SEARCH AI is a tool in development at SBU Libraries designed to help users explore our catalog using natural language queries. Available as a toggle within our catalog webpage, SEARCH AI works by using generative AI to expand upon the topic of interest, dynamically select appropriate filters, and bring the user directly to the search results in our catalog. Currently, SEARCH AI includes the following functionality.
Why SEARCH AI
Getting relevant results from traditional library searches often requires familiarity with Boolean operators and how the catalog works, which can be challenging for many users. We want researchers to be able to find relevant materials to their studies without having to think about the syntax of the search query. Natural language translation to a proper catalog search removes the extra steps needed for individuals to find what it is they are looking for, enhancing discoverability of all our materials. We think SEARCH AI can make our library catalog more approachable, while making the researchers’ work more efficient.
Key features
- Semantic search / natural language queries – Ask questions in plain language.
- Boolean string creation – Generates a set of related keywords for the topic.
- Date range selection – Limit results by creation or publication date.
- Material type filtering – Focus the search on books, articles, dissertations, and more.
- Facet controls – Easily filter by peer-reviewed content, online availability, or items held by SBU Libraries.
What is being worked on
We’re currently iterating on the functionality for the expansion of Boolean searches and the appropriate selection of filters based on natural language input. Testing by library staff and faculty began in August of 2025, with plans for a phased rollout of successive versions of this tool to community members as we progress. The tool is currently only accessible via a closed testing environment, with live implementation to be scheduled.
The roadmap for SEARCH AI is actively being developed internally. There is currently no release date scheduled, but the team is eager to share this tool with the SBU community as soon as it is ready to be implemented on our website.
Large Language Models (LLMs) and Data Privacy
At the moment, SEARCH AI uses OpenAI large language models to complete searches. The OpenAI model gpt-4o-mini is used as a baseline model and fine-tuned gpt-4o models are used as agents. Personal information is not included with user queries. Data transmitted to the OpenAI API is not used by OpenAI for training.
Stony Brook Libraries maintains full control of SEARCH AI data and none of the data is used by external parties. In the future, there is potential for SEARCH AI and our projects to run on local AI servers.
Possible Limitations
SEARCH AI uses artificial intelligence to generate responses and assist with information retrieval, but it has important limitations. AI may produce errors, outdated details, or incomplete information. SEARCH AI has these same limitations. Because of this, responses should be treated as a starting point rather than a final authority. Additionally, AI is not appropriate for processing sensitive, private, or personally identifiable information, and its role is to support and not replace human expertise.
