3. The Secret is with the Prompt. Terry Ewell applies six principles to guide prompts that yield effective results from LLMs. BDP 373, www.terryewell.com. 2026 Jan. 02. Loaded 2026 Jan. 08.

<Hummel’s Bassoon Concerto, Second Movement. Terry B. Ewell and Peter Amstutz, Live Recording, Oct. 1997, Kent State University. >

Hello, I'm Terry Ewell. This video provides successful approaches to interacting with large language models using six principles. One of the most important things to master with large language models is crafting great prompts. Too often, people get poor answers because the prompts were not adequately defined.

In this video, six principles are used to create the prompts. Originally, ChatGPT provided fourteen principles, and you can see all fourteen if you visit the website I provide below. On the screen, you'll see the principles in the left column. In the second column are examples of prompts for creating MLA citations. And in the third column are prompts for creating an essay on Baroque articulation. So, let's explore these principles and how they form our prompts.

So, number one, specify the role. And this is very important because the role conditions, it narrows, the space of probability where the responses come from. If you want something on music, you need to narrow it to an area of music. If you want something that uses grammar or, for instance, sentence construction, you want to narrow it to that area. And that's why we have here in the second column, where we're looking for citations in MLA 9 format, we say, “You are an English professor.” Hopefully, those English professors know how to create citations. And in the third column, where I'm looking for text about music, we have, “You are a university music professor.”

Now, number two, define the task, not just the topic. You want the output, the way in which the response will appear. So, for the second column with the MLA 9 format, your task is to edit the list of citations and, based on the context, provide an additional five peer-reviewed recent citations. So, this is looking again at what the prompt will look like. In the third column, where it talks about the paragraphs, you're going to give text and citations, in-text citations, on the topic of Baroque articulation. And we're asking it to review attachments, articles, and then also notes that I will provide to it as well. And from all of this to put together the text and at the end we'll also be providing some citations.

Number three, constrain the output format. So, here in the second column, where we're looking for these citations, it says, Provided in the ML9 format, no bullet points, and they must be single-spaced. In the third column, we've got here, The in-text citations must conform to MLA9 format. And after the paragraphs, there's going to be a separate thing with the citations and they must include the full citation. So, I'm giving here the output, the format I want.

Number four, provide the context before the instructions. So, I say here, please see the attached list of citations. So I'm going to import citations before ChatGPT does all of its work or whatever AI you use. And in the third column, you see I'm going to actually attach documents and put them in.

Number five, declare the audience. The output I say here should always be at the university graduate level. I'm looking for high-quality academic writing.

And last of all, for the sixth prompt, I have here set the epistemic standard. Epistemic means the knowledge standard, the quality of the response, the accuracy of that response. You'll see in the second column I say flag any non-existent citations. I want the LLM to not make up or hallucinate answers for me. I know it's eager to please, but no, we want correct citations, so it's going to need to search to make sure these citations exist. And then I wanted to give those five peer-reviewed citations. And then in the third column here for the four paragraphs of text, I'm saying as much as possible, include my personal wording. I don't want the AI to sound very much like anyone else. I want it to be from my own wording and then recrafted and put together.

Notice here also, I want the wording to be non-personal. This is the requirement in the academic courses that I am teaching.

OK, very good. I have the prompt here for the four paragraphs of the Baroque articulation. So, I will just cut that. Control C. I'm going to go to ChatGPT. I'm going to paste it there. I'm going to now insert the files. So, I'm adding files, clicking on those, that plus sign. Okay, here we are at the files I'm going to add. So, I'm going to add that one, that one, and that one. And open these. And now it's all loading in here. And let's press this button, see what happens.
Yes, here it comes up. Press the down areas. So here was my input: the articles. And here is the ChatGPT reply. I've got one, two, three, four paragraphs. And I've got in here the in-text citation. This is a correct in-text citation. These appear to be more links to the documents, so I will have to clean that up. But there's something here to work with.

Last of all, I want you to understand that you need to adopt an “editor-in-chief” mindset. Pretend that you are the editor of a publication, like a newspaper or a magazine. You've got all of these eager writers sending you information. But you are responsible for the final result. You are setting the titles. You're setting where the formatting goes. You are setting how things work together. And you may be doing editing yourself to clean up the writing that your other writers are doing for you.

Now, you might want to run these prompts through multiple LLM models. You've only seen me use ChatGPT here, but you can try different things. And if you get different answers, that should be a red flag.

You are the only human in the loop, so it's important that you make the final adaptations and use these skills.

Please visit my website that provides the full prompts, some additional information, and videos that you might find helpful.

As you can see, LLMs are a powerful aid to university students, teachers, and professors, and many others. Using them wisely and being aware of their deficits will minimize problems. Thank you.

<Music: Hummel’s Bassoon Concerto, Second Movement. Terry B. Ewell and Peter Amstutz, Live Recording, Oct. 1997, Kent State University.>