Email phishing attacks are rapidly increasing in frequency for individuals, organizations, and nations, causing financial losses and trust issues. Many traditional methods have been proposed for identifying email phishing attacks, including blacklist-based and signature-driven methods. Recently, traditional machine learning and advance deep learning algorithms have been utilized to overcome phishing issues and provide enhanced detection performance because of their ability to detect sophisticated phishing attempts. However, phishing attacks are becoming increasingly complicated and harder to identify because advanced techniques require more efficient and reliable detection methods. Thus, this paper proposes a hybrid deep learning method that detects phishing attacks by utilizing only text content. The approach involves three main stages: preprocessing, feature extraction, and classification. First, N-grams combined with term frequency-inverse document frequency (TF-IDF) are applied to clean and transform the input data for further processing. Second, a recurrent neural network (RNN) is used to capture the complex and underlying dependencies between email words. Finally, a deep neural network receives the extracted features and acts as a classifier to differentiate between spam and nonspam emails. The performance of the proposed hybrid model is evaluated on the Assassin spam dataset, which consists of 5512 spam email samples for both training and testing. The proposed hybrid model outperforms previously developed deep learning algorithms, with a detection accuracy of 99.2%. In the future, the performance of this model can be enhanced by utilizing large language models (LLMs) and their variants.