ANALYSIS OF PROMPT ENGINEERING EFFECTIVENESS IN STOCK RECOMMENDATION BY CHATGPT: AN EXPERIMENTAL STUDY IN THE INDONESIAN MARKET

Abstract

This study investigates the influence of prompt engineering on the quality of stock recommendations generated by ChatGPT in the Indonesian energy sector. Using four distinct prompt types ranging from general to highly structured the research analyzes outputs related to five IDX-listed energy stocks. Each ChatGPT response was evaluated using four binary-coded indicators: analytical depth, indicator integration, scenario contextualization, and actionability. The findings reveal that structured and specific prompts produce significantly more accurate, relevant, and actionable recommendations. Among all prompt types, time-bound and context-rich prompts delivered the highest performance, while vague prompts yielded generic, low-quality outputs. The results support the importance of prompt literacy and suggest that effective human-AI interaction in financial decision-making depends heavily on input clarity. This study contributes to the growing literature on generative AI in finance and highlights the need for user education in prompt design.