How to write literature review and Research Proposal using ChatGPT: PART 2 of Doing Research with AI

TL;DR

Refine the research title first, then use that exact wording to drive reference searches and ensure the literature review stays on-topic.

Briefing Cornell Notes

Briefing

A practical workflow for building a full research proposal with citations using ChatGPT centers on one idea: break the work into small, sequential chunks, generate each section separately, then assemble everything into a single document in Microsoft Word. The payoff is speed—turning what would otherwise take months of manual searching and writing into a structured draft that already includes a literature review, research objectives, methodology, and ethics—while still grounding claims in references that can be checked.

The process begins by refining a master’s-level public health project title around a specific study question: assessing vaccination rates and identifying key barriers to immunization in rural Uganda communities. From that finalized topic, the workflow shifts into citation gathering. Using ChatGPT (with the Pro capability mentioned as a requirement for browsing), the user prompts the model to “browse the internet” and generate as many relevant references as possible, focused on immunization coverage and obstacles to vaccination in rural Uganda. Each returned reference is treated as a clickable link for validation, and the user can request additional references until satisfied.

If ChatGPT Pro isn’t available, the transcript recommends a fallback: searching Semantic Scholar directly. The user copies the research title, visits semantic scholar.org, and runs searches there. When results are sparse—especially for overly broad titles—the transcript advises trimming irrelevant parts of the title to keep the query focused. The resulting articles are then saved and later fed back into ChatGPT.

Once references are assembled, ChatGPT is prompted to write a detailed literature review that uses citations “naturally” and incorporates both the references generated via browsing and the additional Semantic Scholar links. The literature review is produced as a standalone section, complete with in-text citations tied to the supplied sources. The proposal then expands section-by-section: ChatGPT is asked to generate the title and introduction, then to enlarge the abstract into a full page, and to expand the introduction further (the transcript targets roughly two pages). The literature review is also requested as a multi-page, “bulkier” document.

After the narrative sections, the workflow continues with the core proposal components. ChatGPT is prompted to generate research objectives and research questions, followed by a methodology section. When the methodology comes back “shallow,” the transcript’s method is to copy the weak section back into ChatGPT and request a more detailed, comprehensive version. The final step is ethics: prompting for full ethical considerations, including concerns and approvals required for studies involving human subjects.

The end product is described as a compiled proposal—around 16 pages and over 2,000 words—assembled in Microsoft Word. The transcript emphasizes that the draft is not meant to be final as-is: it should be edited, formatted, and optionally expanded by re-feeding any underdeveloped sections back into ChatGPT. Future steps in the series are framed around paraphrasing to reduce AI-detector traceability, and running plagiarism checks, with the claim that the generated work can pass those tests.

Cornell Notes

The transcript lays out a step-by-step method to generate a complete research proposal with citations using ChatGPT, anchored in one principle: split the task into small chunks and build the document section by section. It starts by rewriting a research topic about vaccination rates and immunization barriers in rural Uganda, then generating references via ChatGPT browsing (or via Semantic Scholar if Pro isn’t available). Those references feed into a literature review written with citations, followed by separate prompts to expand the abstract and introduction to target page lengths. The workflow then produces research objectives, research questions, a detailed methodology (iterated if too shallow), and an ethics section covering human-subject approvals. The final sections are assembled in Microsoft Word and edited/expanded as needed.

How does the workflow turn a research topic into a proposal-ready draft with citations?

It begins by rewriting the study title to sharpen the research focus: assessing vaccination rates and identifying key barriers to immunization in rural Uganda communities. That title is then used to generate references. With ChatGPT Pro, the model is prompted to browse the internet and return many relevant articles with clickable links for validation. Without Pro, the workflow switches to Semantic Scholar: the title is searched there, results are saved, and those links are later provided to ChatGPT. Finally, ChatGPT is prompted to write a detailed literature review using those references with citations integrated naturally.

What’s the recommended approach when reference searches return too few results?

When Semantic Scholar search results are empty or thin—often because the title is too broad—the transcript advises removing irrelevant parts of the title and keeping only the most relevant terms. The goal is to narrow the query so the search returns articles directly tied to immunization coverage and barriers to vaccination in rural Uganda.

How are the proposal sections produced and expanded over multiple iterations?

Each section is generated separately through targeted prompts: title and introduction first, then a short abstract expanded into a full-page abstract, then an introduction expanded to about two pages, then a multi-page literature review. After that, ChatGPT generates research objectives and research questions, followed by a methodology section. If the methodology is “shallow,” the user copies that section back into ChatGPT and requests a more detailed, comprehensive version before inserting it into the Word document.

What role does Microsoft Word play in the workflow?

Microsoft Word is the assembly point. The transcript describes copying the generated title, abstract, introduction, literature review, and subsequent sections into a new Word document. After all parts are pasted in, the document can be organized, formatted with a uniform font, and expanded if needed. The transcript also mentions adding a table of contents in a later video.

What ethical content is explicitly requested at the end of the proposal-building process?

The final prompt asks for full ethical considerations, including ethical concerns and approvals needed for the study—especially because the research involves human subjects. This is treated as a required closing section to complete the proposal package.

Review Questions

What specific prompts and page-length targets are used to expand the abstract, introduction, and literature review?
How does the workflow handle situations where the methodology output is too brief or shallow?
Compare the two reference-gathering routes described (ChatGPT browsing vs Semantic Scholar) and explain how each feeds into the literature review.

Key Points

1
Refine the research title first, then use that exact wording to drive reference searches and ensure the literature review stays on-topic.
2
Generate references in bulk with clickable links via ChatGPT Pro browsing, or use Semantic Scholar as a non-Pro alternative.
3
If search results are sparse, narrow the query by removing irrelevant parts of the title rather than keeping it broad.
4
Build the proposal section-by-section—title/introduction, abstract, literature review, objectives/questions, methodology, then ethics—so each part can be expanded and corrected.
5
When any section is too thin (especially methodology), re-enter that specific text into ChatGPT and request a more detailed, comprehensive rewrite.
6
Assemble everything in Microsoft Word, then edit and format for consistency; the draft is meant to be refined, not blindly accepted.
7
Plan for later steps like paraphrasing and plagiarism checking, using the generated draft as the starting material.

Highlights

The workflow’s central tactic is chunking: generate and expand one proposal section at a time, then merge results into a single Word document.

References can be gathered either through ChatGPT Pro browsing (with validated clickable links) or through Semantic Scholar searches when Pro isn’t available.

The transcript emphasizes iteration: shallow methodology text should be copied back into ChatGPT for a deeper rewrite.

Ethics is treated as a required final component, with explicit attention to human-subject approvals and concerns.