivity. Participants found the method highly useful and easy to use, leading to strong acceptance. While cognitive empathy with AI-generated personas was high, affective and behavioral empathy showed mixed results, indicating challenges in fostering deeper emotional connections with artificial user profiles. The study highlights GenAI’s potential for efficiency and collaboration in software engineering, alongside areas for future improvement in human-like persona generation.
In the fast-paced world of software development, understanding user needs early on is crucial. This is where ‘proto-personas’ come in – early representations of users based on assumptions and limited data. They help guide product definition and align teams during initial stages like Lean Inception. However, creating these proto-personas manually can be a time-consuming, mentally demanding, and often biased process.
A recent study, titled Generating Proto-Personas through Prompt Engineering: A Case Study on Efficiency, Effectiveness and Empathy, explores a novel approach: using Generative AI (GenAI) and prompt engineering to automate and enhance this process. The research team, including Fernando Ayach, Vitor Lameirão, Raul Leão, Jerfferson Felizardo, Rafael Sobrinho, Vanessa Borges, PatrÃcia Matsubara, and Awdren Fontão, empirically investigated how GenAI could improve the efficiency, effectiveness, user acceptance, and the level of empathy elicited by these AI-generated personas.
The AI-Powered Approach
The core of their method is a refined prompt-engineering approach, building on existing guidelines and patterns. They utilized ChatGPT 4o-mini, a GenAI model, to execute a five-step process. This involved first setting the context for the AI, then feeding it key product information like the ‘Product Vision’ and an ‘Is/Is Not/Does/Does Not’ matrix. Next, a predefined template was provided to ensure the proto-personas generated by the AI adhered to Lean Inception standards, including essential demographic data such as age, occupation, marital status, and education. Finally, the AI was instructed to act as a UX/UI designer and generate at least five proto-personas, a number recommended for capturing most user needs.
To combat common AI issues like ‘hallucinations’ (generating incorrect or nonsensical information), the team implemented specific prompt patterns. For instance, a ‘Context Manager’ pattern helped maintain a consistent understanding of the product, while a ‘Persona’ pattern ensured the AI’s outputs remained within the UX/UI design domain. These refinements aimed to make the AI-generated personas more complete and accurate.
A Real-World Case Study
To test their approach, the researchers conducted a case study with 19 participants involved in a real Lean Inception workshop for an AI-driven web application project. This project aimed to assist state attorneys and public sector legal professionals, a complex domain with specific terminology and regulatory constraints. The participants, a multidisciplinary team of software engineers, computer scientists, data scientists, and legal experts, engaged in half-day sessions over a week.
Data was collected through various means, including recording the time taken for persona generation, administering questionnaires based on the Technology Acceptance Model (TAM) and empathy dimensions, and conducting semi-structured interviews. This mixed-methods approach allowed for both quantitative and qualitative insights into the approach’s performance.
Key Findings: Efficiency, Effectiveness, and Acceptance
The results were compelling. In terms of **efficiency**, the prompt-engineering approach drastically reduced the time required for proto-persona creation. What typically took days of manual work was completed in an average of just 5.94 minutes. This significant time saving allowed teams to dedicate more effort to strategic discussions and refining product functionalities, rather than the laborious task of initial persona generation. Participants reported a marked increase in productivity and a reduction in cognitive effort, especially for creative and writing aspects.
Regarding **effectiveness**, the AI-generated proto-personas proved to be valuable tools. Their richness, realism, and contextual alignment fostered empathy among team members, motivating them to adapt the product to better meet user needs. The personas served as a central reference point for refining functionalities and user journeys, particularly useful when direct stakeholder access was limited. They also guided the understanding of the Minimum Viable Product (MVP) and prototypes, introducing new perspectives on potential users.
User **acceptance** of the approach was generally high. Participants overwhelmingly found the approach useful (100% agreement) and easy to use (95% agreement). They appreciated the quality and richness of the generated personas and their utility as a basis for discussion, which streamlined the refinement process. While some noted minor concerns about generalization in highly specialized domains or occasional confusion with the research setup, the overall attitude towards using AI tools for UX activities was positive. Most participants expressed a strong intention to reuse the approach and adopt similar tools in the future, highlighting its compatibility with existing workflows.
The Nuance of Empathy
One of the most intriguing aspects of the study was the evaluation of **empathy**. The researchers assessed three dimensions: cognitive, affective, and behavioral empathy.
Cognitive empathy, which relates to understanding and interpreting the internal states of others, was strongly supported. 100% of participants found the AI-generated proto-personas easy to understand and interpret, largely due to the inclusion of comprehensive attributes like age and function in the template. This allowed participants to more tangibly imagine the users they were designing for.
However, **affective empathy**, the ability to feel the same emotional state as another, showed mixed results. Nearly half of the participants (47%) expressed neutrality or disagreement regarding forming an emotional connection or seeing the personas as real people. This suggests a barrier in developing a deeper emotional bond with AI-generated personas, often due to a divergence in context between the participant and the persona, or a perceived artificiality.
Similarly, **behavioral empathy**, demonstrating intentional behavior based on understanding others’ states, also yielded varied outcomes. While some participants identified with the proto-personas based on their described behaviors, others lacked the motivation to act in favor of them, feeling that the AI-generated personas lacked sufficient human details like routines and preferences. This points to a limitation of GenAI in fully capturing the nuanced human elements necessary for fostering strong behavioral empathy.
Also Read:
- Exploring the Creative Frontier: Can Artificial Intelligence Truly Be Creative?
- AI’s Ethical Compass: How Multi-Agent LLMs Are Shaping Responsible Systems
Looking Ahead
Despite some limitations, such as evaluating only one LLM and conducting the study in a specific domain, the research provides valuable insights. It demonstrates that prompt engineering can significantly boost efficiency in product discovery and enhance collaboration. It also highlights the ongoing challenge of imbuing AI-generated content with the emotional depth needed for full affective and behavioral empathy. Future work aims to develop more intuitive tools, integrate image generation, improve non-functional aspects like accessibility, and explore hybrid strategies that combine AI generation with human input to strengthen empathy and contextual fidelity.
