How to Cite Sources

If you choose to use generative AI tools for course assignments, academic work, or other forms of published writing, you should give special attention to how you acknowledge and cite the output of those tools in your work. You should always check with your instructor before using AI for coursework. 

As with all things related to AI, the norms and conventions for citing AI-generated content are likely to evolve over the next few years. For now, some of the major style guides have released preliminary guidelines. Individual publishers may have their own guidance on citing AI-generated content. 

Here are some fundamental ideas that hold true for citing AI generated content, no matter which citation style you're using:

• Do cite or acknowledge the outputs of generative AI tools when you use them in your work. This includes direct quotations and paraphrasing, as well as using the tool for tasks like editing, translating, idea generation, and data processing. 
• Do not use sources that are cited by AI tools without reading those sources yourself. There are two different reasons for this:
  • Generative AI tools can create fake citations.
  • These tools may cite a real piece of writing, but the cited content may be inaccurate. 

• Be flexible in your approach to citing AI-generated content, because emerging guidelines will always lag behind the current state of technology, and the way that technology is applied. If you are unsure of how to cite something, include a note in your text that describes how you used a certain tool. 

• When in doubt, remember that we cite sources for two primary purposes: first, to give credit to the author or creator; and second, to help others locate the sources you used in your research. Use these two concepts to help make decisions about using and citing AI-generated content. 

Although generative AI chatbots can deliver answers to questions in a way that feels similar to Google, it's important to understand that these tools operate very differently from the search engines and retrieval systems that we're used to.

Google helps users find and retrieve text or other information that already exists and has been published on the Internet. It does this by identifying words and phrases in your search, looking for existing sources that match these terms, and applying a ranking algorithm to identify the results that are likely to be most relevant. 

ChatGPT does not have real-time access to the Internet, and it is not searching an existing body of text for matches. Rather than pointing users to information that is already published on the Internet, it creates new text in response to a query in a conversational tone, based on the information that it was trained on.

Put another way, search engines retrieve text that already exists and connect users to it, while generative chatbots create new text based on highly complex language models that attempt to provide the most likely sequence of words based on the information it has been trained on.

Generative chatbots don't refer to, quote from, or recall information from specific or actual sources. A generative chatbot, like chatGPT, will generate sources or appear to cite sources if you ask it to, but it's important to keep in mind that it does not actually "read" or "understand" these sources. In some situations, the chatbot will make up entirely fake citations. 

There are AI chatbots that have real-time access to the Internet, such as Bing Chat and Google Bard.  These work by using AI to transform a question into a set of queries used to search the Internet, and then generate a text-based response based on the information that it finds. Bing Chat and Google Bard will return internet search results along with the chat response, and in some cases will provide links to the sources used to create the chat response. 

Even though they have access to the Internet, Bing Chat and Google Bard can still have "hallucinations" and deliver false information. These systems frequently generate responses that contain unsupported statements and inaccurate citations. And, of course, the Internet itself is full of false or biased information. 

The free version of ChatGPT is not connected to the Internet, but there is a paid version, ChatGPT Plus, that includes beta features such as web browsing and plugins that allow you to access specific services and websites, and interact with them via ChatGPT. These features are still being tested, and are sometimes temporarily disabled.

Tools like ChatGPT, Bing Chat, and Google Bard are very broad in scope, but there are AI-driven research tools that specialize in navigating and discovering scholarly research. This includes tools like Elicit, which uses Ai to return relevant papers and summaries of key information about those papers in response to a question, or Scite.ai, which provides contextual information about the citations in a given paper. These are just two of many new research tools that are using generative AI to enhance the traditional search process for scholarly literature. 

Publishers of major scientific databases, such as Elsevier's Scopus or Clarivate's Web of Science, are also experimenting with AI-enabled search tools to help researchers discover sources in new ways. With these tools, publishers are attempting to increase the accuracy of responses by limiting the AI-generated responses to information that is retrieved using a more traditional search engine search. 

All of these emerging research tools are experimental, and should be used with caution. In some cases, publishers will not accept papers that have made use of these tools. 

Using AI In Your Search Strategy

Many users of AI tools have observed that using chatbots for straightforward information retrieval is not the best use of this technology. "The chatbots are the least beneficial when we ask them questions and then hope whatever answers they come up with on their are own are true." However, these tools can be used in creative and effective ways to support and enhance the research process.  

The research process is highly iterative, and generative chatbots provide an open and creative space to explore ideas, learn about topics that are entirely new to you, generate research questions, and discover new directions of interest. When you have a new assignment, and don't know where to begin, asking chatGPT to summarize the current state of research on a topic may be just what you need to begin the brainstorming process.

Good for:

• Entry-level knowledge on a topic, domain or discipline
• A conversational and iterative brainstorming session
• Prompting you to think about a direction you may not have previously considered
• Creating basic outlines or tables based on text you input

Use Caution for:

• Highly niche or emerging fields and technologies. The training data for chatGPT is no longer transparently reported, so users don't have key information needed to know if the information you see is up to date. 
• Asking for the names of prominent researchers in a field. chatGPT has been known to display a gender bias for this type of question, entirely make people up, or list people who are deceased even if you ask for living researchers.

In addition to the more well-known chatbots like chatGPT, Bing, and Google Bard, there are new and specialized research discovery platforms being developed to help brainstorm research questions, specifically. One such tool is Elicit.org.

A key skill to develop as a strategic researcher is the ability to brainstorm effective search terms, including synonyms. When you are starting to research a new topic, it can be difficult to brainstorm synonyms, phrases or terms of art that are likely to appear in the sources you most want to find. Ask a chatbot for a list of synonyms, similar concepts, phrases, theories, or methods the next time you begin a new research project!

There are a number of tools that are specifically designed to summarize articles or other documents. Many of these, such as Scholarcy, are fee-based tools that are designed to help students and researchers generate summaries of articles and other readings without having to read the full document. These tools can be useful for managing a large number of articles when you need a basic familiarity with their contents, but in most situations you should fully read sources when you need a deep understanding of the material. 

You can also use popular chatbots like ChatGPT, Bing, or Bard to summarize readings for you by pasting in the text and asking for a summary. You can give additional prompts to refine the summary, such as asking the chatbot to assume a certain level of knowledge on the part of the reader ("Explain it as if I'm 12 years old...") or to focus on specific information ("Summarize the methods section..."). Chatbots have a limited amount of memory for a chat session, so you may not be able to generate summaries of very long document. However, Anthropic Claude, yet another chatbot, has enough memory to summarize a short book. 

It is important to remember that these tools are not actually reading or understanding the documents that you give it, and the summary will be influenced by the dataset that the AI tool was trained on. Before using an AI tool in this way, try testing it out with some readings that you know and understand very well, to get a sense of the quality of the summary. 

When using any of the generative AI chatbots for your research, it is essential to develop your fact-checking skills, including tracing claims, locating original sources, and verifying citations or 'cited' content in the generated response. Depending on the nature of the information you need to verify, you may be able to do this with Google, but when using chatbots for academic research, you may need to use library resources if the information is 'coming' from an academic journal article, for example. Always trace the claim to the original source before using it in your own work. 

Attribution of Images: search engine by Tommy Lau, database by Siti Solekah, factcheck by Tiofei Rostilov, ideation by Kamin Ginkaew, Boolean by Juicy Fish, Aggregate Compil by Eucalyp