Research Workflow · Literature Analysis1 free

Structured Distillation: Generate an Evidence Matrix in 5 Minutes

Reading 30 papers takes a week. With NotebookLM, it takes 5 minutes. Upload your entire literature set into a single notebook, use the Notebook Guide to generate a panoramic overview, then issue a precise extraction command to produce a Markdown table organized by Author — Method — Core Conclusion — Page Number. The result is a citable evidence matrix: traceable, referenceable, and ready to drop into your paper. This isn't a summary. It's a structured knowledge asset with provenance.

Traditional method vs. structured distillation

Traditional

Manual reading + note-taking

5–7 days
30 papers × close reading each one

Read each paper sequentially, annotate by hand, organize notes in Excel or Word. Information scattered across multiple files, nearly impossible to cross-reference. The probability of missing key findings rises exponentially with the number of papers.

Structured Distillation

NotebookLM evidence matrix

5–15 min
30 papers × single extraction pass

Upload all papers into one notebook. AI cross-analyzes every source simultaneously and generates a structured table with page-level citations. Every conclusion traces directly back to the original passage. Zero omissions, fully verifiable.

Why NotebookLM can do this

NotebookLM's core architecture is RAG (Retrieval-Augmented Generation): it won't fabricate information — it only answers based on documents you've uploaded, and tags every claim with a source citation. When you ask it to "extract all findings related to a key variable across all papers," what it actually does is: scan every document → locate relevant passages → extract structured information → format the output → attach provenance. This is exactly the core motion of an academic literature review — just 100 times faster.

The key insight: you're not asking AI to "summarize" your literature. Summaries lose precision. You're asking it to execute a structured information extraction task — pulling specific data points from each paper along the dimensions you define (author, method, conclusion, page number) and returning them in tabular form. This is closer to a database query than a reading report.

The role of Notebook Guide

After creating a notebook and uploading your literature, NotebookLM automatically generates an overview through its Notebook Guide feature. This is your "panoramic map" — it tells you the overall landscape of your literature set: what the papers collectively study, where they agree, where they diverge. Read this overview first, then craft your extraction command. Your prompt will be more precise because you already know the "terrain" of your collection.

The four-step workflow

01

Create a focused notebook and upload your literature

Create a dedicated notebook in NotebookLM for your research question. Upload all relevant literature — PDFs, Google Docs, web links — as sources. Keep the focus tight: one notebook per research question.

Recommended limit: 30–50 papers per notebook. If you exceed this, split into sub-topic notebooks, then create a master notebook that stores the outputs from each sub-topic extraction.
02

Read the panoramic overview

After uploading, review the Notebook Guide's auto-generated summary. This is your terrain map — it reveals the overall themes, key concepts, major points of consensus and disagreement across your literature set. Spend 2 minutes reading it, and you'll write a far more precise extraction command.

If the panoramic overview reveals unexpected thematic branches, consider adjusting your key variable list — the literature may cover broader ground than you anticipated.
03

Issue the structured extraction command

Send NotebookLM a precise extraction prompt specifying your target variables and desired output format. The core template: "Based on all sources in this notebook, extract their findings on [KEY VARIABLE] and generate a Markdown table organized by Author — Method — Core Conclusion — Evidence Page Number."

Extraction dimensions are fully customizable. Beyond "Author – Method – Conclusion – Page Number," you can add columns for sample size, study limitations, relevance score to your research, theoretical framework, and more. The more specific your dimensions, the more useful the output.
04

Verify, iterate, and export

Check each citation in the table — NotebookLM provides direct links that jump to the original passage. For any suspicious entry, click through to verify the source paragraph. If some papers were missed or certain dimensions are incomplete, issue follow-up prompts to supplement the extraction. Export the final matrix as Markdown or copy it directly into your paper framework.

Structured Distillation Pipeline — Flowchart
You

Upload literature to a focused notebook

30–50 PDFs / Google Docs / web links, centered on a single research question

NLM Auto

Generate panoramic overview

Notebook Guide auto-summarizes: themes, key concepts, consensus, disagreements, research gaps

You

Define extraction dimensions + key variables

Based on the panoramic overview, decide which columns you need: author, method, conclusion, page, sample size…

NLM

Structured extraction → evidence matrix

Scans all sources along specified dimensions, generates a Markdown table with citations

Iterate: Verify citations → supplement missed papers → add dimensions → refine categorization. Repeat 1–2 rounds until the matrix is complete and usable.

Output

Citable evidence matrix

Export Markdown → paste directly into your paper / literature review / research report

Sample output: evidence matrix

demo data

Below is a fictional demonstration table showing what NotebookLM's structured extraction output looks like. In actual use, every conclusion includes a clickable citation link to the original source passage.

