Users of social networks and Internet sites face numerous challenges. Problems such as fake news, satire, rumors, misinformation, misleading information, cyberbullying, spam content, offensive language, hate, and offensive speech fall under the category of harmful online content (HOC). This danger has taken advantage of social media's popularity and the abundance of news that spreads quickly, causing problems for individuals and society. Moreover, to combat this danger, researchers in the AI domain have persistently advanced and proposed novel approaches across various domains. Given the progress made in this work, choosing data to evaluate their approaches was always a challenge. Our contribution aims to identify the process and criteria for creating a high-quality dataset for HOC detection, primarily in the Arabic news domain. Therefore, we have collected a list of existing and available Arabic datasets, identified their characteristics, and determined the purpose of their creation. Researchers can use our study's results as a reference to choose an appropriate dataset for their future research.