Product reviews on e-commerce platforms can have a pronounced effect on consumers' decisions. Less is known, however, whether the reviews written by others can shape a person's own written opinion of a product. We hypothesized that people who compose reviews on digital storefronts will try to imitate successful reviews, such that their content will show similarity with other reviews displayed at the time of writing. More specifically, we predicted that reviews would be more semantically similar to the most successful, salient, and readily accessible reviews written by others. To investigate this hypothesis, we extracted over 3 million reviews from a major online distribution platform and traced the reviews that were displayed at the time when each review was being composed. Using word embeddings from a pretrained language model, we quantified the semantic similarity between a given review and other reviews that were visible (or not) to a user. We found that reviewers imitate the most helpful reviews written by others, especially those that are visually salient. Their reviews, in turn, gather more helpfulness ratings in the future, leading to a cascade of similar reviews. Our findings suggest that the default sorting and display format of reviews on online platforms will have a pronounced effect on the style and content of new reviews.