An automatic feature extraction for classifying the clickbait for Thai headlines is presented. The first corpus of 132,948 Thai headline news was collected. To transform Thai words into features, Word2Vec is utilized to overcome the ambiguity of the word segmentation. Then, the features are automatically extract using a Convolutional Neural Network (CNN). A number of experiments for CNN have been conducted to find the suitable value of the parameters that achieve the best classification result. We found that using a non-static modelling technique together with 50 dimension of Word2vec feature, {2, 3, 4} window size, and epoch equal to 5 achieves the accuracy of 95.25%. The experimental results also showed that the proposed method achieves the best result as compared to the other classification methods such as Support Vector Machine (SVM) and Naïve Bayes, which achieve 87.17% and 87.32%, respectively.