Concealing a message in the APPn markers of a JPEG image not only protects the security of the message but also induces no change to the quality of the image. However, from the existing literature, whether it is a plaintext message or a ciphertext message hidden in APPn, it is easy for an attacker to identify the confidentiality of the hidden message, which is not conducive to the security of the message. Inspired by the natural language processing (NLP) and format-preserving encryption (FPE), this paper proposes a data hiding method, which is focused on the quality assurance of the host image and the concealment of the plain text having complete semantics based on the NLP and FPE. This method first uses the NLP and FPE to identify and encrypt the sensitive words in the plain text and then hides the ciphertext text with the plaintext style in the APPn of a JPEG image after replacing the plaintext words with the ciphertext words that have the plaintext style. The experimental results confirm that the structure, size, and quality of the host image do not show any changes before or after the data hiding and the recovered host image is also identical to its original appearance. In addition, the strategy that the semantic similarity is regulated autonomously by the user also makes it possible to obtain the ideal ciphertext words with very low similarity. Moreover, more than 80% of the ciphertext texts have reasonable semantics. Compared with the existing literature, our algorithm has a better performance.INDEX TERMS APPn, common text, format-preserving encryption, natural language processing, quality of host image.