Develops a Bidirectional Encoder Representations from Transformers (BERT) modeling approach to control possible confounds of text and context and isolate the causal effect of specific words on reader engagement, charitable donations and crowdsource funding.