This is the dataset used for the research "The Good, the Bad and the Bait: Detecting and Characterizing Clickbait on YouTube", with DOI: 10.1109/SPW.2018.00018. The dataset consists of three files: 1. groundtruth.json: This is the groundtruth dataset. We have 3443 manually annotated videos (we manually annotated more after the acceptance of the paper), and 17,648 videos that were obtained from channels that post clickbait or not. You can distinguish the method of annotation by observing the field "comments" in "clickbaitClassification" (the ones that have the comment "channels" are the ones obtained from the channels). 2. videos.json: Contains the data for 206K videos that were obtained as described in the paper. 3. predictions.json: It contains the mapping between the video id and the probability of our classifier. In our paper, we treat a video as clickbait if the probability is larger than 0.5. The related software produced for this study may be found here.