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

This study analyzes longitudinal data from Google Trends and Wikipedia to demonstrate a significant global decline in traditional search and page-view activity for physics topics, suggesting a shift in academic information-seeking behavior toward generative AI tools, particularly in non-English-speaking regions.

The Gist

The "Google vs. ChatGPT" Shift: A New Way of Learning Physics

Imagine you are a student stuck on a difficult math problem. For the last twenty years, your go-to move has been to walk into a massive, infinite library (the Internet), pull a specific book off a shelf (a Google search), and read through several chapters to find the answer yourself. This is "Search-based learning." It’s like being a detective: you have to find the clues, verify them, and piece the story together.

But recently, a new character has entered the library: a brilliant, incredibly fast, and very chatty assistant (Generative AI, like ChatGPT). Instead of you hunting for the book, you just tap the assistant on the shoulder and say, "Hey, explain this to me," and they hand you a perfectly written summary. This is "Generation-based learning."

This research paper tracks exactly how this shift is happening globally, using physics students as the "canary in the coal mine."

---

1. The "Digital Pulse" is Slowing Down

The researchers looked at "Google Trends"—essentially the heartbeat of what the world is curious about. They noticed that the "heartbeat" of physics searches is getting much weaker.

Think of it like a fading echo. In 2022, when students were searching for things like "Kinetic Energy," the echo was loud and frequent. By 2025, that echo has significantly quieted down. The researchers believe this isn't because students are suddenly less interested in physics, but because they’ve stopped "searching" and started "asking." They aren't looking for the book anymore; they’re talking to the assistant.

2. Mechanics vs. Electromagnetism: The "Words vs. Pictures" Divide

The study found something fascinating: the decline isn't the same for every topic.

• Mechanics (The "Words" Domain): Concepts like "Force" or "Newton’s Laws" are very easy to describe with words. Because AI is a master of language, it is incredibly good at explaining these. Consequently, searches for Mechanics have plummeted. It’s like the AI has become a world-class tutor for anything you can describe in a sentence.
• Electromagnetism (The "Pictures" Domain): Concepts like "Electric Fields" or "Magnetic Flux" are much more abstract. They rely heavily on complex diagrams, invisible lines, and visual mental models. Current AI is still a bit "clumsy" at explaining things that require deep visual reasoning. Because the AI isn't a perfect "artist" yet, students are still turning to Google and Wikipedia to find the diagrams they need.

3. The "Language Tax" is Being Cancelled

One of the most profound findings is how this affects different parts of the world.

In English-speaking countries (like the US and UK), the shift is happening, but it's relatively gradual. However, in non-English speaking regions (like India, Brazil, or South Korea), the decline in traditional searching is massive.

Think of it as a "Language Tax." For a long time, if you wanted to learn high-level physics, you often had to navigate a world of information written primarily in English. This was a "tax" on your time and mental energy. GenAI acts like a universal translator and local guide. It can take complex scientific concepts and explain them fluently in your native language. This is effectively "canceling the tax," making high-level knowledge much more accessible to everyone, regardless of what language they speak.

4. What does this mean for the future?

The researchers aren't saying AI is "bad," but they are warning that the skills we need are changing.

In the old days, being a good student meant being a great "Searcher" (knowing how to find the right book). In the new era, being a good student means being a great "Editor" (knowing how to check if the AI is lying to you).

If the AI hands you a beautiful, confident explanation of a physics law, you can't just nod and say "thanks." You have to be able to look at that explanation and say, "Wait a minute, that part doesn't actually make sense according to the laws of gravity."

The takeaway: We are moving from an era of finding answers to an era of evaluating answers.

Technical Summary

Technical Summary: From Search to GenAI Queries

Title: From Search to GenAI Queries: Global Trends in Physics Information-Seeking Across Topics and Regions
Authors: Yossi Ben-Zion, Omer Michaeli, and Noah D. Finkelstein
Date: February 11, 2026

---

1. Problem Statement

The emergence of Generative Artificial Intelligence (GenAI) represents a potential paradigm shift in academic information-seeking. Traditionally, students have relied on search engines (e.g., Google) to find information, a process requiring query formulation, source evaluation, and synthesis. GenAI shifts this model from finding information to generating answers.

