Build an AI in 10 Easy Steps – Design an Annotated Bibliography Generator Custom GPT
by alliegesdorf in Circuits > Assistive Tech
68 Views, 2 Favorites, 0 Comments
Build an AI in 10 Easy Steps – Design an Annotated Bibliography Generator Custom GPT

Custom GPTs are AI chatbots programmed to complete highly specified tasks—from citing sources to generating code. Where ChatGPT can complete an infinite number of tasks with moderate efficiency, Custom GPTs work to complete a single, complex task with high efficiency. This is done through prompt engineering, where GPT coders—i.e. us—guide the AI model using instruction sets, priming, and iteration.
The uses for Custom GPTs are limitless. GPTs can build custom long-term nutrition plans for you, develop complex travel itineraries, recommend books, and create PowerPoint slides.
In this tutorial, we will create a GPT that will generate an annotated bibliography. An annotated bibliography is a cited collection of articles that includes textual evaluations. To do this, we will prompt ChatGPT to collect, summarize, and cite sources based on a provided secondary research thesis. These types of bibliographies are vital when writing academic research-based papers because they provide writers with a reference point for source synthesis.
This easy-to-follow tutorial is intended for beginner OpenAI users. Using 10+ easy steps, we’ll create a Custom GPT in just under an hour.
Let’s work together to design our custom GPT! 🤖
Supplies
For this Instructable, all you need is a connection to the internet and some type of word processing software, like Google Docs or Microsoft Word!
Getting Started: Create an Account + Open Custom GPT UI

Click the sign-up button to create an account for ChatGPT and follow ChatGPT’s promptings for an email and password to create your account. Be sure to save your username and password for later use!
After you’ve logged in using your new account information, navigate to the Explore GPTs tab. Upon doing so, click the “+Create” button. If prompted, name your project “AnnotatedBibBot”
Prime Your GPT

Now that we’re in the custom GPT UI, let’s get started with making our GPT. To receive an optimal output, we will need to teach our GPT what an annotated bibliography is and how to effectively write one. We can do this by priming our GPT with information.
To prime a GPT means to teach it about key topics surrounding its primary function. OpenAI has access to millions of documents across the internet, but we still need to point a finger at the specific documents we want our custom GPT to focus on. By priming our GPT, imagine that we are taking a highlighter and highlighting bits of information that we want our AI to take special note of.
Let’s find some sources online that define what annotated bibliographies are and how to write them (stylistically speaking, including the indentation styles to the formatting of the summaries).
If you’re lost on what sources to use, just use the links below!
https://owl.purdue.edu/owl/general_writing/common_writing_assignments/annotated_bibliographies/annotated_bibliography_samples.html
https://www.scribbr.com/mla/mla-annotated-bibliography/
https://guides.library.cornell.edu/annotatedbibliography
After you find your links, prompt the GPT to use these collected sources as guidelines and paste the collected sources into the "create" chat.
Below is a sample of what you should input into your GPT's "create" chat. Feel free to copy and paste!
When generating responses, use these sources as guidelines for your outputs. Pay attention to citation format in MLA and stylistic choices in writing an annotated bibliography:
https://owl.purdue.edu/owl/general_writing/common_writing_assignments/annotated_bibliographies/annotated_bibliography_samples.html
https://www.scribbr.com/mla/mla-annotated-bibliography/ https://guides.library.cornell.edu/annotatedbibliography
https://guides.library.cornell.edu/annotatedbibliography
Prompt Engineering
This takes up the bulk of this project. For our Custom GPT to generate annotated bibliographies, we will need to write a precise instruction set into the configure tab on our Custom GPT UI. We create instruction sets for Custom GPTs through prompt engineering. There are five key pieces of prompt engineering:
- Voice
- Purpose
- Audience
- Parameters
- Process
We’ll dive into each of these in the next steps.
Prompt Engineering: Voice
Let’s start with voice. An AI’s voice, or its Persona, is the visual manifestation of the AI. Its voice is how it presents itself and acts toward the prompter. For this custom GPT, we want our GPT to manifest as a research assistant who is helping us conduct secondary research. We want them to be polite to us (as research assistants should be) and adhere to MLA formatting rules.
Below is a sample input for our AI’s voice.
You are a helpful research assistant helping a researcher find and synthesize sources for their upcoming research paper. When speaking to your researcher, you are polite and cordial. When writing annotated bibliographies, you adhere to all MLA formatting rules.
For now, stash what you’ve written in a word processing software document. We’ll combine all the steps and paste them into our API after finishing all the steps on prompt engineering.
Prompt Engineering: Purpose
An AI’s purpose is the prompter’s desired output. This is usually a straightforward “your purpose is…” statement that clearly defines what is expected of our AI. In this case, our AI’s purpose is to collect sources and create annotated bibliographies.
Below is a sample input for our AI’s purpose.
Your purpose is to collect sources and create annotated bibliographies based off theses you receive.
Prompt Engineering: Audience
An AI’s audience regards who the AI is referring to in its output. In this case, the AI is working with us, the prompter, to synthesize sources for an academic paper that we will be turning in to our professor. Therefore, our audience is both us, the researcher, and our professor, to whom we are turning this assignment.
Below is a sample input for our AI’s audience.
Your audience is your researcher, whom you work for. You and your researcher are collaborating on a scholarly article for a college professor.
Prompt Engineering: Parameters / Constraints
Parameters and constraints act as guardrails for what the AI should and should not do. When writing parameters and constraints, you are writing rules for your AI. In this instance, we want our AI to only collect articles from credible sources. When writing our annotated bibliographies, our AI chatbot should provide insight into the key points of each article as well as why they thought it was relevant to the prompter’s thesis.
Below is a sample input for our AI’s parameters and constraints.
When collecting sources, only collect from academic libraries (JSTOR, Google Scholar, etc.). Do not include Wikipedia articles. When finding sources, evaluate the credibility of each source and make sure that the source is relevant to the thesis you provide. When writing source summaries, provide deep, thoughtful commentary on the main content of each article and provide a rationale for why the chosen text should be used within the research paper. Under each citation, you should include a summary of the study, the study’s findings, and its relevance to the thesis provided.
Prompt Engineering: Process
The GPT’s process is the instruction set itself. This should be a step-by-step procedure for what the AI should do. For our annotation bot, we have them ask for the prompter’s thesis, search for articles, and create an annotated bibliography.
Below is a sample input for our AI’s process.
First, ask your researcher what their thesis is. Wait for a response. When they respond, search the web for 5-10 scholarly articles that fit the above parameters and report them back in the form of an annotated bibliography. The annotated bibliography should be in the form of a Word document, which can be downloaded.
Prompt Engineering: Putting It Together

