I study economic incentives that determine the disclosure softness of corporate language. Using the MD&A section of 10-K filings, I measure disclosure softness by holistically aggregating linguistic attributes consisting of vague, tonal, forward-looking, numerical, and specific information, as well as novel metrics of historical and objective content based on natural language processing and machine learning algorithms. I find that firms provide softer disclosures during poorly performing years, and that this effect is stronger in less ambiguous settings, suggesting that the value of soft/hard information is conditional on the underlying information environment. These results are distinct from the effect of disclosure complexity as measured using the Fog index. In addition, I find that proprietary costs stemming from competition from incumbents versus potential entrants differentially influence disclosure softness. I corroborate my findings using the exogenous variation in entry threat resulting from import tariff rate reductions. My results suggest that performance and competition incentives have a pervasive effect such that managers use an arsenal of linguistic attributes in shaping their disclosure strategy. Overall, I take a first attempt at measuring and studying the economic determinants of disclosure softness, a concept germane to the historical debate on relevance versus reliability.