Studies of morphosyntactic processing have shown that listeners rely on grammatical gender cues and are sensitive to gender agreement violations during language comprehension. However, most research has focused on nouns with unambiguous gender assignment, overlooking cases of variable gender marking. This study examines Spanish dual-gendered nouns such as agua (“water”) and área (“area”), which are prescriptively feminine but, due to historical phonological processes, are sometimes preceded by (presumed to be) masculine forms. A corpus search on X (formerly Twitter) confirmed that Spanish speakers commonly alternate between masculine and feminine agreement in everyday use. Using comprehension- and production-based methods, we show that humans and OpenAI Whisper adapt to real-world distributional patterns over and above prescriptive norms, underscoring the importance of studying variable phenomena in naturalistic contexts.