Decoded: Google's 70-Page PhD-Level Prompting Secrets (We Did The Hard Work For You)

Want unlimited AI, ASAP?

Get up to 6 months of Plus plan + unlimited AI free!

Launch and scale your startup faster with Notion.

Visit the Notion for Startups page to get the offer.

Get Notion AI for free

How to talk to AI like a pro, even if you're just trying to get it to write your grocery list or fix a typo in your email

🧠 INTRO: You’re Already Talking to AI — But Are You Doing It Right?

AI isn’t magic. It’s just really good at predicting the next word based on how you tee it up. If your prompts feel like shouting into the void, you’re not alone — most people are winging it.

Good news? Prompt engineering is learnable, and once you get the hang of it, it’s like having a cheat code for your job, side hustle, or late-night Google rabbit holes.

• 🤖 AI doesn't "know" things — it guesses well based on your prompt.

• 💬 Garbage in = garbage out. Better prompts = smarter outputs.

• 🛠 You don’t need to “learn AI,” you just need to learn how to talk to it.

• 📉 You’re not bad at AI — the default prompt was just mid.

• 🧙 Think of prompts like spells — specific incantations lead to specific magic.

🎯 Example Prompt:

📦 What the Heck is Prompt Engineering?

Prompt engineering is the art of telling AI exactly what you want in a way it understands. You’re not just asking a question — you’re shaping the answer.

This isn't about using big words or sounding technical. It’s about being clear, clever, and a little bossy when needed. The AI is your intern — give it instructions that even a coffee-deprived overachiever can follow.

• 🗣️ Prompt = your instructions. Prompt engineering = instructions that actually work.

• 🧰 You don’t need fancy tools, just a better way to phrase what you want.

• 📚 It’s not writing for AI, it’s writing with AI.

• 🐶 Training a dog and prompting an AI? Same vibes.

• ✍️ Think like a screenwriter — clear stage directions lead to better scenes.

🎯 Example Prompt:

⚙️ LLM Config Settings — AKA, the Dials Behind the Magic Curtain

Okay, you’ve written your prompt. But what if the AI’s tone is too dry or too goofy? That’s where these behind-the-scenes settings come in — they're like seasoning on your steak: too much or too little, and things get weird.

Tweak temperature for creativity, top-K/top-P for randomness, and output length to avoid long-winded rambles. Start small and dial it in like you’re fine-tuning your Spotify playlist.

• 🌶️ Temperature = how spicy and random the AI’s answer is.

• 🎯 Top-K/Top-P = how focused or diverse the output will be.

• 🛑 Output length keeps it from writing a novel when you just want a tweet.

• 🧪 Don’t be afraid to experiment — these knobs are meant to be twisted.

• 🧭 Default settings are fine… if you want “default” results.

Absolutely — here’s a prompt that not only demonstrates clear instructions, but also subtly invites tuning those LLM config dials behind the curtain:

🎯 Example Prompt:

✅ Why it works:

• Role: "snarky food critic"

• Task: "3-sentence review"

• Tone guidance: "witty, slightly rude, creatively over-the-top"

• Constraints: "under 100 words"

• Suggested LLM settings: cranked up temperature for spicy takes, high top-p for creativity, capped max tokens to avoid a monologue

🧠 Think of it like asking your AI intern to write Yelp reviews after three martinis — with just enough control to keep it printable.

Want one that dials things down for more serious or professional content?

🧪 Prompting Techniques That Work (At Home or Work)

Prompting isn’t one-size-fits-all. Sometimes you give an example, sometimes you set a role, sometimes you just talk straight. These are your go-to methods depending on the job.

Whether you’re asking AI to write a blog post or plan a kid’s birthday party, there’s a prompt format that’ll make it smarter — and keep you from tearing your hair out.

• 💬 Zero-shot = no example, just vibes.

• 🧪 Few-shot = show it how it’s done, then let it copy you.

• 👔 System prompts = set the ground rules before it speaks.

• 🎭 Role prompts = cast your AI in a role (chef, coach, CEO).

• 🗂 Contextual = drop some backstory so it doesn’t start from scratch.

You got it — here’s an example prompt that shows off prompting techniques that work, using a few-shot + role + contextual combo to get that chef’s kiss response:

🎯 Example Prompt (Few-Shot + Role + Contextual):

✅ Why it works:

• Role prompt: "witty career coach"

• Contextual prompt: "help recent grads feel better about job hunting"

• Few-shot technique: gives examples for AI to mimic

• Constraints: tone, length, emoji, purpose

This is the "Swiss Army knife" of prompts — structured but not stiff, and packed with cues that steer the AI without smothering it.

Want versions of this using zero-shot or system prompts next?

🧩 Advanced Magic (Still Totally Useable at Home)

Once you've mastered the basics, it's time to go Jedi. These next-level tricks help the AI think deeper, double-check its own work, or act like a mini research assistant.

Sound complicated? They’re really not. Think of these as your AI’s “study hacks” — ways to nudge it into thinking smarter, not harder.

• 🔙 Step-back: Ask a big-picture question before zooming in.

• 🧠 Chain of Thought: Force it to “show its work” like a math teacher.

• 🔁 Self-consistency: Ask multiple times and pick the most reliable answer.

