Is It Time for A New Citation Standard for AI Generated Content?
ChatGPT and friends are here to stay. It’s time for educators to discuss solutions.
In the education journal “Humanising Language Teaching” (ISSN 1755–9715), I recently published an article proposing a new citation standard for AI-generated content. This proposed citation standard would be comparable to APA, Chicago, MLA, etc. citation formats and would provide a means for acknowledging and confirming how AI was used in written work. Citation standards are essential to the fidelity and authenticity of scholarship. As generative AI technology continues its rapid pace of evolution and adoption, educators must lead the discussion of how we will respond to this new challenge, providing practical ways to preserve the integrity of the knowledge landscape.
As detailed in the article, this citation standard would include 5 elements essential to accurate attribution:
- Model name
- Date/year
- Purpose of use
- Retrieved from (URL)
- Prompt used
AI Model Name (Version). (Year). Purpose of use. Retrieved from [URL], Prompt: “Input prompt.”
The remainder of this article will briefly explain and demonstrate the use of each element.
Model Name
In my PhD research, I’ve been exploring the burgeoning effects of generative AI models (especially LLMs) on the field of the digital humanities. Daily, new models are being added to an already astounding number. Aside from the big commercial players (OpenAI’s ChatGPT, Google’s Bard, Meta’s Llama, and Anthropic’s Claude), there are hundreds of lesser-known, open-source models.
It goes without saying, then, that we must begin with the model name. Examples include Claude V2, GPT-4, Llama V2 (beta), etc. But a model is just the starting point. As the landscape of models is changing daily, so are the iterations of these models ever-changing.
Date/Year
To track the iterative nature of these models, it is necessary to locate them on a timeline. Each model has its own nomenclature for different versions. Because of this, simply indicating the model name and version number is not always clear. By adding the date of the model version (if available) or the date of use, interested parties can verify the author’s stated use of AI by interrogating the exact version of the model.
Purpose of Use
Here is an area where the citation standard differs significantly from citations for static resources. With generative AI, we are dealing with a dynamic, general-purpose technology. Working with an AI is not like researching an article or reference text. Such activities involve the investigation of information frozen in a moment in time. The purpose of such an activity, therefore, is obvious and directly related to the medium being cited. A single generative AI model may, by contrast, be used in many different ways. The model may have been used for the purpose of paraphrasing, summarizing, grammar correction, translation, formatting, text or image generation, etc. As such, no citation of the use of an AI model can be considered sufficient without a brief description of the purpose of use.
Retrieved from
Similar to the identification of model name/version and date, an additional means of verifying proper attribution is by including a link to the model used. This URL will not only allow interested parties to verify the model used but will provide them with a convenient (in-document) means of locating the actual model. As the landscape of private and open-source models becomes increasingly diverse, this field will become more important.
Prompt Used
This is where the proposed standard differs most significantly from traditional citations. Essential to the use of generative AI models is the concept of prompting — making a plain language request of the model. By including a direct quotation of the prompt, interested parties can verify the results by replicating the author’s procedure.
Example
Here’s how this would look for citing the use of AI in this article:
Microsoft Bing Image Creator (DALL-E 3). (Oct. 9, 2023). Creation of title image. Retrieved from https://www.bing.com/images/create?FORM=GENILP, Prompt: “A scholar in robes holds an open book to a glowing, circuit-lined AI figure. Between them floats a digitalized citation symbol, merging traditional and futuristic elements. The backdrop shifts from academic blue to techy green.”
OpenAI ChatGPT Plus (GPT-4). (Sep. 25, 2023). Proofreading. Retrieved from https://chat.openai.com, Prompt: “I want you to act as a proofreader. I will provide you with texts, and I would like you to review them for any spelling, grammar, or punctuation errors. Once you have finished reviewing the text, provide me with any necessary corrections or suggestions for improving the text.”
Educators often find themselves on the frontlines of the digital frontier. This moment in our technological history represents what is perhaps the most impactful change since the popularization of the internet. Embracing this forward position has always required evolving our academic traditions. With the pace of technological advancements accelerating, this evolution in education is more crucial now than ever before. As AI cements its role in scholarship, a dedicated citation standard isn’t just practical — it’s imperative.
