Field note
Prompt Patterns for Literature Search and Review
An experimental notebook of prompts I used with GPT-4o to find, organize, and discuss academic literature.
I initially had little patience for elaborate prompt templates. These examples grew out of experiments with GPT-4o in November 2024, when I found that a clear scope and an explicit output format made literature work easier to inspect.
A Markdown reference was useful when preparing the inputs and outputs.
A basic structure
The labels below are prompts for thinking, rather than fields that every request needs:
Role:
Task:
Instructions:
Required steps:
Requirements:
Context:
Examples or input:
Output format:
Finding representative papers
This example comes from work on multimodal discourse analysis in translation studies:
Role:
You are a researcher familiar with multimodal translation studies and
its theoretical development.
Task:
Use web search to identify five representative academic publications
on applications of multimodal discourse analysis in translation studies.
Requirements:
- Use established scholarly databases, publisher websites, or journals.
- Select peer-reviewed journal articles or conference papers.
- Do not invent references. If a bibliographic detail cannot be verified,
state what is missing.
- Consider influential work, authors with substantive contributions,
and papers that illustrate important theoretical or recent developments.
- Treat citation counts as one signal, not proof of methodological quality.
Output:
For each paper, provide its title, authors, year, a short summary,
and a source link that allows me to verify the publication.
I repeated the search with different keywords and perspectives to expand the reading list. Repeating the same request alone does not guarantee new or representative references; I still checked the sources.
Summarizing and synthesizing a supplied reading set
The second prompt deliberately limits the evidence to nine attached papers:
Task:
Summarize and analyze the nine attached papers, then write a focused
literature review on multimodal discourse analysis and translation studies.
Step 1: Summarize and organize the papers.
- Give a concise account of each paper's central question and contribution.
- Group the nine papers into no more than three thematic categories.
- Avoid categories that substantially overlap.
- Identify the authors and year of each paper in every category.
Step 2: Write the review.
- Explain the main research directions and the substance of each category.
- Discuss how multimodal discourse analysis has been applied to translation.
- Identify limitations or open questions supported by this reading set.
Scope:
Use only the contents of the attached papers. Do not expand the review
with unsupported claims about the wider field. Focus on the international
research discussed in these papers and its applications to translation.
Suggested structure:
Paragraph 1: Introduce the thematic categories, explain their main concerns,
and connect representative author-year examples to translation applications.
Paragraph 2: Discuss the limitations and gaps that follow from the evidence.
The useful distinction was between collecting sources and making claims from sources. The search prompt produces a reading list; the synthesis prompt works from the papers themselves.