• 🌳 Tree of Thoughts: Give it space to explore different ideas in parallel.

• 🔄 ReAct: AI thinks and takes action (like Googling for you).

🎯 Example Prompt (Step-back + Chain of Thought):

✅ Why it works:

• Step-back technique: Starts with a broad question ("big picture") before narrowing down

• Chain of Thought: Forces the AI to "show its work" and break the process into smaller, logical steps

• Task: Get deeper into design thinking, rather than just spitting out ideas

• Constraints: Clear, concise, actionable

It’s like training your AI to do a SWOT analysis with a focus on anxiety — getting that detailed, multi-layered response while keeping it smart and structured.

Would you like another prompt that uses Self-consistency or ReAct?

👨‍💻 Code Prompting 101: Not Just for Engineers

You don’t need to wear a hoodie and live in VSCode to write prompts for code. AI can whip up scripts, explain what that weird error means, or convert code between languages.

Perfect for busy folks automating tasks, fixing bugs, or trying to sound like they know what a “regex” is during meetings.

• 💻 Ask it to write scripts like it’s your junior dev.

• 🧠 Drop code in and ask, “Explain this like I’m 5.”

• 🔄 Translate Bash to Python like magic.

• 🧹 Use it for quick automation — rename files, clean up data, whatever.

• 🔐 Confidential stuff? Use private tools like Vertex AI instead of public chatbots.

🛠️ Step-by-Step Prompting Instructions (for Work or Home)

Alright, theory time is over — let’s build your prompt like a pro. These steps will work whether you’re writing an email, creating content, or building a product roadmap.

Treat this like your AI checklist. Skip a step and the answer might suck. Nail all seven? You’ve got yourself a killer prompt.

• ✅ Pick the task. One job per prompt.

• 🗣 Choose the voice — funny, serious, expert, sassy?

• 🧪 Start with zero-shot. Upgrade to few-shot if it fumbles.

• 🎭 Add roles and context to steer tone and content.

• 🎛 Adjust settings for creativity vs. accuracy.

• 🔁 Test, tweak, repeat — don’t settle for version 1.

• 💾 Save your best prompts like templates — future you will thank you.

Here's a fun yet practical example of code prompting using a clear, step-by-step approach:

🎯 Example Prompt (Junior Dev + Step-by-Step Instructions):

✅ Why it works:

• Junior dev approach: AI is treated like a helpful intern who explains code and suggests improvements

• Zero-shot: Starting with a simple prompt and letting the AI explain from scratch

• Step-by-step: "Walk me through it" makes the AI break down the code into manageable pieces

• Context: Directs the AI to focus on a very specific issue (handling missing values)

The result? A detailed breakdown of what’s happening in the code, with suggestions for improvement, like you’re getting help from a junior dev — no hoodie required.

Want a prompt that translates code between languages or tackles automation tasks next?

📈 Business & Industry Takeaways

🔍 External Factors

The demand for smarter, cheaper, and faster AI use is skyrocketing. Companies are under pressure to deliver value while navigating legal landmines and evolving tech. Prompt engineering offers control in a chaotic AI world.

• 🧾 Regulations are looming — precision prompts help reduce risk.

• 💼 Big orgs want fewer “AI oops” moments.

• 📉 Budgets are tight — smarter prompts = fewer model calls = lower costs.

• 🔍 Prompting lets teams get reliable results without retraining models.

• 🧠 Everyone’s asking: “How do we scale AI without breaking stuff?”

📊 Business Metrics

This isn’t just nerd stuff — better prompting impacts the bottom line. From faster content creation to more accurate insights, prompt engineering delivers ROI.

• ⚡ Faster time-to-output for teams.

• 📉 Reduced hallucinations = less human cleanup.

• 💵 Cost-savings on compute and tokens.

• 📊 Improved customer experience via tailored answers.

• 🧑‍💼 Non-technical teams are finally getting value from AI tools.

🚨 Significant Trends

Prompting is moving from fringe skill to mainstream job requirement. More workflows, more use cases, and yes — even more jargon.

• 📈 Everyone’s becoming a prompt engineer, whether they know it or not.

• 🧠 LLMs are moving from chat to enterprise workflows.

• 🖼 Multimodal (text + image + code) is the next big leap.

• 🛠 Prompt frameworks are now part of software stacks.

• 🔄 More automation means prompts are driving entire business functions.

📈 Business Initiatives

Enterprises are getting serious. They’re building prompt libraries, integrating tools like Vertex AI, and treating prompt quality like software quality.

• 🧰 Tooling up with prompt management platforms.

• 👩‍💻 Training teams on best practices (finally).

• 🧠 Investing in reusable prompt templates.

• ⚙️ Automating prompt testing and optimization.

• 🔬 Using prompt experiments to guide product development.

🔮 Forward-Looking Statements

The future’s looking prompt-filled. Think of it as the new programming — except you can do it in English, not Python.

• 🧠 AI will write its own prompts (Automatic Prompt Engineering is real).

• 🔍 Expect built-in evaluators that score your prompt quality.

• 📲 Prompts will soon drive CRM, ops, hiring, and more.

• 🧪 Prompt-based testing will be part of QA.

• 🚀 "PromptOps" might be a job title in 6 months.