I Redesigned 6 Years of Informatics Olympiad Lesson Plans with NotebookLM, and Tripled My Efficiency

I’ve been teaching C++ for the Informatics Olympiad for 6 years, with hundreds of lesson plans scattered everywhere. Last week, I used Google’s NotebookLM to redo them all. Finding a lesson plan went from 30 minutes to 3 minutes. Creating a new lesson plan went from 4 hours to 1.5 hours. This article breaks down the entire workflow for you.

I’m Xiao C.

Dedicated to spreading AI knowledge and educational creation under an anti-effort mindset.

Always on the podium, occasionally passing through the cafeteria.

AI tool hunter.

I’ve been teaching C++ Informatics Olympiad for 6 years.

I’ve taught over 1000 students cumulatively.

I have hundreds of lesson plans on hand.

Word, PPT, Evernote, WeChat Favorites, local folders — scattered everywhere.

Until last week.

I used Google’s NotebookLM to redo this pile of lesson plans.

The time to find a lesson plan dropped from an average of 30 minutes to 3 minutes.

The time to produce a new lesson plan dropped from 4 hours to 1.5 hours.

Reviewing mistakes from a class went from 1 day to 2 hours.

Tripled efficiency.

That’s not an exaggeration.

These are real time metrics I tracked.

In this article, I’ll break down the entire process for you.

From why lesson plans were driving me crazy, to how I used NotebookLM line by line.

At the end, I’ll give you 5 core prompt templates.

If you’re a teacher, coach, curriculum designer, or someone with a pile of “old content” you want to revive, this article is worth bookmarking.

1. After 6 Years of Teaching Informatics, Lesson Plans Drove Me Crazy Countless Times

Let me paint a real scenario.

It was during the 2023 CSP-J semi-finals training class.

11 PM. A parent messaged me.

“Teacher Xiao C, my child was stuck on a dynamic programming problem for 2 hours. Could you help explain the approach?”

Dynamic programming.

I’ve taught it hundreds of times.

But under dynamic programming, there are subsets: knapsack, interval, state compression, tree DP…

Knapsack itself has subcategories: 0/1 knapsack, unbounded knapsack, multiple knapsack, grouped knapsack…

In my mind, I knew which type this problem corresponded to.

But I had to find the lesson plan I had written before.

Turned on the computer.

Evernote, searched “dynamic programming”.

47 results.

Those weren’t knapsack problems.

Those weren’t the ones corresponding to this problem.

Opened the folder.

C++ Informatics Olympiad → Basic Algorithms → Dynamic Programming.

20 PPTs, 30 Words, 10 PDFs.

Clicked the third one — not it.

Clicked the seventh one — also wrong.

Ctrl+F to search keywords — nothing found.

Files weren’t named systematically.

Over time, even I didn’t remember which was which.

In the end, I spent 40 minutes finding the corresponding lesson plan.

But that lesson plan was based on a 2019 past paper problem.

The approach didn’t quite match this year’s problem, so I had to spend another hour modifying it.

One evening, 2 hours gone just like that.

This was my 5th year as an Informatics coach.

Similar things happened every month.

1.1 6 Years of Lesson Plans, Scattered Everywhere

I’m not a messy person.

On the contrary, I’ve always cared about knowledge management.

Word documents: Local folders organized into 6 main categories, dozens of subdirectories.

PPTs: Scattered across 3 USB drives.

Evernote: Over 100 selected teaching notes.

WeChat Favorites: Years of parent questions, screenshots of excellent student work — at least 200 items.

WeChat File Transfer: Temporarily stored materials.

Draft paper: One notebook per semester, several boxes over 6 years.

Printed handouts: At least 200 copies.

Scattered.

Very scattered.

So scattered that sometimes even I didn’t know where the materials were.

You think that’s bad enough?

Not yet.

1.2 The Real Problem with Lesson Plans Isn’t “Can’t Find”

What’s worse than “can’t find” is “can’t use”.

What do I mean?

A concrete example.

In 2019, I wrote a very detailed “Introduction to Dynamic Programming” lesson plan.

Cases, diagrams, code, common mistakes — all covered.

In 2024, I taught this topic again.

This lesson plan was still usable.

But it was written for students from 4 years ago.

Current students:

• Use a newer evaluation environment.
• Face different problem types.
• Have different error patterns.

I wanted to “reuse” that lesson plan.

But what I actually had to do was:

• Delete outdated parts.
• Add new cases.
• Adjust difficulty gradient.
• Regenerate matching practice problems.
• Redraw diagrams, take screenshots, reformat.

