Agentic systems based on large language models (LLMs) are increasingly integrated into requirements engineering (RE) workflows. Among approaches to improve contextual alignment and domain specificity, the persona pattern, where LLMs emulate stakeholders or domain roles, is widely used. However, guidance on designing prompt and context engineering strategies for effective persona adoption is limited, and empirical evidence on whether LLMs truly follow these personas remains scarce. We present a research preview aimed at guiding the adoption and evaluation of persona patterns in LLM-supported RE tasks. Our roadmap includes three steps. First, we conduct a literature review on persona formalization in RE to derive a taxonomy of persona characteristics. Second, we design prompt engineering guidelines to translate these persona models into structured system prompts for LLM-based systems. Third, we evaluate whether LLM outputs reflect the intended persona characteristics within specific LLM-based RE tasks. Preliminary results from an initial iteration include (i) a first version of the persona modelling taxonomy, (ii) a prompt engineering approach grounded in related work, and (iii) an illustrative example demonstrating the effects of applying the persona pattern in dataset annotation. Early findings suggest that commonly used prompting techniques may not reliably induce the intended persona behaviour, motivating more systematic design and evaluation practices.