Agricultural price volatility poses significant challenges for producers, distributors, and consumers, underscoring the need for improved forecasting methods. Traditional statistical and machine learning techniques often struggle to capture the complex, dynamic and unpredictable nature of agricultural markets, which motivates the exploration of alternative approaches. In this study, we propose a novel method that converts time-series data into text prompts to enhance forecast accuracy by utilizing the language processing capabilities of large language models (LLMs). We analyze daily trading prices for 23 crops over 1,862 days to predict price changes over the subsequent 185 days under stable conditions. Unlike traditional methods, our approach employs customized prompt template to convert numerical data into text, enabling LLMs to more effectively interpret and forecast market trends. A comparative analysis with Holt-Winters, LSTM, GRU, and Transformer models demonstrates that our LLM model (Llama 3) outperforms traditional methods, achieving a MAE of 913.61, RMSE of 1477.74, MAPE of 28.89, and an R² of 0.6494. These findings suggest that converting time-series data into text prompts enables the use of LLMs, substantially improving agricultural price forecasting and leading to more accurate and data-driven market decisions.