Using AI: Gen-AI Tools in the Library

AI Insights are generated using a Large Language Model (LLM) prompted through a process known as Retrieval-Augmented Generation (RAG). This method uses the content as grounding to reduce hallucinations and ensure relevance. It does not incorporate user behavior, institutional data, or private content into the model. Additionally, EBSCO applies a Human-in-the-Loop (HITL) process by reviewing a sample of summaries with Subject Matter Experts (SMEs) to ensure accuracy, tone, bias mitigation, and timeliness.

Key benefits include:

• Helping users quickly assess relevance of scholarly articles.
• Supporting users with varying levels of research experience, especially those who struggle with interpreting abstracts.
• Teaching AI literacy and critical reading skills by comparing AI summaries with full articles or abstracts.

⚠️ Insights only display when a full-text article is present and of sufficient length.

⚠️ Although grounded in the source text, the AI summary may miss nuance or context in complex or technical material.

⚠️ Insights are regenerated each time a user clicks the button, so phrasing may differ across sessions.

NLS uses Natural Language Understanding (NLU), a form of artificial intelligence that focuses on interpreting user input rather than generating new content. This system uses large language models that have been designed to better understand and process your academic search questions. It does not use generative AI and is not trained on user data. It breaks your search into important terms and topics, then runs it through EBSCO's search system to find the most relevant and useful results instead of just matching words.

Key benefits include:

• Allowing users to search using everyday language instead of complicated search terms.
• Helping users feel more confident starting their research.
• Supporting interdisciplinary research or unfamiliar topics.

⚠️ Accuracy may vary depending on how clearly the query is written. [See prompt engineering]

⚠️ Some results may still require refinement using advanced filters and traditional search techniques.

⚠️ Not all records will show significant improvements, especially those with limited metadata.

The Ebook Central Research Assistant is powered by OpenAI’s GPT-4o mini using a Retrieval-Augmented Generation (RAG) approach. This means all AI-generated responses are grounded in the specific book chapter text, not the entire book or external sources. The model is not trained on user interactions or licensed ebook content, ensuring privacy and compliance with publisher agreements.

Key benefits include:

• Quickly evaluating if a book chapter is relevant to the user's topic.
• Breaking down key concepts to make materials easier to understand.
• Encouraging deeper reading by offering a starting point to research.

❌ It only analyzes individual chapters, not full books.

❌ Short chapters or those with mostly image-based content (like tables or graphs) may not produce results.

❌ Currently, citations for AI-generated responses are not available.

The research assistant uses OpenAI's GPT-4o mini large language model, implemented with Retrieval-Augmented Generation (RAG) to ensure all AI outputs are based only on the content of the document being viewed.

Key benefits include:

• Quickly assessing relevance of dissertations through concise takeaways and key concepts.
• Discovering new topics with AI-suggested research themes and search prompts.
• Reducing barriers for students who may struggle with traditional search tools.

❌ Unable to analyze images, tables, or graphs.

❌ Short chapters or documents may be excluded due to minimum length requirements for AI processing.

❌ Users cannot cite AI-generated content directly, though features for this are in development.

⚠️ Though hallucinations are minimized by grounding answers in the text, users are encouraged to fact-check.