Evidence Matrix — Key Variable: AI's Impact on Hiring Bias 5 papers · 12 findings
Author (Year) Method Core Conclusion Page
Raghavan et al. (2020) Systematic audit, 15 AI hiring platforms Most platforms lack transparent disclosure of training data bias; audit found gender-correlated features implicitly encoded pp. 469–472
Li et al. (2022) Controlled experiment, N=2,400 resumes AI screening systems passed HBCU graduates at a rate 23% lower than PWI graduates; gap persisted after controlling for GPA and major pp. 15–18
Chen & Zhang (2023) Ethnography, 6 corporate HR departments HR practitioners treated AI scores as "objective" evidence, deferring to algorithmic recommendations even when they contradicted interview impressions pp. 102–108
Bogen & Rieke (2018) Policy analysis, legal framework review Existing anti-discrimination law is insufficient to address indirect discrimination in algorithmic hiring; new regulatory mechanisms needed pp. 28–34
Kim (2021) Mixed methods, survey + interviews N=180 Applicant trust in AI interviews correlated positively with system transparency (r=0.61), but actual transparency levels fell below expectations in practice pp. 44–47

Recommended source types to upload

📄

Journal Article PDFs

Your primary source. Ensure PDFs are text-based (not scanned images) for best extraction results.

📊

Research Reports

Institutional reports, white papers, policy documents. Often contain rich data tables and statistical findings.

🔗

Web Articles

Paste URLs to import directly. Good for preprints, blog-format findings, and news coverage of research.

📝

Google Docs

Your own reading notes, literature annotations, preliminary analysis. Cross-validated against primary sources.

🎓

Dissertations

Master's and doctoral theses typically include complete literature review chapters — ideal input for meta-analysis.

Customizable extraction dimensions

extend columns as needed
DimensionDescriptionWhen to use
Author (Year)First author + publication yearAll literature reviews — foundational column
Research MethodExperiment / survey / ethnography / meta-analysis / policy analysisMethodological comparison — essential
Core ConclusionThe paper's primary finding on the key variableAll contexts — essential
Evidence Page NumberSpecific page(s) or passage location supporting the conclusionAcademic writing — essential (traceability)
Sample Size / Data SourceN=how many, data originEmpirical study comparison — recommended
Study LimitationsAuthor-stated limitationsCritical reviews — recommended
Relevance to My Research1–5 score with brief rationaleLiterature screening — optional
Theoretical FrameworkThe theoretical basis used in the paperTheoretical integration analysis — recommended

Teaser Prompts

1 prompt

All prompts run in NotebookLM. Replace [brackets] with your specific details.

NotebookLM
"Based on all sources in this notebook, extract their research findings on [KEY VARIABLE — e.g., AI's impact on hiring bias / remote work's effect on productivity / social media's influence on adolescent mental health] and generate a Markdown table with the following structure: | Author (Year) | Research Method | Sample Size / Data Source | Core Conclusion | Evidence Page Number |. Requirements: (1) Extract at least one core finding per paper; for key papers, extract multiple findings. (2) In the 'Core Conclusion' column, summarize in one sentence — preserve critical data (effect sizes, percentages, correlation coefficients). (3) 'Evidence Page Number' must cite the specific page or passage location. (4) If a paper doesn't directly study the variable but has indirectly relevant findings, prefix the conclusion with '[Indirect]'." — Core extraction command. Use this template for your first pass.
Unlock All Prompts

Get the complete prompt library for this category.

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 — $19.99

ONE-TIME · 30-DAY GUARANTEE · INSTANT ACCESS

Limitations and practical notes

Structured distillation doesn't replace close reading. The evidence matrix is your navigation tool — it tells you each paper's core findings and where to find them, so you know which sections to read closely. For the key arguments in your own paper, you still need to return to the original text and read in context. The matrix transforms "blindly reading 30 papers" into "closely reading the 5 most critical passages."

PDF quality affects extraction quality. Scanned PDFs (image-based) produce significantly worse results than text-based PDFs. If your literature consists of scanned documents, run them through an OCR tool before uploading. PDFs from Google Scholar and most academic databases are typically text-based.

Always verify citations. Although NotebookLM's RAG architecture dramatically reduces hallucination, for any content you plan to formally cite in your paper, click through the citation link to verify the original passage. The evidence matrix is a first-draft tool, not a verification-free final product.

Cost: NotebookLM's free tier handles most use cases. NotebookLM Plus ($19.99/month via Google AI Plus) provides higher usage limits and faster response times, suitable for researchers who use it frequently or work with large literature sets.

Related Guides
Research Paper Workflow Literature Review Grounded RAG Hypothesis Generation PhD Literature Review Automated Literature Review
← Back to All Guides