The workload was about the same as writing a new one.

That’s the real pain point for Informatics coaches (and all knowledge workers).

It’s not a lack of materials.

It’s too many materials with too high a reuse cost.

The result is reinventing the wheel every year.

1.3 All the Tools I’ve Tried

To solve this problem, I’ve tried almost every mainstream tool over the past 5 years.

Evernote: Weak search capabilities, couldn’t find keywords.

Youdao Note: Slightly better than Evernote, but AI capability was basically zero.

Notion: Powerful, but setup cost was high. Migrating 6 years of materials would take at least a month.

Feishu Docs: Good for collaboration, but not useful for solo work.

Obsidian: Great local tool, but configuration was too complex for me.

OneNote: High flexibility, but cross-device syncing was too slow.

Products specifically designed for teaching: Almost none existed.

The ones that did were SaaS, required payment, learning new operations, and giving up my existing workflow.

Until I met NotebookLM.

Honestly, I didn’t have high expectations at first.

A free tool, how good could it be?

But after using it for a week, I stopped all my old tools.

Below, I’ll break down the entire process for you.

2. What is NotebookLM? Why is It Different from Other AI?

First, let me explain what NotebookLM is.

It’s a personalized AI research assistant from Google.

It started gaining traction in China in late 2024, and fully went viral in 2025.

Its biggest feature, and what sets it apart from ChatGPT, ERNIE Bot, Tongyi Qianwen, and all other general-purpose AIs is:

It only answers questions based on the materials you upload.

What does this mean?

With regular ChatGPT, if you ask “what’s the state transition equation for the knapsack problem,”

it generates an answer based on its training data.

Sounds impressive, right?

But problems arise.

It can confidently generate nonsense.

It can mix up versions from different years and approaches.

It can use terms you think are correct but give wrong explanations.

For a teacher, this is fatal.

Every piece of knowledge you teach a student must be a version you personally endorse.

You can’t stand at the podium and say: “According to AI, this formula is like this, but it seems somewhat uncertain.”

NotebookLM solves this.

Here’s how it works:

• You upload materials (PDF, Word, Google Docs, web links, YouTube video transcripts, audio).
• It answers questions only based on the materials you provided.
• Every answer includes citations showing which material and which section it referenced.

In short, it’s like a “super student” that only reads the books you assign.

The more materials you give it, the smarter it gets.

But it will never “self-learn” things beyond your provided materials.

This feature is practically tailor-made for teachers.

2.1 Three Things It Can Do for You

After a week, I identified three scenarios where NotebookLM is most suitable for teachers.

First: Intelligent search across large volumes of materials.

Throw in 6 years of lesson plans.

Ask: “How should I explain state transition for interval DP in dynamic programming?”

It directly pulls relevant sections from your 2019, 2021, and 2024 lesson plans.

And sorts them by relevance.

Second: Secondary creation based on your materials.

Say: “Using my 2021 lesson plan and adding 2024 past paper cases, create a new lesson plan for me.”

It actually rewrites your old plan into a new one.

Preserving your core ideas while adding new content.

Third: Error analysis and personalized exercise generation.

Throw in a student’s mistakes.

It can provide targeted explanations based on similar problem-solving methods from your past lesson plans.

And generate new practice problems based on the difficulty gradient in your materials.

These three scenarios correspond exactly to the three most time-consuming tasks for an Informatics coach.

2.2 Understand NotebookLM’s Core Features in 5 Minutes

Don’t be intimidated by the name. “NotebookLM” — LM stands for Language Model.

Think of it as “notebook + large model.”

Here’s what the interface looks like (text description):

• Left side: List of “notebooks,” each with a topic.
• Middle: Materials area, where you can upload files and add links.
• Right side: Dialogue area, where you chat with the AI.

That’s it.

No complex settings.

No hotkeys to learn.

If you can type, you can use it.

Extremely beginner-friendly.

3. My 3-Step Practical Workflow: From 6 Years of Lesson Plans to an AI Knowledge Base

Now comes the most important part.

How I turned 6 years of Informatics lesson plans into a usable AI knowledge base in 3 steps.

These 3 steps aren’t just talk.

I’ve stepped into every pitfall and optimized along the way.

Step 1: “Move” All Lesson Plans into NotebookLM

The key here isn’t uploading; it’s organizing.

Don’t just dump hundreds of lesson plans in at once.

NotebookLM will be confused too.

The right approach is: Split into multiple Notebooks by topic.