The researchers aim to quantify this shift by investigating whether the decline in traditional search engine usage serves as a proxy for the adoption of GenAI. Specifically, they seek to understand:

• How longitudinal search trends and seasonal periodicity change post-GenAI.
• If the impact differs between physics domains (Mechanics vs. Electromagnetism).
• How these behavioral shifts vary across different geographic and linguistic regions.

2. Methodology

The study employs a large-scale, longitudinal, and cross-platform approach:

• Data Sources:
  • Google Trends: Used to measure Relative Search Volume (RSV) for core physics concepts. To ensure academic relevance, queries were filtered by the "Science" category and utilized "Topic" entities (semantic concepts) rather than literal strings to ensure linguistic invariance.
  • Wikipedia Page-Views: Used as an independent validation metric across seven major languages to ensure findings were not platform-specific to Google.
  • GenAI Interest: Search volume for "ChatGPT" (filtered by "Science") was analyzed to correlate the decline in physics searches with the rise of AI interest.
• Temporal Scope: Three academic cycles from August 2022 to August 2025, categorized into:
  1. Reference Year ('23): Pre-widespread GenAI adoption.
  2. Transition Year ('24): Rapid diffusion of LLMs.
  3. Integration Year ('25): Established GenAI usage.
• Taxonomy: Topics were curated from seminal physics textbooks, focusing on Classical Mechanics (e.g., Kinetic Energy, Newton’s Laws) and Electromagnetism (e.g., Electric Charge, Electric Field).
• Statistical Protocol:
  • Paired-sample t-tests were used to compare annual means, controlling for academic seasonality.
  • Cohen’s $d$ was used to measure effect size (practical significance).
  • Cross-regional normalization was applied to allow comparisons across different populations and internet penetration levels.

3. Key Results

• Global Decline in Traditional Search: There is a substantial, systematic, and persistent decline in search and Wikipedia page-view activity for most physics topics.
• Domain Divergence:
  • Mechanics exhibited sharper and more consistent declines (large effect sizes, $d > 0.8$).
  • Electromagnetism showed more stability, with some topics even showing slight increases. The authors hypothesize this is because Electromagnetism relies heavily on visual/diagrammatic reasoning, a domain where current LLMs are less proficient than in the verbal/causal reasoning required for Mechanics.
• Geographic and Linguistic Heterogeneity:
  • English-speaking regions (USA, UK, Australia) showed relative stability or only moderate declines.
  • Non-English speaking regions (India, Brazil, South Korea, etc.) exhibited massive reductions in search volume (often $>50\%$).
  • GenAI Correlation: Regions with the highest decline in traditional search (e.g., India, Brazil) also showed the highest relative engagement with "ChatGPT" in scientific contexts.
• The "Language Tax" Mitigation: The data suggests GenAI may be acting as a linguistic equalizer. Non-native English speakers are adopting GenAI more aggressively to bypass the barriers of searching for information in English, effectively reducing the "language tax" in knowledge consumption.
• Validation via Outliers: The study cites Iran as a natural counterfactual; because GenAI is structurally limited there while Google remains dominant, search volumes remained stable, supporting the causal link between GenAI availability and search decline.

4. Significance and Implications

• Epistemic Shift: The study identifies a fundamental change in the skills required for learning. The burden is shifting from information retrieval (finding and synthesizing sources) to information evaluation (critically assessing the accuracy and coherence of generated answers).
• Pedagogical Reorientation: Physics educators must move away from tasks that merely require "finding an answer" and toward tasks that require meta-cognitive evaluation. This includes:
  • Analyzing AI-generated explanations for omissions or misconceptions.
  • Comparing AI responses against formal mathematical or visual representations.
  • Using GenAI as a tool for "prompt-based" inquiry rather than a "black box" answer machine.
• Educational Equity: While GenAI may bridge linguistic gaps, it introduces new risks regarding the depth of understanding (superficial vs. robust) and the potential for "hallucinated" physics principles.