Impact of AI-based Discovery Tools on Academic Reading Habits: An Empirical Study

Thesis Abstract

The rapid expansion of AI-powered discovery tools within academic libraries has reshaped the pathways through which students access, evaluate, and engage with scholarly literature, raising concerns about potential shifts in reading habits, comprehension strategies, and overall research quality. This study investigates how AI-based discovery tools influence academic reading behaviors among graduate students, addressing a gap in empirical evidence on the mechanisms by which algorithmic personalization, relevance ranking, and serendipity features affect time allocation, depth of reading, and critical engagement with sources. The aim is to examine whether AI-enhanced discovery tools foster more efficient information-seeking, promote broader exposure to interdisciplinary materials, or conversely, contribute to narrowing of reading breadth and confirmation-oriented browsing. The specific objectives are (1) to quantify changes in reading time, number of sources consulted, and proportion of primary versus secondary sources used before and after exposure to AI-based discovery tools; (2) to assess students’ perceived depth of reading, critical appraisal practices, and citation quality when utilizing AI-driven recommendations; (3) to explore how user characteristics (discipline, prior AI familiarity, and epistemological orientation) moderate the impact of AI tools on reading habits; (4) to identify coping strategies and behavioral adaptations employed by students to maintain rigorous scholarly engagement; and (5) to develop a conceptual model linking AI-tool features to reading outcomes grounded in the Uses and Gratifications Theory and the Technology Acceptance Model. The study adopts a mixed-methods sequential explanatory design. The population comprises graduate students enrolled in research-focused MSc and PhD programs across five faculties at a major research university. A stratified random sample of 320 students will be selected, with approximately 200 completing a structured survey and 60 participating in in-depth interviews. Data collection instruments include a validated questionnaire measuring reading behavior metrics (time on task, breadth and depth of sources, and critical appraisal indicators) and perceived usefulness and ease of use of AI tools, supplemented by log data from library discovery systems over a 12-week period. Semi-structured interviews will elicit nuanced accounts of decision-making processes, perceived limitations of AI recommendations, and strategies to ensure scholarly rigor. For data analysis, quantitative data will be analyzed using descriptive statistics, paired-sample t-tests to detect pre-post changes, multiple regression to identify predictors of reading depth, and multigroup structural equation modeling to test the proposed theoretical relationships. Qualitative data will undergo thematic analysis guided by Braun and Clarke’s approach, with coding validated through intercoder reliability and triangulated against survey results. The study will address validity through instrument triangulation, pilot testing, and member-checking in interviews, and reliability via Cronbach’s alpha on scale measures and test-retest reliability for reading-behavior items. Anticipated findings include (a) AI-based discovery tools will be associated with increased efficiency in locating relevant materials, but may reduce incidental exposure to non-target topics; (b) depth of reading correlates positively with perceived tool usefulness and perceived trust in algorithmic recommendations; (c) discipline-specific effects emerge, with methodological fields exhibiting greater reliance on primary sources while humanities and social sciences leverage AI suggestions for broader interdisciplinary exploration; (d) users with higher AI familiarity demonstrate more sophisticated critical appraisal and greater adaptation of search strategies. The study contributes to knowledge by empirically linking AI-assisted discovery features to concrete reading outcomes and by proposing a validated model of reading behavior in AI-augmented environments, informing library practice, tool design, and information literacy curricula. Policy and practice implications include guidance for configuring discovery tools to balance relevance with serendipity, developing teaching interventions that cultivate critical evaluation of AI-generated results, and fostering user training that sustains comprehensive and rigorous reading habits. The main conclusion is that AI-based discovery tools hold the potential to enhance research efficiency and scope if accompanied by deliberate design features and targeted information literacy support; otherwise, they risk narrowing exposure and diminishing critical engagement. Recommendations include (1) embedding transparency about ranking criteria and source provenance in AI interfaces; (2) integrating reflective reading prompts and metacognitive training into library instruction; and (3) conducting ongoing, institution-wide evaluations of reading outcomes to monitor unintended biases and preserve academic rigor.