Generative AI

This guidance is subject to change as each style develops its standards. Please check with the publisher of each style for the most current guidance, or consult with your instructor or a librarian.

Official APA Style Blog: How to Cite ChatGPT (Blog post, April 7, 2023, updated February 23, 2024) Includes examples for:
1. Quoting or reproducing the text/output created by ChatGPT in your paper
2. Creating a reference to ChatGPT or other AI models and software

Note: The APA 7th edition is "wider" than just its referencing system - it's a style guide that details how to punctuate, use capitals, format papers etc. In most cases in Trinity, you are NOT being asked to use those guidelines even if you are being told to "use APA" - you are using the referencing style only. Similarly, the APA publish journals and have given guidance on how authors of articles in those journals should acknowledge and cite the use of Generative AI. Unless you are told to specifically refer to those notes, the Library recommends you confine yourself to guidance given on acknowledging and referencing using the APA referencing style rather than as an author in an APA publication.

The APA Style team has likened citing the output from an AI model to sharing an algorithm's output. It must be noted that the output was generated by a Large Language Model (LLM) and the model identified.

In-text citation example, direct quotation:

When prompted with the question "is the beaver a destructive animal?", the ChatGPT-generated text indicated that "the influence of beavers on their environment can be both constructive and destructive, contingent on the particular context" (OpenAI, 2023).

Note: In their updated blog post, the APA Style Team suggests that "you may also put the full text of long responses from ChatGPT in an appendix of your paper or in online supplemental materials, so readers have access to the exact text that was generated." In this case, your in-text citation would be adjusted; e.g., (OpenAI, 2023; see Appendix A for the full transcript).

In-text citation example, paraphrased:

ChatGPT-generated text indicated that the categorization of beavers as 'destructive' or 'constructive' is multifaceted and not simply binary (OpenAI, 2023; see Appendix A for the full transcript).

Note: ChatGPT may not cite any sources unless 'web access' is turned on, or you've specifically requested them in your prompt. Even if references are requested, it is known to make up, or "hallucinate" sources that do not exist. Other AI models such as Bing AI Chat do cite sources, but it is important to assess the quality and reliability of each one individually. For example, when prompted with "Is the beaver a destructive animal?", Bing AI chat generated a response that cites a source promoting pest control practices and exterminators, which should be recognized as potentially representing a biased or unbalanced perspective (prompt output example, highlighted in blue):

Beavers are known to be one of the most destructive animals in the world1. They build dams on waterways which prevents water from flowing downstream and causes problems for the ecosystem because it affects other living things too1. Beavers can also be very destructive to property1. However, beavers are also paradoxical creatures; they create and destroy at the same time. Beaver dams flood valuable timber and beavers cut down thousands of trees around their dams. However, the flooding leads to increased fertility in the land, which is favorable to new plant and animal life2.

Learn more:

1. pestkill.org

2. courses.hamilton.edu

To cite the AI model itself, treat the model as software as outlined in Section 10.10 of the Publication Manual (American Psychological Association, 2020, Chapter 10). This will involve ;crediting the author of the algorithm (i.e., the creator/developer of the LLM) with a reference list entry and the corresponding in-text citation.

Basic reference list format:

Author/Creator/Developer. (Year). Model name (Version) [Large language model]. URL

Example:

OpenAI. (2023). ChatGPT (May 24 version) [Large language model]. https://chat.openai.com/chat/

• Author. APA recommends citing the creator (i.e., author/developer) of the model as the author (note: this is not the case for all citation styles, e.g., MLA).
• (Publication Date). Date of content creation - only need to include year. 2023.
• Name of the Model. The name of the model serves as the title and is italicized in your reference. ChatGPT.
• (Version specification). Identify the version you used in the format the author or publisher provides, which may be a numbering system (e.g., Version 2.0) or other methods. (May 24 version).
• [Additional description to provide context for the reader]. Bracketed text providing additional context is often included in APA citations for less common sources. For text-based generative AI, APA recommends providing the descriptor “Large language model” or "Large multimodal model" in square brackets. Pending further recommendations, use your best discretion based on how the creators/publishers describe the model you're citing. [Large language model].
• URL. When the publisher name and the author name are the same, do not repeat the publisher name in the source element of the reference, and move directly to the most direct URL available to access the model.

Note - these added colours are just to assist in recognising the elements of the citation, they aren't to be used in a real reference!