Personas are a cornerstone of user-centred design, but traditional methods for developing them are difficult to validate, prone to bias and labour-intensive. Data-driven approaches have improved scalability, but often lack the narrative richness and empathy that make personas effective. We present a methodology that uses large language models (LLMs) to accelerate the creation of personas while underpinning and constraining the results with contextual and empirical data. Our approach emphasises transparency and traceability: each generated persona attribute can be linked to its source material, including project documentation, workshop transcripts, survey results or other contextual corpora. By combining the narrative strengths of LLMs with the rigour of an evidence-based foundation, the method generates personas that are both descriptive and verifiable. We present a five-step workflow methodology: (1) generation of persona candidates from contextual data using LLMs, (2) iterative refinement to ensure representativeness of personas, (3) selection of the most relevant profiles through expert evaluation, (4) design of detailed persona profiles, and (5) enrichment with empirical evidence to ensure traceability and validation. The methodology is illustrated with a case study from the field of soil health, but can also be applied to other design contexts where alignment between different stakeholders is crucial. We argue that this approach positions LLMs not as a substitute for human expertise, but as an accelerator of persona work that improves accountability, reduces bias and facilitates communication in collaborative design processes.
Blažica et al. (Fri,) studied this question.