{"id":19014,"date":"2026-04-03T14:04:15","date_gmt":"2026-04-03T14:04:15","guid":{"rendered":"https:\/\/news.nnlm.gov\/region_7\/?p=19014"},"modified":"2026-04-03T14:05:37","modified_gmt":"2026-04-03T14:05:37","slug":"ai-tools-in-the-medical-library-notebooklm","status":"publish","type":"post","link":"https:\/\/news.nnlm.gov\/region_7\/2026\/04\/03\/ai-tools-in-the-medical-library-notebooklm\/","title":{"rendered":"AI Tools in the Medical Library: NotebookLM"},"content":{"rendered":"

This is part of a series of blog posts from hospital librarians using AI in their work. This post was s<\/em>ubmitted by Maureen Dunn<\/strong>, Director of Library Services at Concord Hospital (NH).\u00a0<\/em><\/p>\n

In an attempt to harness the good in AI tools rather than simply feel overwhelmed by them, I decided to try an experiment with one: Google\u2019s NotebookLM. Way back in the fall of 2024, I attended a fantastic class on AI tools taught by the technology librarian at the NH State Library, Bobbi Slosser, where she piqued my interest in NotebookLM by demonstrating a Notebook she had created using digitized town reports. I had been looking for a good use case to try it out with ever since.<\/p>\n

NotebookLM<\/a> is not a generative AI tool itself, but it uses Google\u2019s Gemini AI technology to answer questions based on the user\u2019s own uploaded content, consisting mainly of documents and web site URLs. The tool uses a retrieval-augmented generation (RAG) framework to create answers based ONLY on the content provided by the user, improving specificity and reducing hallucinations. All answers produced are referenced to the specific content from which they came.<\/p>\n

As co-chair of my hospital\u2019s Evidence-Based Practice Committee (EBPC), and a solo librarian, I am often charged with pulling together information for the committee to review on various clinical topics. I send the pertinent information (usually a handful of journal articles and a synopsis I\u2019ve created) out to committee members ahead of meetings so that they might have a chance to review it, but time and lives being what they are, this rarely happens as planned! Either I don\u2019t have time to create a useful synopsis, or the members don\u2019t even look at the email prior to the meeting.<\/p>\n

I end up putting together a brief presentation on most topics, so that everyone can be on the same page to start the meeting, which takes up even more of my time. EBPC recently took up the question of anal cancer screening (anal pap testing), about which there is a decent amount of evidence, but not widely disseminated knowledge, and I decided it was time to put AI to work for me.<\/p>\n

I started by doing a traditional literature search as I always do, but when I had identified a list of about 20 pertinent and good-quality journal articles, I created a new Notebook in NotebookLM, and uploaded the articles to it. I then asked for a brief synopsis of the articles, specifically requesting inclusion of items I knew the committee would want to know, such as the populations most at risk of anal cancer, the most effective screening tools and the potential impact on public health if screening were implemented. In just under a minute, I was presented with a beautifully written, thoroughly referenced synopsis that would have taken me at least an hour to pull together.<\/p>\n

I then used one of NotebookLM\u2019s Beta features to request a slide deck based on my uploaded content. That took longer \u2013 about 10 minutes \u2013 but the results were impressive, with figures and graphs I never could have created myself, and a logical presentation flow. If anything, they were probably too in-depth for what I needed for my meeting, but I had used the default setting, and I could easily have tailored my request to create a more appropriate output.<\/p>\n

The best part of using NotebookLM came in the actual meeting, mid-discussion, when someone asked what the difference in specificity and sensitivity was between the different available screening methods. I knew I had read that in a few different articles, and I had a stack of printed articles sitting in front of me. Rather than start flipping through them or just giving a general answer to the question (which option was considered the most effective), I queried my Notebook, and within 30 seconds I had a specific answer, and links to the articles it had used to find it. (And yes, I did double-check them for accuracy and <\/em>was transparent with the group about what I was doing \u2013 they were intrigued!)<\/p>\n

A few things to know before using NotebookLM for this or similar purposes:<\/p>\n