Towards an AI-Augmented Textbook (2509.13348v2)

Abstract: Textbooks are a cornerstone of education, but they have a fundamental limitation: they are a one-size-fits-all medium. Any new material or alternative representation requires arduous human effort, so that textbooks cannot be adapted in a scalable manner. We present an approach for transforming and augmenting textbooks using generative AI, adding layers of multiple representations and personalization while maintaining content integrity and quality. We refer to the system built with this approach as Learn Your Way. We report pedagogical evaluations of the different transformations and augmentations, and present the results of a a randomized control trial, highlighting the advantages of learning with Learn Your Way over regular textbook usage.

• The paper demonstrates that leveraging generative AI and LLMs for content personalization significantly improves both immediate comprehension and retention.
• The Learn Your Way system uses a two-stage pipeline to tailor curriculum content by grade level and individual interests, generating multi-modal outputs such as slides, mind maps, and immersive texts.
• A randomized controlled trial confirmed that AI-augmented textbooks offer statistically significant gains in engagement, learning, and satisfaction compared to standard digital readers.

AI-Augmented Textbooks: Personalization and Multi-Modal Transformation with Learn Your Way

Introduction

The paper "Towards an AI-Augmented Textbook" (2509.13348) presents a comprehensive framework for transforming static, one-size-fits-all textbooks into dynamic, personalized, and multi-modal learning experiences using generative AI. The proposed system, Learn Your Way, leverages LLMs and generative image models to adapt high-quality, curriculum-aligned source material to individual learner attributes, including grade level and personal interests. The system further augments the content with multiple representations and embedded formative assessments, aiming to enhance engagement, comprehension, and retention.

System Architecture and Content Transformation

Learn Your Way operates via a two-stage generative pipeline:

1. Text Personalization: The source material is rewritten to match the learner's grade level (using Flesch-Kincaid Grade metrics) and personal interests. The system selectively rewrites segments amenable to personalization, maintaining factual integrity and curriculum alignment. This approach is supported by learning science literature on prior knowledge activation and conceptual mapping.
2. Content Transformations: The personalized text serves as the basis for generating multiple representations:
   • Slides and Narrated Slides: Class-like slide decks with generative narration, adapted to the learner's interests.
   • Audio-Graphic Lessons: Simulated teacher-student dialogues, generated via independent LLM personas, paired with dynamic graphical concept maps.
   • Mind Maps: Hierarchical, expandable visualizations of the material, annotated with personalized text and images.
   • Immersive Text: The centerpiece view, interleaving personalized examples, embedded questions, timelines, mnemonics, and visual illustrations.

     Figure 1: The Immersive Text view, showing 6th-grade level content on immune system disruptions, adapted for a learner interested in gaming, with generative add-ons and modality switching.

(Figure 2)

Figure 2: Two-step personalization and expansion: Newton's Third Law example rewritten for 'basketball' and 'art', then transformed into multiple formats.

(Figure 3)

Figure 3: Generated slide for economic systems, personalized to 'soccer', illustrating market concepts with domain-specific examples.

(Figure 4)

Figure 4: Mind map for early human evolution, with expandable nodes and annotated visuals for granular exploration.

(Figure 5)

Figure 5: Personalized visual illustration for economic systems, adapted to 'soccer', with both text and image tailored to the learner's interest.

Embedded Practice and Assessment

Learn Your Way integrates formative assessment at multiple levels:

• Embedded Questions: Dynamically generated, context-grounded multiple-choice questions, accessible within the Immersive Text.
• Quizzes: Section-level assessments with targeted feedback, including "Glows" (strengths) and "Grows" (areas for improvement), aligned with Bloom's taxonomy.

(Figure 6)

Figure 6: Example of an embedded question for economic systems, generated and grounded in the personalized material.

Pedagogical Evaluation

A panel of pedagogical experts evaluated Learn Your Way's components using a rubric encompassing accuracy, coverage, emphasis, engagement, cognitive load, active learning, metacognition, motivation, adaptability, and clarity of learning intentions. Ten OpenStax textbook chapters were used as source material, with three grade levels and three personal interests, resulting in 30 unique configurations.

(Figure 7)

Figure 7: Expert ratings for Learn Your Way components, showing high scores across pedagogical dimensions, with visual illustration scoring lowest due to current generative model limitations.

Key findings include:

• All components achieved average ratings above 0.90 across axes.
• Slides alone scored lowest on engagement, but narrated slides significantly improved engagement.
• Visual illustrations lagged due to generative image model limitations in producing pedagogically effective diagrams.

Efficacy Study: Randomized Controlled Trial

A randomized controlled trial (RCT) was conducted with 60 students (ages 15-18), comparing Learn Your Way to a standard digital reader (Adobe Acrobat Reader) using an unfamiliar textbook chapter on adolescent brain development. Pre-paper reading comprehension and topic familiarity were controlled.

• Immediate Assessment: 15-minute test post-learning session.
• Retention Assessment: 5-10 minute test three days later.

(Figure 8)

Figure 8: Experimental protocol for the RCT, detailing participant flow and assessment timeline.

Results:

• Immediate Assessment: Learn Your Way cohort scored significantly higher ($p = 0.03$).
• Retention Assessment: Learn Your Way cohort maintained a significant advantage ($p = 0.03$).
• Learner Experience Survey: Learn Your Way was rated significantly higher across all measures of engagement, perceived learning, and satisfaction.

(Figure 9)

Figure 9: Immediate and retention assessment scores, showing statistically significant improvements for Learn Your Way over the digital reader.

(Figure 10)

Figure 10: Survey responses post-assessment, with Learn Your Way outperforming the digital reader on all learner experience metrics.

Implementation Considerations

• Model Selection: Gemini 2.5 Pro was used for text and persona generation; a fine-tuned image model was required for educational illustrations.
• Personalization Attributes: Current implementation focuses on grade level and interests; future work could incorporate additional learner attributes (e.g., language proficiency, learning gaps).
• Scalability: The generative pipeline enables rapid adaptation of high-quality, curriculum-aligned content without manual intervention, supporting large-scale deployment.
• Limitations: Visual illustration quality remains a bottleneck; further fine-tuning and dataset curation are needed. The RCT focused on a single chapter; broader topic coverage and component-level efficacy analysis are future directions.

Theoretical and Practical Implications

The Learn Your Way framework operationalizes key principles from learning science, including dual coding theory, self-regulated learning, and adaptive personalization. The integration of multi-modal representations and formative assessment aligns with evidence-based pedagogical strategies. The system demonstrates that generative AI can deliver personalized, engaging, and effective learning experiences at scale, with statistically significant improvements in both immediate and long-term learning outcomes.

Future Directions

Potential avenues for further research and development include:

• Expanding personalization to additional learner attributes and implicit signals.
• Enhancing adaptivity via real-time performance monitoring and dynamic content adjustment.
• Improving generative image models for educational diagram synthesis.
• Integrating Learn Your Way into existing learning management systems, providing teachers with analytics and control.
• Conducting component-level efficacy studies to isolate the impact of individual transformations.

Conclusion

"Towards an AI-Augmented Textbook" provides a robust, empirically validated framework for leveraging generative AI to transform static educational content into personalized, multi-modal, and assessment-rich learning experiences. The Learn Your Way system demonstrates significant pedagogical and efficacy advantages over traditional digital readers, substantiating the potential of AI-augmented textbooks to advance personalized learning at scale. Future work should focus on expanding personalization, improving generative modalities, and integrating adaptive feedback mechanisms to further optimize learning outcomes.