Use Prompts Like Recipes for Designing Full Conversations with AI
Discover the "Recipe Prompting" technique that helps you design conversations and get exactly what you need from ChatGPT, Claude, or any AI assistant.
Have you ever stopped to think that we're the first generation in human history who can talk to machines in our own language?
Before ChatGPT, communicating with computers meant speaking their language, not ours. Only people who knew Python, C++, JavaScript, Turbo Pascal, and other programming languages had access to this privileged club. The rest of us were locked out.
Well, the world has changed. Today anyone can communicate with a computer in plain English (or their language of choice!), though it often feels like we're still speaking different languages. We ask for one thing and sometimes (more often than we'd like to admit) the AI gives us something completely different.
Imagine asking someone who has never cooked before to make lasagna by simply saying "Make lasagna." The result? It might technically be edible, but it definitely won't be the delicious, layered Italian masterpiece you had in mind. This is exactly why humans invented recipes in the first place!
For some time now, I've been using a prompting technique that's all about exactly this: giving detailed "recipes" to AI. I call it Recipe Prompting.
I chose this name because recipes are the most familiar algorithms we encounter in daily life – they're just step-by-step instructions that tell us how to reach a delicious result. Most people would run away from "programming an algorithm," but everyone understands following a recipe.
And that's exactly what we do with this technique – transform our prompts into recipes that AI can follow step by step, instead of asking for something vague and crossing our fingers hoping it somehow reads our mind.
I actually thought about calling it "Algorithmic Prompting," but then I imagined that phrase stamped across a 300-page corporate manual that no one would ever read. So I stuck with the more appetizing recipe concept.
Note: I'm not claiming I invented this approach – I've probably just combined various prompting techniques under a simple, catchy name. I doubt I'm the first person to use this phrase. But hey, I can always rebrand later if needed! 😊
What is recipe prompting?
When we use ChatGPT, Claude, or any other AI assistant, most of us simply type something like "Help me write a proposal my boss can't refuse" or "Create a plot where a robot falls in love with a blender" and then we expect the AI to understand exactly what we want. Sometimes it works! But far too often, we get responses that are nothing like what we had in mind – just like when you cook without following a recipe.
I'll confess something: I love cooking but rarely follow recipes precisely. A bit more of this, a bit less of that, what if I throw in this random ingredient from the fridge? This freestyle approach usually works out fine... except for that one time I improvised with a butter fish dish that my family has collectively agreed never to speak of again.
But when it comes to baking (spoiler: definitely not my thing), everything changes. You can't just casually add half a kilo more flour or decide to skip three eggs from the recipe. If you do that, instead of a delicious cake, you'll end up with a brick suitable for construction work.
It's exactly the same with AI – if you want precise, useful results, you need to provide clear, detailed instructions just like you would follow a baking recipe.
The Fine Art of Prompt Pastry
Recipe prompting consists of three essential ingredients:
Clear step-by-step instructions: Like a good recipe, you tell the AI exactly what to do in sequence, leaving nothing to chance.
Adaptive decision branches: You include "if-then" scenarios to handle different situations - like "if the dough is too dry, add a splash of milk" or "if the user is vegetarian, don't suggest meat recipes."
Iteration loops: You build in feedback cycles by asking the AI to repeat steps until the result meets your standards, such as "keep adjusting the text until it's fewer than 100 words"
Here's a before-and-after example that shows the dramatic difference:
Instead of asking:
Help me plan my meals for the weekYou could try:
Let's create a meal plan for my week. Follow these steps:
1. Ask me one question at a time and wait for my response before continuing:
- First, ask if I have any allergies or dietary preferences
- Then, ask how much time I have available to cook each day
2. If any of my responses seem unusual or unclear (like saying I'm allergic to something strange or that I cook 25 hours a day), ask for clarification before continuing.
3. With the information I've given you, create a simple meal plan for 5 days (lunches and dinners only).
4. Ask me if I like the plan or if I want to make specific changes.
After my adjustments, present the final plan in a weekly calendar format.With the traditional prompt, you'd likely get generic, one-size-fits-all suggestions like "Monday: Pasta with vegetables, Tuesday: Roasted chicken..." But with the Recipe Prompting approach, you receive a meal plan perfectly tailored to your specific situation, preferences, and time constraints.
Here's what's fascinating: you're essentially using programmer thinking without realizing it! You're applying the same logical structure that software developers use, but in such a natural, conversational way that anyone can do it without any technical knowledge or coding skills!
Why Recipe Prompting Works So Well
Have you noticed that the clearest instructions you've ever received likely came from an IKEA assembly manual or a well-written recipe? That's no coincidence. When humans need to communicate exactly how to do something, step-by-step instructions with decision points are universal. We've been using "algorithms" long before computers existed - we just called them recipes, manuals, and how-to guides!
Now let's explore why this approach works so well with AI...
Recipe prompting works for three fundamental reasons:
AI doesn't read minds or between lines: Humans are amazingly good at inferring what someone wants even when instructions are vague. We fill in gaps using context and shared experiences. AI simply can't do this (yet). It processes exactly what you tell it - nothing more, nothing less. Recipe prompting makes your intentions crystal clear at each step, eliminating those gaps of understanding.
AI thrives on structure: While modern AI can create the illusion of understanding casual conversation, underneath they're still pattern-matching machines that perform best with clearly structured information. They're like that friend who needs things explained step-by-step rather than in a rambling story.
AI has conversation amnesia: During longer conversations, AI can "forget" important context from earlier messages. A recipe format provides a consistent framework they can follow from beginning to end, keeping everything organized even as the conversation grows.
When Should You Use Recipe Prompting?
This technique is like a specialized kitchen tool - not needed for everything, but invaluable in specific situations:
When you need genuine personalization and want the AI to fully understand your context before generating a response.
