ChatGPT as Co‑Tutor: AI Literacy, Metacognitive Use, and Portfolio Performance in an Undergraduate Investment Course

This paper reports on an AI-assisted, Python-based portfolio project embedded in a final-year undergraduate finance course enrolling accounting majors in the United Arab Emirates. Motivated by the gap between spreadsheet-centered teaching and data-driven industry practice, the intervention positioned ChatGPT as a “co-tutor” to support students’ transition toward computational and algorithmic thinking while fostering AI literacy and critical use of generative AI. The study adopts a qualitative-dominant mixed-methods, design-based research approach within a constructively aligned curriculum. Thirty final-year accounting majors enrolled in an Investment Analysis course each implemented a pre-assigned diversified portfolio in Google Colab, downloaded live market data, estimated CAPM alpha and beta, and computed performance indicators. ChatGPT was integrated into the workflow for code generation, debugging, and conceptual explanation. Data sources comprised (a) portfolio performance summaries, (b) a post-project survey measuring learning and confidence in Python/finance, metacognitive strategies, and AI literacy, and (c) students’ open-ended reflections on their ChatGPT use. Portfolios exhibited realistic but generally negative returns, with an average total return of − 4.5% and CAPM betas clustering around 1.0, indicating technically coherent implementation rather than systematic computational failure. The survey scales demonstrated excellent internal consistency and moderately high mean scores, suggesting that most students perceived learning gains, engaged in checking and revision practices, and recognized ChatGPT’s limitations. Usage patterns indicated cautious, opportunistic use of ChatGPT rather than blanket reliance, and only weak, non-significant correlations emerged between AI-related scales and portfolio performance. Qualitative analysis highlighted four themes: ChatGPT as coding assistant rather than replacement for understanding; the centrality of verifying financial logic; the development of prompting and debugging strategies; and the tension between accelerated experimentation and ethical responsibility. The paper concludes with design principles for AI-assisted experiential learning in finance and provides an open replication package (notebooks and datasets) to support adaptation of the intervention in other quantitative courses.