Extract author relationships and citation connections from your uploaded papers in NotebookLM, then have ChatGPT write the code that turns those relationships into a visual network graph — no data science background required.
Upload papers → prompt NotebookLM to extract author-citation pairs → paste structured output into ChatGPT → get networkx or D3.js visualization code → run and read the graph.
This workflow has been tested with literature sets from sociology, computer science, public health, and education research. The prompt templates below handle the most common edge cases. Written by a team of AI superusers who teach visualization workflows to graduate researchers. Updated March 2026.
A literature network map is a graph where each node represents an author or paper, and each edge (connecting line) represents a citation relationship or shared reference. The map reveals things that are invisible in a reading list: which scholars are the hubs of a field, which research clusters cite only each other, and which papers bridge otherwise disconnected groups.
For researchers writing literature reviews, grant proposals, or dissertation chapters, a network map answers the question: "Who are the 5 people I absolutely must cite, and which intellectual communities am I implicitly positioning myself within?" That question is hard to answer by reading papers one at a time. It takes about 60 seconds to read in a good network graph.
The challenge is data extraction. Building the edge list — the table of "Paper A cites Paper B" relationships — from a collection of PDFs is tedious by hand. NotebookLM automates this with the right prompts, and ChatGPT writes the visualization code so you never need to learn Python or JavaScript.
The end result is a rendered network graph you can export as PNG, SVG, or an interactive HTML file. Nodes are sized by centrality (how many connections they have). Clusters are color-coded by research group or sub-topic. You can click nodes in the interactive version to see paper titles. This is a publication-ready figure suitable for a dissertation chapter, conference poster, or grant background section.
Upload 20–50 papers. Use NotebookLM's Mind Map feature first to get a visual sense of which concepts appear across multiple sources — this helps you decide which relationship type to extract (co-authorship, shared citation, or shared concept).
Use the extraction prompt below. Ask NotebookLM to list, for each paper, which other papers in the notebook it cites. The output is your raw edge list — the data that defines the network graph.
Ask NotebookLM to reformat the relationship list as edge pairs: "Source → Target" or CSV format. This structured output pastes cleanly into ChatGPT's code generation step without manual cleanup.
Paste your edge list into ChatGPT with the code generation prompt. Specify: Python (networkx) for a static publication figure, or D3.js for an interactive web graph. ChatGPT writes the complete runnable script.
Execute in Google Colab (free) or locally. Look for: hub nodes (high degree), isolated nodes (potential gaps in your literature), and bridge nodes (scholars connecting distinct clusters — often the most interesting for your own positioning).
| Task | Tool | Why this assignment |
|---|---|---|
| Extracting which papers cite which | NotebookLM | RAG grounding — only cites what's actually in your uploaded documents |
| Formatting edge pairs for export | NotebookLM | Structured output prompts work well in NLM's chat interface |
| Writing visualization code | ChatGPT | Canvas + code interpreter — iterates on code with live output preview |
| Debugging and styling the graph | ChatGPT | Code interpreter runs the script and shows errors inline |
| Interpreting what the graph means | Either | Claude for analytical interpretation; ChatGPT for visual annotation |
Copy any prompt. Replace bracketed placeholders with your field details.
Every prompt in this guide plus all prompts across the full category — advanced workflows, specialized use cases, and production-grade templates.
Category Bundle — one-time access
Unlock Category Prompts — $19.99ONE-TIME · 30-DAY GUARANTEE · INSTANT ACCESS
NotebookLM extracts citation relationships from what appears in your uploaded PDFs. If a paper's references section is poorly formatted or scanned as an image, extraction quality drops. For best results, upload text-layer PDFs from sources like Google Scholar, JSTOR, or Semantic Scholar rather than scanned documents.
For fields with very dense citation networks (100+ papers with 50+ references each), start with a subset: the 20 most-cited papers in your collection. Build the core graph first, then expand. A graph with 20 well-connected nodes is more readable and analytically useful than a hairball of 200.
ChatGPT's code interpreter can run your networkx code directly in the chat. You don't need to set up a local Python environment — just upload your CSV and ask it to execute. The rendered PNG will appear inline in the chat.