Library Services
As we are increasingly seeing the integration of GenAI capabilities into web search and academic databases, it is useful to know the difference between how traditional search tools and GenAI tools function. Knowing the difference helps in understanding their capabilities, what is being returned when using them for search, and why caution needs to be taken when using outputs.
To put it simply, search engines and databases take a query of user inputted keywords and match them to sources (websites or articles in a database) to direct the user to information that already exists.
More advanced web search engines (such as a traditional Google search not using the AI overview) use AI but this is in the form of algorithms to enhance its ability to retrieve relevant results. For example, it can spot spelling mistakes, search for words with similar meaning, personalise your results based on your search history, and look at your location to apply context to a search query. This is know as semantic search. It can be seen in action if you were to search for "coffee shop" - the results returned wouldn't just be websites with the words "coffee shop" in them. It would provide a list of coffee shops that are close to your location, links to websites and menus, an address, directions, opening hours and customer reviews. This information is pulled together form existing data or links can be easily followed to find further information.
Rather than retrieving existing information which the user then selects (go to a website or choose an article to read in a database) GenAI tools create a new content based on the prompt that has been provided. Initially the output from GenAI was based solely on the dataset it was trained on to predict an answer. This put limitations on their ability to provide accurate responses to questions that required up to date knowledge. The tools weren't understanding a prompt, searching the internet for information, retrieving relevant information and summarising it - they were just providing outputs based on the probability of one word appearing after another in a sentence based on what they had learnt from their training data.
To improve the tools' ability to create up to date answers, some GenAI tools now have Retrieval Augmented Generation (RAG). This allows the tools to pull information from external data sources rather than solely relying on the training data (this could be in the form of connecting to the internet to access real time data on websites for tools like ChatGPT, or can be seen in a Google search AI overview). RAG also allows the tool to pinpoint its sources (from documents or websites) and provide citations, so it is clear where the information has come from.
Some GenAI tools are also incorporating Agentic AI features as the tools are starting to work autonomously and achieve more complex tasks beyond the original user-initiated prompts. You may see this happening in tools like ChatGPT as they ask you questions to clarify your prompts or the option to provide you with further details or resources.
The important thing to be aware of is even if a GenAI tool is accessing the internet, it still uses its LLM capability to summarise the information and create an answer. It will use scans of websites and documents, plus the training dataset, to create an output. Some of the data may be pulled directly from a source and summarised, some of the data may be entirely generated. This can result in misleading information or hallucinations in outputs. This is why it is especially important to always check outputs from GenAI tools (this includes the ‘AI Overviews’ that sometimes surface at the top of Google search results pages) - don’t assume their outputs are correct. If a citation is provided, follow it to check the data matches the output from the GenAI tool. For further guidance on evaluating output see the Critical evaluation page.
With increasing numbers of GenAI tools available, especially ones specifically focused on research, it can be tempting to ignore the library databases. However, GenAI research tools are often limited to only open source or publicly available sources, so not using library databases could result in useful research being missed.
To get the best results when using traditional databases and search engines, it is best to use established search skills such as keyword searching, phrase searching, use of search strings and Boolean operators. You can also apply filters and search individual fields such as author, date, title, or subject. There is further guidance on traditional search skills on our Search tips page.
GenAI tools provide the best results when using a conversational style using natural language, asking questions and clarifying what you want. Think of your search as giving a series of instructions. If we think of the earlier example of searching for “coffee shop”, entering this as a search in ChatGPT would result in the tool needing to ask further clarifying questions: “Are you thinking about opening a coffee shop, looking for coffee shop recommendations, or something else related to coffee shops? Let me know how I can assist!”. A better approach would be to tell it that you are visiting a certain location and would like it to provide you with a list of coffee shops at that location.
There is further guidance on effective prompt writing on the Prompt engineering page.
We’ve mentioned some uses for general and research GenAI tools and discussed their use as search tools, but how else can they be used to help with search and study skills?
Remember that you need to critically evaluate the information the tools find and verify their sources.
Producing reference lists can be tedious … wouldn’t it be amazing if you told ChatGPT to generate a list of references in your chosen style?
Although tools like ChatGPT can create a convincing looking reference list output, remember, GenAI tools are models using mathematical algorithms to predict the likelihood of words appearing next to each other. It won’t have access to all the details of your sources especially if they are from subscription databases. It may also combine several of your listed sources together to summarise them which will likely result in it producing fake references.