|
Field
Trial Title
|
Before
Gemini Notebook and Study Habits
|
After
AI Study Tool with Grounded Content and Study Habits
|
|
Field
Abstract
|
Before
This study evaluates whether access to Google Gemini Notebook, an AI-powered study tool available to ASU students, helps undergraduate students build more effective study habits and improve academic performance, and whether the tool is more effective when paired with personalized guidance about evidence-based study practices. The study is a three-arm randomized controlled trial among undergraduate students enrolled in targeted intermediate economics and business courses at Arizona State University. Consenting students complete a baseline survey after the first course exam and are randomly assigned to: (1) a control group receiving the standard course experience and neutral study-related contact; (2) a Gemini Notebook-only group receiving access-oriented guidance, including an introductory video and tutorials about using Gemini Notebook as an AI study partner; or (3) a Gemini Notebook plus personalized study guidance group receiving the Gemini Notebook introduction plus personalized feedback on their own study habits and guidance on using Gemini Notebook to support active learning practices such as self-testing, practice question generation, flashcards, and spaced practice. Outcomes combine survey measures of study beliefs, study effort, study time allocation, AI tool adoption, Gemini Notebook feature use and perceptions, with administrative outcomes including course grades, exam or course scores, Canvas activity, and student academic records. The central hypothesis is that AI study tools can improve learning partly by lowering the effort cost of high-return study practices such as retrieval practice and question generation.
|
After
This study evaluates whether access to a Grounded Large Language Model (LLM) Study Platform, an AI-powered study tool available to ASU students, helps undergraduate students build more effective study habits and improve academic performance, and whether the tool is more effective when paired with personalized guidance about evidence-based study practices. The study is a three-arm randomized controlled trial among undergraduate students enrolled in targeted intermediate economics and business courses at Arizona State University. Consenting students complete a baseline survey after the first course exam and are randomly assigned to: (1) a Control group receiving the standard course experience and neutral study-related contact; (2) a Grounded Large Language Model (LLM) Study Platform-only group receiving access-oriented guidance, including an introductory video and tutorials about using the AI study tool as an AI study partner; or (3) a Grounded Large Language Model (LLM) Study Platform plus personalized study guidance group receiving the same AI study tool introduction plus personalized feedback on their own study habits and guidance on using the AI study tool to support active learning practices such as self-testing, practice question generation, flashcards, and spaced practice. Outcomes combine survey measures of study beliefs, study effort, study time allocation, AI tool adoption, AI study tool feature use and perceptions, with administrative outcomes including course grades, exam or course scores, Canvas activity, and student academic records. The central hypothesis is that AI study tools can improve learning partly by lowering the effort cost of high-return study practices such as retrieval practice and question generation.
|
|
Field
Last Published
|
Before
September 28, 2026 12:20 PM
|
After
October 02, 2026 02:01 PM
|
|
Field
Intervention (Public)
|
Before
Three arms: (1) Control: standard course experience, neutral ASU-related video/activity, and balanced Canvas/email contact. (2) Gemini Notebook-only: introductory video and tutorials explaining how to create notebooks, add course materials, ask questions, check citations, and use Studio features such as summaries, study guides, flashcards, quizzes, mind maps, reports, and audio/video overviews, while remaining neutral about study strategy. (3) Gemini Notebook plus personalized study guidance: the same Gemini Notebook access/onboarding plus personalized feedback about the student's study habits, comparison with peers, evidence-based guidance encouraging retrieval practice over passive rereading, and concrete advice on using Gemini Notebook for self-testing, practice questions, quizzes, and flashcards. Students are not informed of their group assignment in order to preserve study integrity and reduce behavior that could undo randomization or contaminate the comparison across arms.
|
After
Three arms: (1) Control: standard course experience, neutral ASU-related video/activity, and balanced Canvas/email contact. (2) Grounded Large Language Model (LLM) Study Platform-only: introductory video and tutorials explaining how to create workspaces, add course materials, ask questions, check citations, and use features such as summaries, study guides, flashcards, quizzes, mind maps, reports, and audio/video overviews, while remaining neutral about study strategy. (3) Grounded Large Language Model (LLM) Study Platform plus personalized study guidance: the same AI study tool access/onboarding plus personalized feedback about the student's study habits, comparison with peers, evidence-based guidance encouraging retrieval practice over passive rereading, and concrete advice on using the AI study tool for self-testing, practice questions, quizzes, and flashcards. Students are not informed of their group assignment in order to preserve study integrity and reduce behavior that could undo randomization or contaminate the comparison across arms.
