This paper uses a transformer-based model for text classification in the area of mental health-related content. We apply XLNet to classify text data based on its mental health relevance and toxicity level. Using Mental Health Corpus dataset, the model categorizes text into two main classes: toxic, which may indicate mental health concerns such as emotional distress or harmful thought patterns, and nontoxic, which is neutral or supportive. The model detects harmful content while encouraging mental health awareness. XLNet effectively identifies mental health cues in online discussions. The proposed model achieves 96.4% accuracy, 96.9% precision, and 96.4% F1-score that demonstrates its effectiveness in detecting inappropriate mental health content.