Principle 1
Specify the usable output.
Weak: “Tell me about these papers.”
Better: “Create a comparison table of research question, method, sample, finding, and limitation.”
Learning Product · Prompt Engineering · Updated July 2026
Stop collecting prompts. Learn to design them.
NotebookLM works best when you treat prompting as source-grounded task design, not magic wording. This guide teaches a reusable system for building, testing, diagnosing, and chaining prompts across an entire workflow.
01 · Quick Win
See the difference in 30 seconds.
Upload one useful source, paste the prompt below, and compare the result with a generic “summarize this” request.
Copy this prompt
Using only the uploaded source, create a decision-ready briefing. Include: 1. The source's main claim in one sentence 2. The three strongest supporting points 3. Any limitations, uncertainty, or missing evidence 4. A table with: claim | evidence | source citation 5. Three questions a careful reader should ask next Do not add facts that are not supported by the source.
02 · Mental Model
A NotebookLM prompt is an instruction for evidence work.
The source supplies the knowledge boundary. The prompt tells NotebookLM what operations to perform inside that boundary and how to present the result.
01
Goal
What decision, question, or learning outcome must the response support?
02
Evidence
Which uploaded sources may be used, and what should happen when evidence is missing?
03
Reasoning
Should the model compare, classify, synthesize, trace, critique, or evaluate?
04
Output
What structure, audience, depth, and format will make the answer usable?
05
Evaluation
How will citations, uncertainty, omissions, and quality be checked?
One system, three scales: the Prompt Design Canvas builds one prompt; Prompt Anatomy diagnoses its parts; the PPAE loop organizes an entire prompting session.
03 · Build the Skill
Use the Prompt Design Canvas.
Start with the job the prompt must perform. Add only the instructions that help NotebookLM complete that job reliably.
Annotated prompt
Act as a research synthesis assistant.
Using Sources 2, 4, and 7 only, compare how the authors define knowledge mobilization.
Identify agreements, disagreements, and differences in underlying assumptions.
Present the result as a three-column comparison table followed by a 200-word synthesis for doctoral students.
Cite every substantive claim. Flag any interpretation that is not directly stated in the sources.
Context / roleSets the working perspective when it materially changes the task.
Evidence scopeDefines the allowed sources, topics, dates, or sections.
TaskNames the intellectual operation: compare, trace, synthesize, test, classify.
Output designSpecifies structure, audience, length, and level of detail.
Quality checkRequires citations, uncertainty labels, missing evidence, or a verification pass.
Principle 2
Constrain the evidence scope.
Weak: “What does the research say?”
Better: “Use Sources 1–6 only. Separate direct findings from your synthesis.”
Principle 3
Name the reasoning operation.
Weak: “Analyze the sources.”
Better: “Compare assumptions, identify contradictions, and explain which evidence supports each interpretation.”
Principle 4
Design for iteration.
Weak: one prompt expected to produce the final answer.
Better: inventory → compare → challenge → synthesize → verify.
04 · Practice
Repair three weak prompts.
Do not memorize the improved versions. Identify what is missing, then rebuild each prompt with the Canvas.
“Summarize these studies.”
Add a purpose, evidence boundary, comparison dimensions, output structure, and quality check.
View one repair
Using all uploaded studies, create a table of research question, population, method, key finding, limitation, and citation. Then identify two recurring patterns and one unresolved disagreement. Do not merge findings that use materially different populations.
“Make me a study guide.”
Define what must be learned, the exam format, difficulty, and a feedback loop.
View one repair
Using Chapters 3–5, create a study guide for a closed-book short-answer exam. Organize it into core concepts, common confusions, five retrieval questions, and two application problems. Provide answers in a separate section with citations.
“Write an executive summary.”
Specify the decision, audience, risk tolerance, and what evidence should be excluded.
View one repair
Using the uploaded project documents, prepare a one-page briefing for a leadership decision on whether to proceed. Include recommendation, supporting evidence, unresolved risks, assumptions, and three questions requiring human judgment. Cite each factual claim.
05 · Build the System
Move from isolated prompts to a PPAE workflow.
The Canvas improves a single prompt. The PPAE loop manages the sequence of work, including source inspection, task planning, execution, and verification.