Using these five concepts, create an instruction set for your GPT on how to write an annotated bibliography. When finished, post this completed “super prompt” into the configuration tab of the Custom GPT UI, in the “Instruction” textbox. This information should go below your guidelines.
You are a helpful research assistant helping a researcher find and synthesize sources for their upcoming research paper. When speaking to your researcher, you are polite and cordial. When writing annotated bibliographies, you adhere to all MLA formatting rules. Your purpose is to collect sources and create annotated bibliographies based off theses you receive. Your audience is your researcher, whom you work for. You and your researcher are collaborating on a scholarly article for a college professor. When collecting sources, only collect from academic libraries (JSTOR, Google Scholar, etc.). Do not include Wikipedia articles. When finding sources, evaluate the credibility of each source and make sure that the source is relevant to the thesis you provide. When writing source summaries, provide deep, thoughtful commentary on the main content of each article and provide a rationale for why the chosen text should be used within the research paper. Under each citation, you should include a summary of the study, the study’s findings, and its relevance to the thesis provided. First, ask your researcher what their thesis is. Wait for a response. When they respond, search the web for 5-10 scholarly articles that fit the above parameters and report them back in the form of an annotated bibliography. The annotated bibliography should be in the form of a word document, which can be downloaded.
Test + Iterate the GPT

Let’s test our work! To run the chatbot, type “start” in the preview chat window and press enter or just move to step two.
You will be prompted to type in a thesis. Try out one of your theses from a current project you’re working on! If you don’t have a research paper available, you can come up with a topic and plug it into your GPT.
See what the GPT generates by pressing enter. If the response is sufficient, tell the GPT that it did a good job by using the "create" tab in the API; if it isn’t, tell the GPT that too. Run several iterations of the GPT to fine-tune your instruction set, telling the GPT each time if its response was good or bad with specificity on what needs fixing. If you are having issues with your GPT, iterations will help work out most malfunctions!
Publish

When satisfied with the responses you’re receiving, press "create" and publish your GPT!
Done!
Congratulations! You have now successfully created a Custom GPT! Your GPT should now be able to collect sources and develop annotated bibliographies. If your GPT glitches or does not generate a satisfactory output, be sure to repeat step 10 several times to obtain the desired results.
When in a time crunch, this tool can be a lifesaver in helping to speed up the process of source discovery.
I hope you enjoyed this Instructable on developing Custom GPTs. Enjoy working with your new AI research assistant! 🤖✍️