🍽️ Tasting Menu

The best way to understand what good prompting looks like is to see the actual output. These are real examples — prompt in, result out, notes on why it worked.

From the recipe: Cold Email That Gets Replies →
The prompt 
Write a cold email to a senior product manager at Notion.

Purpose: I want a 20-minute call to discuss how they approach onboarding.

About them: They recently posted about reducing time-to-value in B2B SaaS on LinkedIn.

About me: I'm a UX researcher studying onboarding patterns — no sales pitch.

Constraints:
- Max 120 words
- No "I hope this finds you well"
- Lead with what I know about their work
The output 
Subject: Your take on time-to-value in onboarding

Hi [Name],

Your LinkedIn post about reducing time-to-value in B2B SaaS hit close to home — I've spent the last year studying exactly that.

I'm a UX researcher looking at onboarding patterns across 20+ tools. Not selling anything — I'm writing up findings and Notion keeps coming up as a model worth studying.

Would you be open to a 20-minute call? I'd share what I'm seeing across the industry in exchange for your perspective.

[Your name]
💡 Why it worked

Note how it leads with their work, not yours. The 'no sales pitch' line does real work — it removes the main objection before they think it.

From the recipe: Draft → Critique → Revise →
The prompt 
Draft a short LinkedIn post announcing that I'm joining a new startup as Head of Design. Tone: genuine, not humblebraggy. Audience: former colleagues and hiring managers. Max 150 words.
The output 
After 4 years at [Company], I'm making a move.

Starting next month, I'll be Head of Design at [Startup] — a small team building [what they do].

Honestly, the decision came down to two things: the problem they're solving (which I think actually matters) and the chance to build a design practice from scratch.

Looking forward to what comes next. And grateful for the work and people that got me here.

[Name]
💡 Why it worked

The prompt's 'not humblebraggy' constraint is doing the heavy lifting. Without it, you'd get 'thrilled and honoured to announce this exciting new chapter.'

From the recipe: Teach Me Like →
The prompt 
Explain how transformer neural networks work. Teach me like I'm a competent software engineer who understands data structures and algorithms but has no ML background. Use analogies. Stop after each major concept and check if I'm following.
The output 
Let's start with the core problem transformers solve: understanding sequences where order and context matter.

Think about the sentence "The bank was steep." vs "The bank was closed." Same word, different meaning — and you only know which from context.

Old approaches (like RNNs) read left to right, word by word — like reading with a sticky note covering everything you haven't read yet. By the time you reach "bank" in a long sentence, the early context is fuzzy.

Transformers throw that out. Instead, every word "looks at" every other word simultaneously and scores how relevant each one is to understanding it. This is called **attention**.

Does that distinction between sequential vs. parallel context-reading make sense before we go further?
💡 Why it worked

The 'stop and check' instruction produces the conversational back-and-forth that makes explanations stick.

Want to try these yourself?

Each example comes from a full recipe in the Recipe Book — with the prompt template, remix variations, and notes.

Browse the Recipe Book →