Generative AI Use and Reading Comprehension: Associations with Metacognitive Experience in Higher Education—A Quasi-Experimental Study

(1) Background: The integration of generative artificial intelligence (AI) into academic reading raises the question of whether such tools improve genuine comprehension or merely alter readers’ perception of it. (2) Objective: This study examined the association between generative AI use and reading performance and cognitive experience among university students reading a conceptually dense regulatory text. (3) Method: In a quasi-experimental, post-test-only design, with a single cross-sectional measurement per participant, 531 students from two Spanish higher education institutions read a Spanish-language regulatory text and were assigned to six conditions combining two systems (Microsoft Copilot Chat and ChatGPT 4 and 4.5) with varying pedagogical mediation, from unassisted reading to guided use with prior training. Literal comprehension and conceptual interpretation scores were assessed alongside perceptual measures of interest, attention, perceived learning and explanatory confidence. (4) Results: Literal comprehension did not differ across conditions (F(5, 525) = 0.85, p = 0.518, η2 = 0.008), whereas conceptual interpretation did (F(5, 525) = 3.55, p = 0.004, η2 = 0.033), although no AI condition outperformed unassisted reading. The clearest effect emerged for explanatory confidence (F(5, 525) = 12.58, p < 0.001, η2 = 0.107), followed by attention (p < 0.001, η2 = 0.027). Prior AI training was associated with the lowest interpretation and confidence scores. (5) Conclusions: Generative AI use was associated with changes in the subjective experience of reading, particularly explanatory confidence, more consistently than in actual performance. The observed pattern suggests that how students interact with generative AI may warrant at least as much attention as mere tool availability, although the present design does not permit this effect to be isolated from institutional and phase-related confounders.