Oxford Academic’s AI Discovery Assistant

by Oxford University Press
Mar 17, 2025 | Filed in Academic Libraries

Included with any subscription to or purchase of content on Oxford Academic, our AI Discovery Assistant helps to make your discovery of scholarship quicker, whatever your specialism, as it identifies the ten pieces of content most relevant to your research. Think of it as a more intuitive search, where you no longer need to have exact matches on words or terms.

How to use the tool

Oxford Academic’s AI Discovery Assistant is easy for anyone to use, regardless of experience. To input a prompt, you must have access to content on Oxford Academic, so make sure you’re signed in. Once you’re signed in, input your prompt:

The assistant will provide ten pieces of content from across Oxford Academic, with the most relevant content right at the top.
Each piece of content is accompanied with an AI-generated two-line synopsis wherever an original abstract is available.
You can ask questions to get more detail, or to clarify your intent.
Then simply click on one of the ten results to go to the content on Oxford Academic that you’re interested in to read the full text. You can use the availability markers, just like in our traditional search, to see which content you have access to.
Start again by clicking “New chat” and input your next prompt.

Please be aware that we do not keep a conversation history. If this is needed for your future reference, please copy and paste information into your digital notebook or word processing application.

If you don’t have access to any content on Oxford Academic please consider recommending it to your librarian or organisation’s purchaser directly by filling out our form.

Sign up to a webinar to find out more about the tool and how to use it.

Browse our frequently asked questions

General information

Why does the tool say “beta”?
Oxford Academic’s AI Discovery Assistant has launched as a beta. While we strive for accuracy, even state-of-the-art language models can make errors. We are continuously improving the assistant as more advancements happen in the generative AI space.

How does the AI Discovery Assistant work?
Oxford Academic’s AI Discovery Assistant is built on ChatGPT’s state-of-the-art, lighter weight ChatGPT 4o-mini large language model (LLM). We’ve implemented a retrieval-augmented generation (RAG) system that includes the publicly available metadata on our trusted content, which includes every title, abstract, keyword, contributor, and publication date.

Whenever we ingest content into our RAG system, we have the LLM generate a two-line synopsis based on the original abstract when that is available. This is to provide a quick, consistent overview of the content with a standard length, and help you decide what you’re most interested in. Please note, these synopses aren’t dynamically generated based on your prompt.

Where/when will the AI Discovery Assistant option show up?
You can find our AI Discovery Assistant at the top-right of your screen wherever you are on Oxford Academic, next to the traditional search box. To input a prompt you must have access to at least a piece of content on Oxford Academic, so make sure you’re signed in.

What content does the AI Discovery Assistant refer to? Can the AI Discovery Assistant be used to search the web?
We’ve implemented a retrieval-augmented generation (RAG) system for Oxford Academic’s AI Discovery Assistant that includes the publicly available metadata on our trusted content, which includes every title, abstract, keyword, contributor, and publication date. The list of ten results it provides includes availability markers, just like in our traditional search, so that you can quickly find the content you have access to. The AI Discovery Assistant does not pull from the wider web.

How does the AI Discovery Assistant determine what’s the most relevant content for my prompt?
When you type a prompt or query into Oxford Academic’s AI Discovery Assistant, the Assistant takes this and finds matching information from our platform to help you get the information you need.

Understanding your query: the chatbot uses a powerful language model to turn your prompt into a special format called "embeddings." Think of embeddings as capturing the essence of your query in a mathematical form.
Searching for matches: these embeddings are then compared against all the publicly available metadata about content on Oxford Academic.
Finding the best matches: the chatbot measures how similar your query is to each piece of content. It calculates similarity scores to determine which pieces of content are the most relevant to your prompt.
Returning results: finally, the chatbot returns the ten pieces of content with the highest similarity scores. These are the most relevant results based on your query.

What makes a good prompt for the AI Discovery Assistant?
We recommend using Oxford Academic’s AI Discovery Assistant the way you might use any search engine, using specific keywords or prompts depending on your needs.