I created 6 Notebooks:

• Basic Syntax Notebook: Variables, loops, arrays, functions, structs.
• Basic Algorithms Notebook: Sorting, recursion, binary search, greedy.
• Dynamic Programming Notebook: Knapsack, interval, state compression, tree DP.
• Graph Theory Notebook: DFS, BFS, shortest path, minimum spanning tree.
• Past Paper Analysis Notebook: Solutions for CSP, Blue Bridge Cup problems over the years.
• Student Mistakes Notebook: Typical errors and incorrect thought process analysis.

In each Notebook, put all related materials.

Word lesson plans, PPT screenshots, PDF past papers, web links — all work.

How to organize faster?

I used the lazy version:

• Upload all materials to cloud storage in one go.
• Organize into topic folders.
• Upload all files to the corresponding Notebook in one batch.

The whole process took me about 6 hours.

Faster than I expected.

Three pitfalls to avoid in this step (pre-warning):

Pitfall 1: Too many materials backfires.

I first put all lesson plans into one Notebook.

The AI kept quoting the wrong sources.

Later I realized: NotebookLM recommends keeping materials in each Notebook to under 50 files.

Above 50, accuracy drops.

Pitfall 2: Images and scans need separate handling.

I had many handwritten lesson plans photographed.

Directly uploading images gives low AI recognition rates.

Recommend using a scanning app to convert to PDF, or an OCR tool to extract text before uploading.

I use the scanner built into my phone (CamScanner), free version is enough.

Pitfall 3: Links must be stable.

I initially used some cloud storage links.

Later the links expired, and the AI couldn’t access them.

Now I consistently upload local PDFs.

Step 2: Replace Ctrl+F with “Conversation”

This is the core of the core.

Previously, finding materials was like this:

• Ctrl+F, search keyword.
• Manually browse through N results.

NotebookLM lets you:

• Ask questions as if chatting with a student.
• AI gives you the answer directly, plus citations.

How big is the difference?

A real example.

Last week a student asked: “Teacher, what’s the core difference between 0/1 knapsack and unbounded knapsack? I always get confused when solving problems.”

If I manually searched for lesson plans:

• Browse folders → 5 minutes.
• Find 3 relevant lesson plans → read 10 minutes.
• Organize thoughts → 5 minutes. Total: 20 minutes.

With NotebookLM:

• Opened “Dynamic Programming Notebook”.
• Typed: “What’s the core difference between 0/1 knapsack and unbounded knapsack?”
• After 3 seconds, AI gave an answer, citing 3 lesson plans from 2019, 2021, and 2023.
• I spent 1 minute checking the cited originals to confirm.
• Then explained to the student. Total: 1 minute.

Efficiency improved 20x.

I’m not exaggerating.

This is the fundamental upgrade in tools.

Three effective question formats (copy directly):

Format 1: Direct knowledge question

“How do I write the state transition equation for dynamic programming? Give me the simplest example.”

Format 2: Comparison question

“What’s the core difference between 0/1 knapsack and unbounded knapsack? Give me a comparison table.”

Format 3: Application question

“For a student who just finished learning greedy algorithms, how can I smoothly transition them to dynamic programming? Based on my lesson plans, give me a 5-week teaching plan.”

These three formats cover 90% of lesson preparation scenarios.

You can use them directly or adapt them to your subject.

Step 3: Let AI Help You “Recreate” Lesson Plans

This step is the key to truly unlocking productivity.

What does it mean?

As I said before, the biggest pain point with lesson plans isn’t finding them — it’s reusing them.

Old plans need updating, but you don’t want to rewrite entirely.

NotebookLM’s “Audio Overview” and “Study Guide” features can help.

Here are 5 prompts I commonly use (saved for you, see appendix for compiled version):

Prompt 1: Generate new lesson plan from old one

“Based on my uploaded 2021 dynamic programming lesson plan, add 2024 CSP-J semi-final past paper cases, and create a new lesson plan. Retain my original teaching approach, but update all cases and practice problems.”

Prompt 2: Adjust difficulty level

“Rewrite this lesson plan into two versions: one for absolute beginners, and one for students who have studied once and want to advance. Give me a detailed [X]-lesson teaching plan for each version.”

Prompt 3: Generate supporting exercises

“Based on this ‘Greedy Algorithm’ lesson plan, generate [N] supporting practice problems. Difficulty from easy to hard. First [N1] problems are basic concept questions, last [N2] are adapted from [competition] past papers. Give me detailed solution approaches and reference code for each.”

