Skip to Main Content
London Metropolitan University
Library LibGuides
Skills Hub | Subject Guides | Library Search | E-Resources A-Z

Artificial Intelligence (AI) Library Guide: Checking Outputs

References from AI

decorative imageAlways check out the information provided by the AI tool's creators to find out the tool's latest capabilities before using it, and whether there is an added capacity to cite and reference sources. 

Citation and referencing problems with Generative AI Tools

One of the current features of AI is a failure to cite or reference at all, or if it does to do so accurately.  AI models rely on statistical patterns based on what likely word comes next (predictive text) and don't have a real understanding of how a citation or a reference is presented. This can lead to inaccuracies in the citations and references which might include:

  • Providing incorrect authors for a source;
  • Not providing DOIs or URLs, which means the reader is unable to link to the original sources;
  • Places of publication and dates may be incorrect or missing; 
  • Confusing resource types, for example, mixing up book chapters with journal articles;
  • Mixing up or giving incorrect journal and issue numbers for academic articles including grey literature; 
  • Incorrect page numbering for publication sources;
  • Incorrect edition statements;
  • Even when requested, a reference can be generated in the wrong referencing style.
     

Users need to check and verify sources given by any AI tool they have used.
It is always advisable to keep screenshots of AI output as the information is not replicable. This will enable you to have that information to hand when it comes to providing information on your use of the AI tool and the specific task carried out as well as sources used by the AI tool.
 

NOTE: Not all reference management tools have yet listed AI as a source material. If any citation is provided and has been checked for accuracy the user will have to manually enter the details on text, images or illustrations within their collection of references on a topic.


Recent research highlighting this problem can be found in Walters and Wilder research which provides an in-depth analysis of types of errors and fabricated citations generated by ChatGPT 3 and 4 in response to research queries across various disciplines.

Walters, W.H., Wilder, E.I. (2023) 'Fabrication and errors in the bibliographic citations generated by ChatGPT.' Sci Rep 13, 14045. Available at: https://doi.org/10.1038/s41598-023-41032-5.