Since 2001, 76 regional trade agreements (RTAs) have incorporated provisions on electronic commerce, representing 44% of all RTAs signed over the period. We build an original dataset identifying these agreements (which we call E-RTAs) using an algorithm that combines machine learning and text analysis to detect e-commerce provisions in 449 trade agreement texts. We document that E-RTAs are geographically concentrated among a handful of developed economies and that signatory pairs trade less and are farther apart than pairs linked by other RTAs. Using a gravity framework estimated via Poisson pseudomaximum likelihood (PPML), we find that E-RTAs increase trade by less than the average RTA: conditional on the standard RTA dummy, the E-RTA coefficient is -0.245, implying approximately 22 percent less trade. Neither the number of e-commerce provisions nor their legal bindingness affects this result. However, when we interact E-RTA status with a measure of the combined development level of the trading pair, the interaction coefficient is 0.134, and e-commerce provisions increase trade between sufficiently developed country pairs. These results are robust to alternative specifications, estimators, product-level disaggregation, services trade, and tariff controls.