This research addresses a fundamental challenge in Search-Based Software Engineering (SBSE): the high cognitive load and expertise required by Decision-Makers (DMs) to select appropriate objective functions during Product Line Architecture (PLA) design. In many-objective scenarios, such as those in the MOA4PLA approach, architects must often choose from 20 conflicting metrics without fully understanding their interdependencies or their impact on the final optimization. The primary contribution of this work is an approach that integrates Large Language Models (LLMs), specifically ChatGPT, to assist DMs by suggesting relevant objective functions based on natural language preferences. The research represents a transition from a strictly automated search paradigm toward a collaborative human-AI optimization framework. While earlier iterations of the OPLA-Tool required manual, expert-level selection of metrics, this study introduced a dedicated module that translates high-level architectural requirements—such as enhancing modularity or reducing structural coupling—into the corresponding technical indicators used by the search algorithm. The refinement process involved iterative implementation validation across four distinct stages, ensuring the model could handle domain-specific nuances, identify positive and negative correlations between various functions, and consistently provide structured data for seamless tool integration. Empirical evidence was significantly strengthened through a qualitative experiment involving domain experts from both academia and industry. The results demonstrate that the LLM-assisted approach achieves an 85% precision rate in aligning suggestions with DM preferences. Furthermore, the integration yielded a 40% average reduction in the time required for objective selection and a 25% perceived improvement in the quality of the resulting architectural solutions. Presenting this work at the ICSE 2026 Showcase highlights how Brazilian research is successfully leveraging generative AI to solve complex decision-making problems in search-based PLA design.