P · Perceive
Inspect the evidence.
Inventory sources, identify gaps, note conflicts, and learn what the notebook can actually support.
P · Plan
Design the task.
Choose scope, reasoning operation, output, and evaluation criteria before asking for a final product.
A · Act
Run the sequence.
Use multiple focused prompts instead of compressing every intellectual operation into one request.
E · Evaluate
Inspect and revise.
Check evidence coverage, citations, uncertainty, omissions, and whether the result serves the original goal.
1 · Inventory
List each source, its type, date, main claim, and likely relevance. Identify missing evidence.
Best fit: NotebookLM
2 · Extract
Pull claims, definitions, evidence, quotations, methods, or examples into a structured table.
Best fit: NotebookLM
3 · Compare
Identify agreement, contradiction, variation in assumptions, and differences in evidence quality.
Best fit: NotebookLM
4 · Synthesize
Create an explanatory structure that preserves distinctions rather than flattening the sources.
NotebookLM → Claude/ChatGPT
5 · Produce
Adapt the grounded synthesis into a briefing, lesson, article, presentation, or study asset.
Claude / ChatGPT / Gemini
6 · Verify
Return claims to the source set. Flag unsupported additions and unresolved uncertainty.
NotebookLM + human review
06 · Prompt Clinic
Diagnose before you rewrite.
Paste a prompt below. This lightweight local checker looks for explicit components. It does not judge truth or source quality; it helps you notice structural omissions.
07 · Reference Library
Use examples as scaffolds, not substitutes for thinking.
Each prompt below demonstrates the same architecture in a different context. Adapt the evidence scope and quality criteria to your actual sources.
Literature map
Map a source set without prematurely forcing consensus.
For each source, identify research question, conceptual lens, method, finding, limitation, and citation. Then group sources by genuine conceptual similarity and explain the basis for each cluster.
Contradiction audit
Find disagreements and determine whether they are substantive or methodological.
Identify claims that appear to conflict. For each, trace the supporting evidence, population, definition, and method. Classify the conflict as conceptual, empirical, contextual, or only apparent.
Retrieval practice set
Turn a chapter into an active study sequence.
Using the selected sources, create 12 retrieval questions across definition, comparison, application, and error diagnosis. Put answers and citations in a separate section.
Misconception detector
Focus practice on likely reasoning errors.
List five plausible misconceptions a learner might form from these materials. Explain why each is wrong, cite corrective evidence, and create one diagnostic question per misconception.
Grounded repurposing brief
Preserve the epistemic core while changing form.
Extract the central claim, essential evidence, qualifications, and non-negotiable distinctions. Then produce a handoff brief another AI can use to adapt the material without adding unsupported claims.
Source-to-article pipeline
Separate evidence synthesis from prose generation.
First create an evidence ledger with claim, support, citation, confidence, and caveat. Then create an outline that uses only ledgered claims. Export the outline and ledger for drafting.
Executive briefing
Build a concise, inspectable decision document.
Prepare a one-page briefing with decision, recommendation, evidence, risks, assumptions, and unresolved questions. Cite every factual claim and distinguish source evidence from interpretation.
Meeting preparation system
Convert documents into a decision-focused agenda.
Using the project sources, identify decisions required, unresolved dependencies, conflicting claims, and missing owners. Produce an agenda plus a pre-meeting evidence packet.
08 · Assessment
Can you design the prompt without copying one?
The goal is capability, not page completion. Check only what you can do independently.
Conditional Next Step
Continue through the knowledge mesh.
Your next page should depend on what you are trying to accomplish, not on a generic “related articles” list.
Researcher
Source Organization
Improve the evidence environment before building a research workflow.
Researcher
Grounded Research Loop
Place prompts inside a complete research operating system.
Creator
Content Repurposing
Move from source extraction to evidence-preserving distribution.
Student
Study Workflow
Build retrieval practice, feedback, and exam preparation systems.
Free implementation asset
Prompt Design Canvas + PPAE Worksheet
Turn the frameworks on this page into a reusable working document. Includes the five-part Canvas, a prompt repair checklist, and a six-stage workflow planner.