AI-assisted search lets you search using everyday language.

OneSearch and other library databases normally require you to limit your search query to a few keywords that might be found in a resource’s title, abstract, or various other fields. That is because the database you are querying provides an index, and the search system retrieves that index’s items by matching the language in your search query. In some cases the system might supply synonyms so that, for example, if you search for heart attack, you also get results for cardiac.

AI-assisted search, by contrast, takes your search query and sends it to a large-language model (LLM). The LLM has been given a task: to transform your query into a query suitable for searching the index. This means that instead of you having to decide which keywords will match results, you can construct your query in a way that makes sense to you, using ordinary, conversational language, and the LLM will choose the keywords that will be used for the search. You can review what was actually used for the search by clicking generated query or refined query.

Is it like asking a chatbot to do my research?

This process is different from what happens when you ask an AI-powered chatbot to perform some research for you. When AI agents compile information in response to a query, they access stores of information that have a very different structure from an index, and make use not just of the words, but of the meanings those words are associated with. Library databases are not capable of this sort of activity.

How well does it work?

Your librarians have found that the performance of AI-assisted search is mixed. While it does a good job of finding appropriate synonyms, it ends up doing a very broad search; some parts of the query may be far broader in scope than what you sent to the LLM. In addition, while sometimes you can ask it to limit to certain formats (for example, "scholarly articles about…”) or time periods, in other cases it handles these limits poorly.