Prompt 4: Mistake analysis template

“I uploaded a student’s mistake collection. Based on the teaching methods for similar problem types in my lesson plans, give targeted analysis for each mistake, and summarize the student’s [N] main weak points.”

Prompt 5: Parent communication scripts

“I have a parent in [scenario] who [specific issue]. Based on the teaching philosophy in my lesson plans and the student’s progress data, give me a warm yet persuasive reply.”

I’ve been using these 5 prompts for a week, every day.

This is not simple “AI content generation;” it’s generating results in your style based on your own materials.

That’s the biggest difference from ChatGPT.

ChatGPT’s output is its style.

NotebookLM’s output is your style.

4. Real-World Comparison: Old Method vs. New Method

To avoid sounding like I’m bragging, here’s a real comparison.

All data based on my actual work over the past week.

Task	Old Method (Time)	New Method (Time)	Efficiency Gain
Find lesson plan for an old knowledge point	30 min	3 min	10x
Rewrite old lesson plan (update cases + adjust difficulty)	4 hours	1.5 hours	2.7x
Compile mistakes from a class of 30 students	1 day	2 hours	4x
Prepare materials for a new lesson	3 hours	1 hour	3x
Prepare materials for parent meeting presentation	6 hours	1.5 hours	4x
Create personalized practice problems for a student	2 hours/set	20 min/set	6x

On average, efficiency improved 3 to 4 times.

What can I do with the time saved?

• Spend more detailed code reviews with students.
• Research new CSP past paper trends.
• Write blog posts and short video scripts.
• Have dinner with family.

That’s the real meaning of tool upgrades.

Not to make you do more things.

It’s to let you do more important things in the same amount of time.

5. Why Does This Method Work? The Underlying Logic Explained

Some teachers might ask: Why can NotebookLM do all this?

Let me break down the underlying logic into 3 layers.

5.1 Layer 1: It Solves the Fundamental Pain Point of “Knowledge Retrieval”

Teachers don’t lack knowledge.

Teachers have too much knowledge and can’t find it.

Traditional Ctrl+F is essentially “keyword matching.”

It requires you to remember keywords and file locations.

NotebookLM uses semantic understanding.

You ask “what’s the core idea of 0/1 knapsack?” It understands you’re asking about the algorithm’s essence, not matching the string “0/1 knapsack.”

Then it picks the section from your lesson plans that truly explains the essence.

That’s the retrieval method the AI era should have.

5.2 Layer 2: It Turns “Personal Experience” into a Reusable Asset

A teacher’s 6 years of lesson plans are their most valuable asset.

But without AI, these assets are hard to scale and reuse.

Because each lesson plan was born for a specific student, year, and environment.

A new group of students means the plan doesn’t fully apply.

NotebookLM makes this simple:

• Throw old plans in as “raw materials.”
• Feed new requirements as “orders.”
• It produces finished products from the raw materials according to the orders.

Your experience is no longer one-time.

It becomes an assembly line.

5.3 Layer 3: It Preserves the “Teacher’s Flavor”

This is the most crucial point and the biggest difference between NotebookLM and general AIs.

Lesson plans generated by ChatGPT clearly don’t look like my style.

Sentence structures, word choices, case preferences — all have ChatGPT’s flavor.

Students can tell immediately: “This doesn’t sound like Teacher Xiao C wrote it.”

Content generated by NotebookLM, because it’s based on my own lesson plans, learned my way of expression.

For example, I commonly use “loading into QQ” as an analogy to explain function calls.

NotebookLM will continue using that analogy when generating new lesson plans.

That’s the fundamental reason why AI won’t replace teachers.

AI replaces repetitive labor, not the teacher’s style, judgment, or care for students.

6. The Boundaries of This Method: Where Is It Not Suitable?

But I don’t want you to think NotebookLM is a silver bullet.

It has several clear limitations.

Limit 1: You Need Your Own “Raw Materials”

If you haven’t accumulated any materials and rely entirely on AI to generate from scratch, NotebookLM can’t help you.

AI is an amplifier, not a generator.

Without original accumulation, even the best tool is useless.

That’s why I say: It’s never too late to start accumulating now.

Limit 2: Limited in Complex Reasoning Tasks

NotebookLM excels at “knowledge retrieval + simple rewriting.”

But when it comes to complex multi-step reasoning or creative breakthroughs, it’s somewhat weak.

For example, asking it to “predict 2026 CSP-J question trends based on these lesson plans.”

It will give you a seemingly reasonable answer, but the depth is insufficient.

