The term “Clickbait” refers to content that has the express intention of grabbing the reader’s attention. It has become an annoyance for social media users, because of the deception contained in the titles of Clickbait. Many studies detect Clickbait using Deep Learning and Machine Learning models. But detecting Clickbait in Arabic titles was addressed by a few studies, all of which used Machine Learning techniques, and here is where our turn came from. In our proposed work ”Wals” which is an expression of lying in Arabic, we will detect Clickbait in the Arabic titles using Deep Learning techniques which are Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNNs), Long Short Term Memory (LSTM), Gated Recurrent Unit (GRU), and Bidirectional Gated Recurrent Unit (BiGRU). We have followed two tracks to train models to reach the Wals model to find an optimal model with the suitable optimizer and best case of different cases (with or without pre-processing and with or without Word2vec). The models were trained using an unbalanced dataset in the first track and using a balanced dataset in the second track. According to the best results obtained from the first track, we built the first Wals hybrid model consisting of LSTM+BiGRU using the Adam optimizer, and by applying Word2vec and pre-processing, the LSTM model obtained the highest Macro-F in the track equal to 0.79 using Adam and by applying both Word2vec and pre-processing, the BiGRU model also obtained the same LSTM value in the same path, but by applying pre-processing only. For the second track, we built another Wals hybrid model consisting of LSTM+CNN, the LSTM model got the highest accuracy in the track equal to 0.79 using Adam and by applying both Word2vec and pre-processing, the CNN model also got the same LSTM value with the same track but without applying either Word2vec and pre-processing. Its results did not show a clear difference between the two tracks, and for the use of Early Stop, some models showed better results than those not using it such as LSTM, BiGRU, and CNN. Also, all the two hybrid models of Wals gave close results, reaching an accuracy equal to 0.77 and with the Early Stop application as well.