This research investigates the impact of specific emotions in electronic word-of-mouth (eWOM) on the monthly donations received by a non-profit organization (NPO). We employ the empathy-helping (empathy-altruism) hypothesis as a theoretical foundation, proposing that donation motivations should inform eWOM fundraising appeals. To do so, we analyzed 71,462 tweets about a charity from 23,430 users, categorizing them as either marketer-generated content (MGC) or user-generated content (UGC). To automatically detect six distinct emotions in the text, we utilized a transformer-transfer learning approach for emotion detection. This model was trained in a sequential manner, starting with self-reported emotions in over 3.6 million tweets and progressing to socially agreed-upon emotion datasets to mimic social-emotional development stages. Our findings revealed that emotions prompting empathy (such as sadness in MGC) and positive empathy (like joy in UGC) positively influence donation amounts in line with the empathy-helping hypothesis. We offer insights on how social media marketers can leverage these results to create and manage tweets that boost donations. This study contributes to marketing research and practice in three ways: (1) by being the first, to our knowledge, to examine the effect of specific emotions in eWOM on donation decisions, (2) by introducing a novel machine learning model capable of detecting emotions in large-scale eWOM, and (3) by providing actionable recommendations for NPOs to increase donations via emotionally driven social media messaging. As a result, marketing managers can more effectively use social media platforms to foster emotional connections between NPOs and donors.