Step1 Notebooklm Guide

Supplementary Sources

Chapters 20, 21, and 22 are NOT standalone podcast chapters — they are supplementary material to enrich all other chapters:

• Chapter 20 (Clinical Vignettes): Short clinical scenarios with classic presentations → use to teach chapter concepts in Step 1 question stem language
• Chapter 21 (Principles of Management): When to order what test, imaging types, management protocols → use to add management context to chapter topics
• Chapter 22 (Test Taking Skills): Exam strategy → not used in prompts

Chapter-to-Vignette Mapping

When generating a prompt for chapter N, extract vignettes from chapters 20/21 that match chapter N's topics. Key mappings:

Proven Prompt Template

This opening block MUST appear at the start of every prompt:

USMLE Step 1 review podcast for a med student in dedicated study mode. The PRIMARY source is the {Chapter Name} chapter — cover EVERY topic in it, skip nothing, including all tables and drug lists. The clinical vignettes file is supplementary: use those vignettes to TEACH the chapter concepts in Step 1 question stem language, not as extra content. Use Socratic back-and-forth: one host quizzes, the other answers. Compare confusable pairs side-by-side. Flag commonly tested tricky concepts and classic board traps. After every 3-4 topics, do a 30-second rapid-fire recap of key takeaways.

COVER ALL CHAPTER TOPICS IN THIS ORDER:

High-Yield Patterns to Encode in Every Prompt

1. Confusable Pairs

Step 1 loves testing distinctions. Always identify and request side-by-side comparison for pairs like:

• Hypertrophy vs Hyperplasia
• Reversible vs Irreversible injury
• Apoptosis vs Necrosis
• Red vs Pale infarcts
• Competitive vs Non-competitive inhibitors
• First-order vs Zero-order kinetics
• Type A vs Type B drug reactions
• Caseating vs Non-caseating granulomas
• Dystrophic vs Metastatic calcification
• Sensitivity vs Specificity

2. Classic Board Traps

Flag with "Trap:" in the prompt:

• Dysplasia is pre-neoplastic but STILL reversible
• Brain does NOT undergo coagulative necrosis (liquefactive instead)
• Eccrine sweat glands are sympathetic but cholinergic
• Chronic alcohol = CYP450 inducer, Acute alcohol = inhibitor
• Atropine reverses muscarinic effects only, NOT nicotinic (muscle weakness persists)
• Neostigmine does NOT cross BBB (quaternary amine)

3. Arrow Notation for Pathways

Always encode pathways as A→B→C chains:

• LPS→macrophages→IL-1/TNF-α→COX→PGE2→posterior hypothalamus→fever
• Rolling(L-selectin)→tight binding(ICAM-1)→transmigration(PECAM-1)
• Ischemia→Na/K pump failure→cell swelling(reversible)→membrane damage(irreversible)

4. Drug-Pathway Integration

For every pathway, name at least one drug that targets it:

• COX pathway → NSAIDs, Aspirin
• TNF-α → Infliximab, Adalimumab
• PD-1/PD-L1 → Pembrolizumab, Nivolumab
• CDK4/6 → Palbociclib

5. Management Pearls

Integrate from Chapter 21 where relevant:

• "Any toxicity → measure plasma drug concentration"
• "Alpha-block FIRST then beta-block for pheochromocytoma"
• "All ICU patients get PPIs"
• "Abscess → drain + antibiotics (Metronidazole below diaphragm, Clindamycin above)"

Lessons from Testing

What achieved 100% coverage (Chapter 18):

• Every topic explicitly named with specific details
• Prompt was ~4,800 chars for ~8 pages of source
• Every drug, enzyme, and clinical association was listed by name
• Confusable pairs were explicitly requested

What achieved 83% coverage (Chapter 1 V1):

• Topics were named but some sub-details were omitted
• Missing items were ones NOT explicitly named: Isoproterenol, Mirabegron, Phentolamine, Bisphenol A, Radon, Echinacea, Pyruvate metabolism

What achieved ~97% coverage (Ch1 V2/V3, Ch15A, Ch16A, Ch17A/B/C):

• Revised prompt explicitly named every detail → recovered 15/15 previously-missed items
• PDF-only source was sufficient — adding OCR markdown did not improve coverage
• Strategy scaled across 10 podcasts consistently at ~97%
• ~97% appears to be the practical ceiling — last 3% are isolated sub-details NotebookLM occasionally skips

What caused 94% coverage (Ch16B — ETC skipped):

• The prompt was under the char limit (3,964 chars) but covered TOO MANY DENSE TOPICS (7 biochemistry sections)
• NotebookLM ran out of podcast time (~77 min) before reaching the ETC section at the end
• Lesson: The constraint is content density per podcast, not just prompt chars
• Fix: Split into B1 (5 sections) and B2 (6 sections) → each covers fewer topics in depth

Chapter Splitting Strategy

Split decisions are based on two constraints: the 4,900 char prompt limit AND content density per podcast.

Pages-per-Podcast Guidelines

Prompt Sections-per-Podcast Guidelines

Split Decision Flowchart

1. Count source pages
2. Classify content density (high/medium/low)
3. Divide pages by max-pages-per-podcast → number of parts
4. Find natural topic boundaries for split points
5. Write prompts with appropriate section counts per part
6. Validate: if a prompt has >6 dense sections or >9 medium sections, split further
7. After validation, split PDF and OCR markdown at matching boundaries

Tested Split Examples

Hard Limit

NotebookLM's customization box silently truncates at ~5,000 characters. Prompts MUST stay under 4,900 chars. Use dense notation: = instead of "is", → instead of "leads to", drop articles, abbreviate (HR, BP, HTN, HF, Rx, DOC).

Source Upload Strategy (Tested)

Testing across 3 podcasts (Ch1 V2, Ch1 V3, Ch15A) proved that the prompt is the primary driver of coverage, not the source format:

The OCR markdown adds no measurable coverage improvement. The prompt's explicit topic enumeration is what forces NotebookLM to cover content.

Source upload order:

1. Primary: Chapter PDF
2. Optional enrichment: OCR'd markdown (not required — use it to help you write a better prompt by identifying table/diagram content)
3. Supplementary: Vignettes file

When writing prompts, pay special attention to content from tables, diagrams, and flowcharts — explicitly name every item since these are at highest risk of being skipped if not in the prompt. Use the OCR markdown as a reference while authoring the prompt, even if you don't upload it to NotebookLM.

The Golden Rule

If a topic, drug, pathway, or association exists in the source material and is NOT explicitly named in the prompt, NotebookLM will likely skip it. There is no shortcut — you must name everything.

Vignettes File Format

# Clinical Vignettes Relevant to {Chapter Name}

Use these vignettes to teach concepts from the {Chapter Name} chapter in Step 1 question stem language.

---

## {Topic Group 1} Vignettes

- {Clinical scenario} → {Diagnosis}
- {Clinical scenario} → {Diagnosis}

## {Topic Group 2} Vignettes

- {Clinical scenario} → {Diagnosis}

## Management Principles

- {Management pearl relevant to chapter}
- {Management pearl relevant to chapter}

Only include vignettes that directly map to chapter topics. Do not include tangential vignettes from other organ systems.