The growing adoption of Electric Vehicles (EVs) introduces new uncertainty into power system operation, increasing importance of EV charging-session forecasting. This study addresses the capability of Large Language Models (LLMs) in forecasting EV charging sessions, across such variables as start time, duration, and amount of transferred energy. LLM-based prompt strategies are benchmarked against mean and median baselines, and classical machine-learning models using an 80/20 chronological split with a walk-forward protocol. The analysis is conducted on a real-world dataset from Electric Vehicle Supply Equipment (EVSE) operated in Porto, Portugal, containing 2,933 sessions from 69 users and 13 EVSEs. The results indicate that LLMs outperform the considered baselines and machine-learning models in predicting the start time of the next charging session. In contrast, transferred energy and duration predictions show no consistent improvement, with simpler models performing better in some cases. This suggests LLMs can capture useful temporal charging patterns, but hybrid approaches may be required for robust transferred energy and duration forecasting.