Should I cite the AI tool that I used? *** by Dr. Kristin Terrill, Iowa State University
Our last blog post about citing/acknowledging AI tools was very well received. One such response was worthy of publishing. Dr. Kristin Terrill is the Graduate Student Services Specialist at Iowa State University’s Center for Communication Excellence. The response is very specific, so we are publishing it in two related posts: “Should I Cite AI Tools?” (the post below) and “How to Cite AI Tools.” Enjoy …
First, let’s disambiguate between two questions: should I cite the AI tool that I used, and how should I cite the AI tool that I used? The first question rests on the nature of your AI tool use, and to answer it, I will break down aspects of research into parts. The second question is procedural, and to answer it, I will consider what software is. I will also address acknowledgments as an alternative to citation.
Should I cite the AI tool that I used?
Scholars do a lot of different things, and they use a lot of different tools. One thing a lot of scholars do is research, and by this, I mean formulating a research question and then collecting and analyzing data to answer the research question. They also write about the research they conduct and publish their writing in peer-reviewed journals.
Literature Review
In formulating a research question, scholarly researchers start by reviewing what is already known about that topic. This process typically involves reading papers by other researchers and synthesizing the knowledge into a cohesive narrative that leads to a specific question. Scholars then write up the narrative in a literature review and provide citations for the papers they read.
Literature reviews can be informal or systematic. Informal literature reviews, such as those found at the beginning of empirical research articles, usually don’t include citations for software that was used to facilitate the literature review process. These articles usually don’t report that cited articles were selected with the help of Google Scholar or the Readers Guide to Periodical Literature. So if you used an AI tool like Scite or Elicit while putting together an informal literature review for your empirical research article introduction, you don’t really need to cite it. Readers can go directly to the papers themselves.
You might be wondering about citing information directly generated by a chatbot. Chatbots like ChatGPT and Bing Chat advertise that these interfaces can provide reliable answers to questions. You might have even seen statistics claiming that answers provided by chatbots are 97% non-hallucinated. As a scholar, you should not cite a chatbot, however. Even if a chatbot provides the correct answer to a question, it still cannot be cited. The reason for this is that the answers generated by a chatbot are not archived publicly, so readers can’t follow the reference to evaluate your citation[1]. Academic citations include detailed information about where readers can find the information from the outside source, usually down to the exact page number. This is not possible with a chat bot-generated prompt response, so it cannot be cited.
If you are doing a systematic literature review (often called a research synthesis or meta-analysis), then the way you approach searching for and selecting sources is considered methodological, and the sources themselves should be treated as data. Therefore, you need to precisely report on the tools that you used to conduct the literature review. In this case, using an AI tool to find or screen sources should be reported as a methodological instrument, meaning that you need to cite the tool, just as you need to cite the database and any software that you use to analyze the content of the sources selected for the review.
In short, if the literature review is just for the introduction of an empirical research article or thesis/dissertation, then citing an assistive AI tool is probably not helpful. If the systematic literature review is the project itself, then using AI tools to conduct it is methodologically relevant and so you should cite the tool.
Be aware that as the author of the paper, you are responsible for the content of the paper. If your literature review relies on an AI-generated summary, and that summary turns out to be invalid, you are on the hook. Citing the AI tool that generated the invalid summary will not release you of responsibility. Inaccurate and misleading quotations, paraphrases, and summaries of related literature are a form of plagiarism, so … don’t rely on an AI-generated summary alone for generating your literature review.
Methods
Now, if you used an AI tool to assist with collecting and analyzing data, then you do have to cite it. The point of a methodological report (and this also applies to systematic literature review methodology) is to enable fellow researchers to replicate the results. If fellow researchers cannot replicate the results following the same method, then any theory derived from the research is rightly called into question. How might you use an AI tool as a method? An example might involve describing the linguistic features of chatbot-generated text. For a study like this to be replicated, follow-up studies would need to know specific details of the chatbot technology used to generate text for the analysis. You can understand why a full citation would be warranted in this case.
Article Writing and Publishing
If you’re a scholar using AI tools, you’re probably at least considering using them to help you with your article writing. Generating accurate, cohesive text is one of the most exciting applications of these technologies, so it’s natural to want to use it for laborious and high-stakes writing that you do professionally. However, it’s controversial, and many publishers require authors to disclose if they used assistance in their writing. But should writers cite an AI tool that they used to help them write the article? No. Writing is part of the research process, but it is not research. Scholars have always used tools for writing, and these tools are not cited. It is assumed, for instance, that research articles are composed with a word processor like Microsoft Word or Google Docs, but authors don’t cite those word processors. The tool used for writing a paper doesn’t impact the meaning of the content of the paper.
Well, that’s not entirely true. Imagine if your word processor had an error where it automatically deleted every fourth line of text when you converted your file to a PDF. That would certainly impact the meaning of the content of the paper—it would make it completely incomprehensible. In that case, however, the responsibility would lie with the author. No publisher would agree to publish such a paper with the caveat that the author agreed to cite that word processor and lay the blame for the problem at the feet of the software publisher.
In the same way, if you use a chatbot to generate text for your paper, you are responsible for the quality of the paper, just as you would be if your word processor garbled the content. You can’t deflect responsibility with a citation.
Recommendation Table for AI Tool Use Cases
Takeaways
Should I cite the AI tool I used? It depends:
Informal lit reviews - No
Systematic reviews - Yes, cite search/screening methods
Data collection/analysis - Yes, cite as method
Writing assistance - No citation needed; author responsibility
When AI tools directly impact methodology or findings, cite them. When they merely aid the process, no citation is necessary. Authors retain responsibility for the final published content.
[1] Rarely, academic writers do include citations to personal correspondences, which can’t easily be accessed by readers. However, it’s not the only evidence provided for a claim, and it almost always involves personal correspondence with either a renowned expert or someone who has a unique kind of knowledge on the topic. A chatbot response lacks the ethos that would warrant citing it in the same vein as personal correspondence.
