Recently, I’ve participated in discussions among writers and clients alike as to whether, in a non-fiction book that provides footnotes, ChatGPT should be cited if information or text from the algorithm has been included in the manuscript.
Generally, in a non-fiction book, you cite a source with a footnote if that source has provided:
1. A direct quotation.
2. Specific information that would be considered proprietary to that source, or over which the source could claim a copyright. For example, a published academic study that produced unique results would require a citation. Keep in mind that outside of academia, ideas cannot be copyrighted. Only the actual words and sentences can be copyrighted.
3. Optionally, if you want to give your assertion additional authority, you can credit the source even if it’s not strictly required. Such as, “The New York Times reported that the endangered piping plover shorebird has gained in population on East Coast beaches.” You could probably dig around and find these data in one or more primary sources, but it’s simpler to credit the newspaper.
Primary sources offer firsthand, original accounts or data, such as diaries, research reports, or interviews. In a published book, the citation of a primary source may be required. Secondary sources interpret, analyze, or summarize primary sources, like textbooks or historical articles. The key difference is the creator’s relationship to the information: primary sources come directly from the experience, and secondary sources are removed, providing analysis or context. In a published book, you do not cite a secondary source.
ChatGPT is proficient at searching the internet for information on a topic, analyzing it, and then—this is the really innovative part—composing new text to convey it.
But ChatGPT is not a primary source. It does not produce any first-hand information. It’s akin to Wikipedia, with the difference being that the text on a Wikipedia page is copyrighted by Wikipedia, just like the text in a New York Times article is copyrighted by the newspaper.
If you use Wikipedia, you’ll see that every piece of information is cited. So if you’re writing about a subject and you discover something about it on Wikipedia, instead of citing Wikipedia, which is a secondary source, you click on the footnote link and go to the primary source. You read the primary source and make your decision.
ChatGPT Is Like Your Summer Intern
If you use ChatGPT to produce rough drafts of text for a non-fiction book, it’s like having a human intern do the same thing. You possess the rough draft the algorithm has written or helped you write, and if necessary, ChatGPT will provide the primary sources for specific information it has uncovered. As the author of record, it’s your responsibility to verify the accuracy and fidelity of data provided by ChatGPT, just as you would verify the work done by your summer intern. While it’s getting better at being truthful, ChatGPT is still subject to hallucinations, especially with contemporary topics that are not yet historically settled.
You are not required to cite or acknowledge the use of ChatGPT or any other research/composing tool, just as you’re not required to publicly acknowledge the contribution of your intern (although it might be nice if you did!). And no matter how you choose to thank people or algorithms who have helped you write your book, if they are not primary sources, then you owe them no citation.