We study the disclosure of influencer posts on Twitter across a large set of brands based on a unique data set of over 100 million posts and a novel classification method to detect undisclosed sponsorship. Using our preferred empirical specification, we find that 96% of sponsored posts are not disclosed. This result is robust to a series of specification tests, and even a lower-bound classification still yields an undisclosed share of 82%. Despite stronger enforcement of disclosure regulations, the share of undisclosed posts decreases only slightly over time. Compared with disclosed posts, undisclosed posts tend to be associated with young brands with a large Twitter following. Using an online survey, we find that many consumers are not able to identify sponsored content without disclosure. Our findings highlight a potential need for further regulatory scrutiny and suggest that researchers studying influencers must account for undisclosed sponsored content.