This paper presents a modular pipeline for generating synthetic doctor-patient dialogues in European Portuguese using OpenAI models. The system is structured around the SOAP framework. Synthetic patient profiles are generated and used to prompt GPT-3.5-turbo, GPT-4, and GPT-4o to produce complete medical consultations. We conduct a lightweight structural evaluation of 30 dialogues, with 10 generated by each model, focusing on sentence count, section length, and content distribution. Results show that GPT-4o produces longer and more balanced outputs, with greater attention to symptom description and diagnostic reasoning. In contrast, GPT-3.5-turbo often produces outputs dominated by treatment planning. Qualitative examples further support these findings, revealing improvements in conversational depth and formatting consistency in newer models. This study demonstrates the feasibility of generating structured clinical dialogues in Portuguese and highlights the need for future work involving clinical validation and grounding through retrieval-augmented generation.