This thesis examines how a portfolio constructed with support from a large language model performed in comparison with selected actively managed funds and market indices. The study focuses on both total return and risk-adjusted return, since a portfolio that generates a positive return is not necessarily more attractive if it also involves higher volatility. The study is therefore designed as a short-term comparative analysis of one LLM-assisted portfolio with the aim of examining its performance in relation to selected funds and market indices rather than evaluating LLM-based portfolio management in general. The comparison is based on the same observation period and the same performance measures for all investment alternatives. The study covers the period from 23 March to 7 May 2026, corresponding to 33 price observations and 32 daily returns. Performance was evaluated using Holding Period Return, annualized standard deviation, and the Sharpe ratio. The study uses a quantitative and comparative research design. The language model Claude was used as decision support to select and weight 30 global large cap stocks. The stock selection was based on established financial criteria: momentum, valuation, profitability, and risk. The portfolio was then compared with three actively managed global equity funds: Swedbank Robur Globalfond A, SEB Global Aktiefond A and Handelsbanken Global Selektiv. It was also compared with three market indices: OMXS30GI, MSCI World Net USD and S&P 500 TR.    The results show that the LLM-assisted portfolio generated a positive return during the study period. Its Holding Period Return was 8.94 percent, which was higher than SEB Global Aktiefond A but lower than the remaining funds and market indices. The portfolio's annualized standard deviation was 13.48 percent, which indicated that the portfolio had higher volatility than several of the comparison groups. Its Sharpe ratio was 4.87. This was higher than OMXS30GI but lower than the actively managed funds, MSCI World Net USD and S&P 500 TR. Overall, the portfolio matched some comparison objects but was not superior to the comparison alternatives. The study indicates that Claude can support portfolio construction by applying predefined financial criteria and structuring publicly available information. At the same time, the results do not show that a prompt-based LLM-assisted strategy outperformed traditional fund management or broad market exposure during the short study period. The findings should therefore be interpreted with caution, especially since the study is based on one language model, one prompt, one portfolio, and a limited observation period.   

Keywords: Large language models, portfolio management, risk-adjusted returns, Sharpe ratio, active funds, market indices