Your prompt can be as short as “hypertension” which will search for “hypertension” and potentially other variations like articles on high blood pressure.

You can improve the specificity in your prompt by typing “chapters on hypertension” or “articles on hypertension” which will pre-filter the ten results to only come from either book chapters or journal articles. The AI Discovery Assistant can understand and filter by:

Book
Chapter
Journal
Article
DOI
ISBN
Publication date

Prompts like “beta blocker impact on hypertension” and “good introduction on hypertension” will help the tool drill down into the specific areas you are interested in.

What are you doing in response to the high energy demands of AI?
We recognize there are concerns about AI’s energy demands within the technology industry, and it’s something we continue to monitor. That includes working with our suppliers and partners to understand what they are doing to minimize their environmental impact.

Oxford University Press’s commitment to achieve net zero by 2050 encompasses tackling the footprint of our digital products. We will continue to evolve our strategy to ensure we are addressing the areas that have the biggest impact on our operations.

Development and Capabilities

How was the AI Discovery Assistant developed?
Oxford Academic uses a powerful Boolean-based search engine (Solr), giving you the opportunity to deep-dive into the content on Oxford Academic. The different filters that can be applied enable you to refine your search and find the information most relevant to you.

With advancements in generative AI, we wanted to see what would be possible to give a new option for searching. One that allows for natural language queries and doesn’t rely on exact term matching.

With editorial and technical experts within our organization, we developed a retrieval-augmented generation (RAG) AI pipeline. For the content set, we included publicly available metadata on our published content: the titles, contributors, publishing dates, content formats, etc.

Our testing focused on improving the retrieval of relevant Oxford Academic content through:

experimentation with the combination of metadata and content elements made available to the tool
“Top-k” adjustments: fine-tuning the score for retrieved content and content used in response generation. This is presently set to k=10, which means that 10 suggested reading items will be provided in each response.
“Temperature” reduction: making responses less creative and more deterministic for consistency
Prompt engineering: ensuring appropriate language and consistent formatting of responses

How does the AI Discovery Assistant ensure the accuracy of the information provided?
Oxford Academic’s AI Discovery Assistant ensures reliability by being grounded in our trusted content on Oxford Academic. The list of recommended content that it returns is listed in reverse order of relevancy. You can then ask follow-up questions to get more options or clarify your intent.

Can the AI Discovery Assistant hallucinate?
There is always some risk of hallucination (defined as when a chatbot generates untrue information in its response). However, because Oxford Academic’s AI Discovery Assistant is designed to provide ten search results that closely match your query, it is unlikely that you will encounter hallucinations. Any results are returned based on the LLM’s training data on semantic relevancy.

Can I input prompts into the AI Discovery Assistant in languages other than English?
Yes, Oxford Academic’s AI Discovery Assistant can respond to prompts written in languages other than English. However, it is optimized for English, and the content it references are primarily written in English.

Can I upload documents?
No, Oxford Academic’s AI Discovery Assistant only allows text-based input in the prompt box. It is not designed to analyse or help prepare any longform documentation.

Can I access previous conversations?
No, we do not keep conversation history within Oxford Academic’s AI Discovery Assistant. If this is needed for your future reference, please copy and paste information into your digital notebook or word processing application.

Compatibility

What are the system requirements for using the AI Discovery Assistant?
Oxford Academic’s AI Discovery Assistant supports the latest versions of major web browsers, including Microsoft Edge, Mozilla Firefox, Apple Safari, and Google Chrome. It is responsively designed for a variety of screen sizes including smartphones. For more detailed information, please visit our accessibility page.

Security and Privacy

This information relates to the Oxford Academic AI Discovery Assistant. View Oxford University Press’s Privacy Policy.

How does the AI Discovery Assistant protect confidentiality and data privacy?
None of your inputs to Oxford Academic’s AI Discovery Assistant are ingested into the training model of the assistant. There is no way for another user, your school, your employer, or Oxford University Press to see a recording of your conversation. The assistant is a safe and secure place for your research.