|
|
Field
Primary Outcomes (Explanation)
|
Before
The core estimands compare each treatment arm to control and compare Gemini Notebook plus personalized guidance to Gemini Notebook-only. Academic performance outcomes will be linked from instructor or administrative course records for consenting students. Study behavior outcomes will be measured in the endline survey and supplemented by usage logs where available and consented.
We will report intent-to-treat estimates. Because untreated classmates may adopt Gemini Notebook, we will also report estimates that use the usage data collected for all consented students in all arms to account for control-group adoption. These are (i) the treatment–control difference in adoption, (ii) instrumental-variable estimates of the effect of tool use for the tool-only arm relative to control, with random assignment as the instrument
|
After
The core estimands compare each treatment arm to control and compare Grounded Large Language Model (LLM) Study Platform plus personalized guidance to Grounded Large Language Model (LLM) Study Platform-only. Academic performance outcomes will be linked from instructor or administrative course records for consenting students. Study behavior outcomes will be measured in the endline survey and supplemented by usage logs where available and consented.
We will report intent-to-treat estimates. Because untreated classmates may adopt the Grounded Large Language Model (LLM) Study Platform, we will also report estimates that use the usage data collected for all consented students in all arms to account for control-group adoption. These are (i) the treatment–control difference in adoption, (ii) instrumental-variable estimates of the effect of tool use for the tool-only arm relative to control, with random assignment as the instrument
|
|
Field
Sample size (or number of clusters) by treatment arms
|
Before
Approximately one-third of consenting students per arm: about 667 control, 667 Gemini Notebook-only, and 667 Gemini Notebook plus personalized guidance, for a planned total of about 2,000.
|
After
Approximately one-third of consenting students per arm: about 667 control, 667 Grounded AI study tool, and 667 Grounded AI study tool plus personalized guidance, for a planned total of about 2,000.
|
|
Field
Secondary Outcomes (End Points)
|
Before
Beliefs about the effectiveness of study strategies; perceived costs and time needed to implement self-quizzing; expected exam performance; familiarity with and confidence using AI tools; Gemini Notebook adoption; number of weeks used; feature use including uploaded sources, quizzes, study guides, summaries, audio overviews, questions, and flashcards; perceived usefulness or harmfulness; trust in AI-generated outputs; intent to continue using AI-assisted study tools; spillovers to other courses; Canvas activity; demographic and academic covariates including prior academic preparation. In addition to this, the effectiveness gap between rereading and retrieval practices.
|
After
Beliefs about the effectiveness of study strategies; perceived costs and time needed to implement self-quizzing; expected exam performance; familiarity with and confidence using AI tools; Grounded Large Language Model (LLM) Study Platform adoption; number of weeks used; feature use including uploaded sources, quizzes, study guides, summaries, audio overviews, questions, and flashcards; perceived usefulness or harmfulness; trust in AI-generated outputs; intent to continue using AI-assisted study tools; spillovers to other courses; Canvas activity; demographic and academic covariates including prior academic preparation. In addition to this, the effectiveness gap between rereading and retrieval practices.
|
|
Field
Secondary Outcomes (Explanation)
|
Before
Secondary outcomes come from the baseline and endline surveys, Gemini Notebook usage data under the Google-ASU data-sharing process when available, Canvas activity, and ASU administrative records. These outcomes are intended to measure mechanisms: whether Gemini Notebook changes the cost of active studying, whether personalized guidance changes how students use the tool, and whether intensity or feature choice is associated with outcomes.
|
After
Secondary outcomes come from the baseline and endline surveys, Grounded Large Language Model (LLM) Study Platform usage data when available, Canvas activity, and ASU administrative records. These outcomes are intended to measure mechanisms: whether the Grounded Large Language Model (LLM) Study Platform changes the cost of active studying, whether personalized guidance changes how students use the tool, and whether intensity or feature choice is associated with outcomes.
|