For such tasks, you still need general AIs like ChatGPT or Claude.

Limit 3: Chinese Material Recognition Rate Needs Improvement

NotebookLM is a Google product.

Its recognition rate for English materials is significantly higher than for Chinese.

For my Chinese lesson plans, the AI’s answer accuracy is about 85%.

Not 100%.

So I always manually verify every answer it gives.

Cannot fully trust it.

This flaw will hopefully improve with Google’s optimization for Chinese.

Limit 4: Free Version Has Limits

NotebookLM is currently free (as of writing), but:

• Each Notebook can have at most 50 materials.
• Single conversation has length limits.
• Some advanced features may require payment.

For individual teachers, the free version is sufficient.

But for institutions or large teams, a paid version may be needed.

7. Who Is This Method Best For?

After all this, let me summarize: Who is this method best for?

7.1 K-12 Teachers / Tutoring Teachers / Coaches

You are the core target audience.

As long as you have years of accumulated teaching materials, this method can directly improve your efficiency by over 3x.

You don’t need to be a tech expert.

If you can use a browser and type, you can use it.

7.2 Curriculum Designers / Trainers

If you need to repeatedly update courses and generate new versions of handouts, this method lets you “stand on your own shoulders” instead of starting from scratch each time.

7.3 Content Creators / Social Media Influencers

If you write blog posts or produce video scripts and have accumulated materials, this method can quickly generate drafts that match your style, which you then polish manually.

7.4 Lawyers / Doctors / Consultants / Any Knowledge Workers

As long as your work involves “knowledge reuse,” this can work for you.

Throw in cases, documents, reports you’ve handled over the past years.

When needed later, retrieve them conversationally.

This is the skill every knowledge worker should master in the AI era.

8. My Final Verdict

Back to that frustrating night in 2023.

11 PM, a parent messaged, their child stuck on a dynamic programming problem.

If NotebookLM had existed then.

I wouldn’t have needed to browse folders.

I wouldn’t have needed to manually sift through 47 search results.

I wouldn’t have needed 40 minutes to find a 2019 lesson plan.

I would only need to open “Dynamic Programming Notebook” and type 3 lines.

After 3 seconds, the AI gives 3 versions of the lesson plan.

Citations are clear, cases match what the student recently learned.

I spend 3 minutes verifying, then explain to the student.

Whole process: 5 minutes.

Saved 1 hour 55 minutes.

What can I do with 1 hour 55 minutes?

Help the student with one more problem.

Have an uninterrupted dinner with family.

Write a blog post.

Get a good night’s sleep.

AI doesn’t replace teachers; it amplifies teachers’ experience.

AI doesn’t steal time; it gives time back to teachers.

AI doesn’t make people cold; it gives people time to do what truly matters.

That’s my biggest takeaway from this week with NotebookLM.

It hasn’t changed the essence of my teaching.

It just gives me more time to be myself.

Finally.

If you want to use NotebookLM but don’t know where to start.

First, upload your 10 most commonly used materials and try them out.

Even if it’s just one notebook with 5 files.

You’ll feel the difference.

Action is more important than perfection.

Starting is more important than mastery.

I’m Xiao C.

Dedicated to spreading AI knowledge and educational creation under an anti-effort mindset.

Always on the podium, occasionally passing through the cafeteria.

AI tool hunter.

See you next time.

Appendix: 5 Prompt Templates (Copy Directly)

Prompt 1: Generate new lesson plan from old one

Based on my uploaded [year] [topic] lesson plan, add [year/competition] past paper cases, and create a new lesson plan. Retain my original teaching approach, but update all cases and practice problems.

Prompt 2: Adjust difficulty level

Rewrite this lesson plan into two versions: one for absolute beginners, and one for students who have studied once and want to advance. Give me a detailed [X]-lesson teaching plan for each version.

Prompt 3: Generate supporting exercises

Based on this [topic] lesson plan, generate [N] supporting practice problems. Difficulty from easy to hard. First [N1] problems are basic concept questions, last [N2] are adapted from [competition] past papers. Give me detailed solution approaches and reference code for each.

Prompt 4: Mistake analysis template

I uploaded a student’s mistake collection. Based on the teaching methods for similar problem types in my lesson plans, give targeted analysis for each mistake, and summarize the student’s [N] main weak points.

Prompt 5: Parent communication scripts

I have a [scenario] parent who [specific issue]. Based on the teaching philosophy in my lesson plans and the student’s progress data, give me a warm yet persuasive reply.