Back to Modules
Module 4
intermediate

AI for Research & Literature

Use AI effectively for literature review, research synthesis, and academic writing while avoiding common pitfalls.

45-60 min7 sections
Learning Objectives
  • Use AI to summarize and synthesize research papers
  • Understand limitations for literature review
  • Apply AI to research writing workflows
  • Verify AI-generated citations and claims
Contents
  1. 1.AI in Research Workflows
    text
  2. 2.Paper Summarization
    text
  3. 3.Example: Research Synthesis
    example
  4. 4.The Citation Problem
    tip
  5. 5.Writing Assistance
    text
  6. 6.Exercise: Verify AI Output
    exercise
  7. 7.Module Summary
    text
text
AI in Research Workflows

AI can significantly accelerate research workflows, but understanding its limitations is crucial for maintaining academic integrity.

Where AI Helps:

  • Summarizing papers you've already obtained
  • Synthesizing themes across multiple sources
  • Improving writing clarity and flow
  • Brainstorming research questions
  • Explaining complex statistical concepts

Where AI Falls Short:

  • Searching current literature (training cutoff issues)
  • Providing reliable citations (hallucination risk)
  • Accessing paywalled content
  • Knowing about recent publications
  • Replacing systematic review methodology

Critical Warning:

Never cite a reference provided by an AI without independently verifying it exists and says what the AI claims. LLMs frequently generate plausible-sounding but non-existent citations.

text
Paper Summarization

AI can help quickly extract key information from papers you've already obtained (not for finding papers).

Effective Summarization Prompt:

Summarize the following research paper. Include:
1. Main research question/objective
2. Study design and population
3. Key findings (with numbers if provided)
4. Main limitations acknowledged by authors
5. Clinical implications

[Paste the abstract or full text]

For Radiation Oncology Papers:

Add domain-specific requests:

Additionally, note:
- Treatment details (dose, fractionation, technique)
- Primary and secondary endpoints
- Toxicity outcomes
- Follow-up duration

Limitations to Remember:

  • AI summarizes what's in the text—it can't critique methodology
  • Numbers may be misquoted; verify key statistics
  • Context from figures/tables may be missed
  • AI may not recognize important nuances in the field
example
Example: Research Synthesis

Scenario: You've read 5 papers on SBRT for early-stage NSCLC and want to synthesize the findings.

Prompt:

I've reviewed several papers on SBRT for early-stage NSCLC. Help me synthesize the following findings into a coherent narrative for a review article:

Paper 1: [Paste key findings]
Paper 2: [Paste key findings]
Paper 3: [Paste key findings]
...

Focus on:
- Local control rates
- Overall survival outcomes
- Toxicity patterns
- Dose-fractionation schemes used
- Patient selection factors

Note areas of agreement and any conflicting findings.

What AI Can Help With:

  • Identifying common themes across papers
  • Organizing information into logical sections
  • Suggesting structure for your synthesis
  • Highlighting discrepancies to address
  • Drafting transitional language

What You Must Still Do:

  • Verify the synthesis accurately reflects the source papers
  • Add your expert interpretation and context
  • Properly cite each claim to its source
  • Consider methodological quality differences
  • Draw your own conclusions
tip
The Citation Problem

This is the most important caution for academic AI use.

LLMs will confidently generate citations that:

  • Don't exist at all
  • Exist but don't say what's claimed
  • Have wrong authors, journals, or dates
  • Mix up details from multiple real papers

Example of AI Hallucination:

AI Output: "SBRT for early-stage NSCLC shows 3-year local control rates of 90-95% (Timmerman et al., JAMA 2018; Chang et al., Lancet Oncol 2019)."

Reality: These citations may be fabricated or the claims may be misattributed.

Safe Practice:

  • Never ask AI for citations - Find them yourself through PubMed, Google Scholar, etc.
  • If AI mentions a paper, search for it independently before citing.
  • Use AI to summarize papers you've already found, not to find new ones.
  • Verify every claim that you'll include in your manuscript.

Better Approach:

"I found this paper [full citation]. Summarize its key findings about local control outcomes."

Rather than:

"What are the key papers on SBRT for lung cancer?"

text
Writing Assistance

AI can be a powerful writing assistant for improving clarity, flow, and structure—without generating content you claim as original.

Appropriate Uses:

  • Editing for clarity:

"Improve the clarity of this paragraph while maintaining the scientific accuracy:"

  • Restructuring:

"Suggest a better organization for these points..."

  • Grammar and style:

"Check this text for grammar issues and awkward phrasing:"

  • Condensing:

"Shorten this paragraph to approximately 100 words while keeping the key points:"

  • Explaining feedback:

"A reviewer commented that my statistical analysis is unclear. Here's what I wrote: [text]. How might I clarify this?"

Avoid:

  • Having AI generate novel content you present as your own analysis
  • Using AI to write sections you should be writing based on your expertise
  • Hiding AI use from collaborators or in contexts where disclosure is expected

Disclosure:

Many journals now require disclosure of AI use in manuscript preparation. Check your target journal's policies and be prepared to disclose appropriately.

exercise
Exercise: Verify AI Output

Scenario: You asked an AI to help with your literature review on hypofractionation in breast cancer. It provided this text:

AI-Generated Text:

"Hypofractionated whole breast irradiation has become standard of care following publication of several landmark trials. The START-B trial randomized 2,215 women to conventional (50 Gy/25 fractions) versus hypofractionated (40 Gy/15 fractions) treatment, demonstrating equivalent local control and improved cosmesis with hypofractionation at 10-year follow-up (Haviland et al., Lancet Oncol 2013). Similarly, the Canadian trial by Whelan et al. (NEJM 2010) showed no difference in local recurrence with 42.5 Gy in 16 fractions. The ASTRO evidence-based guideline (Smith et al., IJROBP 2018) subsequently endorsed hypofractionation as the preferred approach for most patients."

Your Task:

Identify what you would need to verify before using this in a manuscript:

  • Do the START-B results and citation check out?
  • Is the Canadian trial correctly described and cited?
  • Does an ASTRO guideline with that citation exist?
  • Are the dose/fractionation numbers correct?

What to Actually Do:

  • Search PubMed for each citation to verify it exists
  • Check that the claims match the actual paper abstracts
  • Look up the actual ASTRO guideline and its authors
  • Cross-reference the dose-fractionation details

This verification step is essential before using any AI-assisted research content.

text
Module Summary

Key Takeaways:

  • AI cannot reliably search literature due to training cutoffs and hallucination risk. Use PubMed, Google Scholar, and other databases.
  • Never trust AI citations. Verify every reference independently before citing.
  • AI is excellent for summarizing papers you provide and synthesizing across multiple sources.
  • Use AI as a writing assistant for editing, restructuring, and clarifying—not for generating original analysis.
  • Disclose AI use according to journal policies and collaborator expectations.

Best Practices Summary:

TaskAI Appropriate?Notes
Finding papers❌Use databases
Summarizing papers✅You provide the paper
Generating citations❌Always verify
Synthesizing themes✅From sources you provide
Editing prose✅Improves clarity
Writing analysis⚠️Should be your own work

Next Steps:

Practice these concepts in the AI Playground with research-focused scenarios.

Patient Communication with AITry the Playground