Using LLMs for Literature Review

Introduction

"Publish or perish" is the adage of academic medicine. But between clinical duties, administrative burdens, and teaching, finding time to write is increasingly difficult.

Large Language Models (LLMs) act as a force multiplier for the physician-scientist. They can shave hours off the "blank page" phase of writing, help refine non-native English, and even critique the logic of your arguments.

Using them requires navigating a minefield of ethical considerations. This guide explores how to use LLMs effectively and ethically in your research workflow.

The Ethical Landscape: Disclosure is Key

Before we discuss how to use these tools, we must address the rules.

• Authorship: AI cannot be an author. It cannot take responsibility for the work.
• Disclosure: Most major journals (Nature, JAMA, Elsevier) now require a statement disclosing the use of AI tools in the preparation of the manuscript.
• Plagiarism: While LLMs generate original text, they can inadvertently mimic existing phrases. Always run AI-generated text through a plagiarism checker.

Standard Disclosure Statement Example:

"During the preparation of this work, the author(s) used [Name of Tool / Service] in order to [Reason]. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication."

Use Case 1: The "Zero Draft"

The hardest part of writing is starting. LLMs excel at creating a "Zero Draft" — a rough, ugly, but structured piece of text that you can actually edit.

Prompt Strategy:

"I am writing a discussion section for a paper about [Topic]. Here are my 3 key findings: [List Findings]. Here are 2 limitations: [List Limitations]. Please draft a 4-paragraph discussion structure that:

1. Summarizes the main findings.
2. Contextualizes them with existing literature (I will add citations later).
3. Discusses the limitations.
4. Proposes future directions."

Result: You get a structure. You can now fill in the citations, correct the nuance, and make it your own.

Use Case 2: The Abstract Editor

Condensing a complex 4,000-word paper into a 250-word abstract is an art form. AI is a great editor for this.

Prompt Strategy:

"I am pasting my manuscript below. Please draft a structured abstract (Background, Methods, Results, Conclusion) under 300 words. Focus heavily on the quantitative results."

Pro-Tip: Ask for 3 variations. "Give me one version that is conservative and academic, and one that is punchy and highlights the novelty."

Use Case 3: The "Reviewer 2" Simulator

Before you submit, let the AI critique your work. It is surprisingly good at finding logical holes.

Prompt Strategy:

"Act as a harsh, critical peer reviewer for a high-impact medical journal. Read the Introduction and Methods section below. Identify:

1. Logic gaps in the hypothesis.
2. Methodological weaknesses that are not addressed.
3. Clarity issues. Be specific and constructive."

Use Case 4: English Language Editing

For researchers for whom English is a second language, AI is a game-changer. It levels the playing field.

Prompt Strategy:

"Please edit the following paragraph for clarity, flow, and academic tone. Do not change the scientific meaning. Fix any grammatical errors."

What NOT To Do

1. Do not ask AI to "Find Citations": It will often invent them (Hallucination). Use tools like Perplexity or Elicit for finding papers, then read them yourself.
2. Do not paste unpublished data into public models: If you have proprietary data or novel molecules, putting them into ChatGPT is technically a disclosure. See our safety guide for more on protecting sensitive research.
3. Do not accept the output blindly: AI tends to use "hedge words" (e.g., "It is important to note," "Furthermore"). It often lacks the decisiveness of a subject matter expert. You must inject your voice back into the text.

Conclusion

Think of the LLM as a very fast, very eager graduate student. It can produce a draft in seconds that would take a human hours. But like a student's draft, it requires the steady hand of the attending physician to correct, verify, and finalize. Used wisely, it is the most powerful research assistant you will ever have.