Using AI with literature searches

MediaCentral Widget Placeholderhttps://mediacentral.ucl.ac.uk/Player/22BDiEIg

Watch on MediaCentral
Download the transcript (docx).
Only got a minute? Jump straight to Zahra and Alia's top tip

Introduction

The aim of this project was to develop a mutual understanding of how GenAI may be used for the development of systematic literature searches. This included building a picture of how our partner students approached GenAI for their systematic review projects (if they have used it at all), and how we, as educators, can support students going forward, and learn from them. 

Using ChatGPT to refine the research question

We started off by identifying a broad research topic and we used ChatGPT to refine it. We asked ChatGPT to break down the research question into PICO (Population, Intervention, Comparison and Outcomes). We prompted ChatGPT to identify synonyms and alternative terminology for each of the PICO concepts, which was useful to clarify the Comparison and Outcomes. 

Developing and comparing search strategies

We moved on to develop a systematic search strategy using bibliographic databases, e.g. Ovid Medline and Ovid Embase. We used this more traditional approach to searching as a benchmark against which to test ChatGPT’s ability to develop a comprehensive search strategy providing a series of prompts that we engineered and tested.  

Findings from using ChatGPT

The results of using ChatGPT for systematically searching revealed that: 

1. It helped us refine the question and identify more specific interventions by giving us an overview of the topic. 
2. ChatGPT was only partially helpful in providing a more comprehensive list of synonyms and alternative terminology and it was quite broad in providing advice about developing a comprehensive search strategy. 
3. ChatGPT could not create meaningful search strings for translating our initial search strategy across Embase and Web of Science databases. Since our study, Deep Research models in GenAI tools like ChatGPT, Gemini and Perplexity have emerged that are more promising in this area.   

Conclusions and next steps

As a result, our student-staff partnership enabled us to develop a deeper understanding of the conscientious use of GenAI tools for systematic reviews projects. 

UCL’s recommended GenAI tool is Microsoft Copilot – follow the link to find out more about the contract UCL has with Microsoft to protect data privacy.  If you’re using different tools, make sure you’re aware of how they use the data you upload. UCL’s GenAI Hub has more information. 

Although ChatGPT was the GenAI tool we utilised for our student-staff partnership, we envisage that these steps could be tailored and transferred to the use of Copilot as well.  

Zahra and Alia's top tip

C – Clarify the research question 
Ensure the research question is clearly defined and specific. 
H – Harness relevant databases 
Identify and utilize appropriate databases. 
A – Assemble search terms and strategies 
Develop a comprehensive list of search terms and construct search strategies. 
T – Tailor search parameters 
Adjust search parameters such as publication date ranges, language, and study types. 
G – Gather and organize results 
Collect search results systematically, ensuring that they are organized and documented accurately for easy reference and analysis. 
P – Perform iterative searches 
Refine and adjust search terms and strategies. 
T – Track and report search process 
Maintain a detailed record of the search process to ensure transparency and reproducibility. 

Contributors:

• Ms Veronica Parisi, Former Training and Clinical Support Librarian, UCL Cruciform Hub
• Dr Zahra Mohri, Associate Professor and MS Aesthetic Programme Lead, Division of Surgery and Interventional Science
• Dr Alia Galadari, MS Aesthetic student, Division of Surgery and Interventional Science