From Search to GenAI Queries: Global Trends in Physics Information-Seeking Across Topics and Regions

TL;DR

This paper documents a global shift in physics information-seeking from traditional search to generative AI-enabled access by analyzing longitudinal RSV data from Google Trends and corroborating with Wikipedia page views across multiple languages. Using topic-entity and Science-category filters, it defines robust cross-regional RSV metrics and compares three academic years to capture pre- and post-GenAI dynamics. The findings show substantial, domain- and region-dependent declines in search activity—Mechanics more so than Electromagnetism—with pronounced non-English-language declines and preserved seasonal structure, suggesting a redistribution of information access toward GenAI tools. The work discusses implications for teaching, advocating task designs that foster critical evaluation and interpretation of generated explanations, while highlighting limitations and the need for further context-specific research.

Abstract

The emergence of generative artificial intelligence (GenAI) marks a potential inflection point in the way academic information is accessed, raising fundamental questions about the evolving role of search in student learning. This study examines this shift by analyzing longitudinal trends in physics-related search and page-view activity, using declines in traditional search behavior as a quantitative proxy for changes in independent information-seeking practices. We analyze Google Trends data for core concepts in Classical Mechanics and Electromagnetism across three academic years (2022-2025) in more than 20 countries, and complement this analysis with Wikipedia page-view data across seven major languages to establish platform independence. The results reveal a substantial, systematic, and persistent global decline in search and page-view activity across most examined physics topics. The magnitude of this decline is domain-dependent, with Mechanics-related content exhibiting sharper and more consistent reductions than Electromagnetism-related content. Pronounced geographic and linguistic heterogeneity is observed: while English-speaking regions show relative stability or only moderate declines, non-English-speaking regions exhibit substantially larger reductions in traditional, search-based information-seeking activity. Despite the overall decrease in volume, the seasonal structure characteristic of academic activity remains robust. Taken together, these findings indicate a redistribution of physics-related information-seeking behavior in academic contexts where generative tools are increasingly available.

Figures (8)

• Figure 1: Temporal distribution of search interest for the topic Kinetic energy in the United States over the 2024–2025 academic year, filtered within the Science category, with annotated markers corresponding to key academic calendar events. Numbered markers indicate: (1) Fall semester start, (2) Thanksgiving break, (3) Fall final examination period, (4) Winter break, (5) Spring semester start, (6) Spring final examination period, and (7) Summer break.
• Figure 2: Temporal distribution of search interest for the topic ChatGPT in the United States over the 2024–2025 academic year, filtered within the Science category, with annotated markers corresponding to key academic calendar events that provide contextual reference for potential seasonal patterns. Numbered markers indicate: (1) Fall semester start, (2) Thanksgiving break, (3) Fall final examination period, (4) Winter break, (5) Spring semester start, (6) Spring final examination period, and (7) Summer break.
• Figure 3: Distribution of cumulative percentage changes ($\Delta\%$) in search volume from 2023--2025 across regions and groups of physics-related search content. Box plots represent the median and interquartile ranges, while individual topics are shown as distinct points. A divergence is observed between Mechanics (blue) and Electromagnetism (pink), particularly in the United States, where Electromagnetism-related search content shows a positive median shift. For visualization purposes only, one extreme outlier (+220% in the worldwide electromagnetism dataset) was excluded to ensure an interpretable scale. Its exclusion does not affect the reported medians or the qualitative conclusions.
• Figure 4: Standardized effect sizes (Cohen's $d$) for physics-related search content across regions. Dashed horizontal lines indicate thresholds for small ($0.2$), medium ($0.5$), and large ($0.8$) effects. In the Worldwide and India datasets, the median effect size for Mechanics-related search content exceeds the threshold for a large effect ($d > 0.8$), whereas the United States displays substantially smaller effect sizes.
• Figure 5: Effect-size (Cohen's $d$) summary for Mechanics-related search content, comparing the 2022--2023 reference period with the 2024--2025 integration period across three regions: the United States, Worldwide, and India. Bars represent the corresponding Cohen's $d$ values by region. Dashed vertical lines mark the conventional thresholds for small ($d=0.2$), medium ($d=0.5$), and large ($d=0.8$) effects. Statistically significant differences ($p<0.05$) are highlighted in color (pink for the United States, blue for Worldwide, and yellow for India), whereas non-significant results ($p\ge 0.05$) are shown in gray.