When you need precise results and a generic, one-size-fits-all answer simply won't cut it.
When your task involves multiple steps and you need to guide the AI through a process, not just give a single instruction.
When you want control over the outcome by building in review points and adjustment opportunities.
If any of these scenarios sound familiar, recipe prompting will become your new AI superpower.
On the flip side, you don't need this approach when:
You just want a quick yes/no answer without complications.
You're asking straightforward factual questions like "What's the capital of France?"
You're just brainstorming or exploring ideas without a specific goal in mind.
The difference in results is dramatic. With a recipe, you know exactly what you'll get. Without one... well, you might end up with radioactive lasagna.
Real-World Applications (with a Fun Example)
This prompting method works for practically anything you need an AI to help with:
Creating content that sounds genuinely human and resonates with your audience
Running simulated job interviews tailored to specific roles and industries
Designing personalized plans for everything from fitness routines to learning paths
Generating creative ideas that actually match your specific vision
And countless other possibilities...
Let's look at a simple example that should make you go, "Aha! Now I get what Germán is talking about!"
Behold the mighty Band-name generator
I have this weird fascination with band names that follow a specific format: a number greater than 1 followed by an animal name in plural form, all in English – things like "22 Bears," "35 Whales," or "47 Porcupines" (don't bother Googling these, I made them up).
If I simply ask ChatGPT to "give me band names," it'll likely generate perfectly fine suggestions like "Neon Shadows," "Quantum Pulse," or "The Germán Martínez Experience." Not bad (especially that last one), but they're nothing like the specific format I'm looking for.
Now look at this Recipe Prompt:
The Band Name Generator!
Let's create a band name following this precise procedure:
1. Ask me for an animal. If what I tell you isn't actually an animal, politely ask again until I provide a valid animal name.
2. Ask me for a number greater than 1 but less than 999. If I don't give you a number, or if the number is 1 or less or greater than 999, ask again until I provide a valid number.
3. Create a band name by combining the number and the animal name in plural form (in English). Example: "37 Monkeys"Want to see this in action? I've actually created a custom GPT with this exact prompt and named it "BandNameology" – you can try it yourself !
When you chat with it:
It won't be fooled (hopefully): Try telling it "table" is an animal, and it'll politely ask you for a real animal instead.
It follows the procedure precisely: It won't get distracted or suddenly start suggesting band logos instead of names. It stays on track!
You'll get exactly what you're looking for: A perfect band name in exactly the format you wanted!
This is a simple example, but it demonstrates why this method is so powerful. Instead of just giving instructions, you're designing an entire conversation flow.
It’s not about giving detailed instructions, it’s about designing a structured conversation. And even if we’re just talking about band names, this approach can be used in many other (more important) situations.
You can design conversations to solve complex problems –you’d be the MacGyver of prompts (just missing the ’80s mullet and Swiss army knife)
Creating Your Own Recipe Prompts
After experimenting with this technique for months, I've discovered it's incredibly versatile. You can apply it to virtually anything – from highly structured analytical tasks to guiding creative processes. It works for technical writing, creative brainstorming, and even design thinking workshops!
Start With Your End Goal
Before you write a single word of your prompt, take a deep breath (seriously – take another one), and get crystal clear about exactly what successful output looks like.
There's a world of difference between asking "help me with my podcast" and saying "I need a structured script for a 20-minute technology podcast that will keep my audience of non-technical professionals engaged and build credibility for my consulting business."
The clearer your vision, the more precise your Recipe Prompt can be.
Map the Conversation Journey
What makes Recipe Prompting truly powerful is that you're designing the entire conversation flow, not just the initial request:
What background information does the AI need before it can start generating useful content?
At which critical points should you review progress and potentially redirect?
How should the AI handle confusion or unclear responses from you?
This approach lets you guide the AI precisely where you want it to go, creating a reliable path rather than hoping the AI somehow reads your mind.
What if…?
My favorite aspect of this technique is how it handles different scenarios with "if-then" branches:
"If I like this direction, develop it further by..."
"If I'm not satisfied, ask me specifically what aspect I want to change..."
"If my answer is vague or ambiguous, ask follow-up questions about..."
These conditional branches transform what would be a static, one-way information dump into a dynamic, adaptive conversation that responds intelligently to your feedback.
Programming Tricks Anyone Can Use (No Coding Required!)
Don't worry - you absolutely don't need any coding knowledge to use Recipe Prompting! But I want to show you how some basic programming concepts are actually just formalized versions of everyday thinking patterns you already use naturally. At their core, programming and recipes are both just ways of giving clear step-by-step instructions. 😊
"If This, Then That" Branches
In programming this is called an "if-then-else statement," but we use this logic every day: "If it's raining, bring an umbrella. If not, wear sunscreen." It's just making different choices based on conditions.
In your prompt:
After showing me each potential blog article idea:
- If I respond with "interested," develop a complete outline with introduction, three main sections, and conclusion.
- If I respond with "not convinced," ask me specifically what I don't like (topic, approach, or style) and then generate three new ideas modifying only that aspect.This creates a conversation that adapts intelligently to your feedback instead of just giving you static responses.
Do This Several Times
Programmers call this a "for loop," but it's just a fancy way of saying "do this exact same thing X times" - like making multiple cookies from the same recipe.
In your prompt:
Generate 5 titles for articles about AI applications in different industries, following this exact format for each one:
- Each title must begin with "How AI is Transforming"
- Followed by a different industry name (healthcare, finance, education, etc.)
- Ending with a specific benefit or advantage
Follow this precise format for all 5 titles, using 5 different industries.This is perfect when you need the AI to generate multiple variations of something using the same pattern, saving you from having to repeat yourself.
Keep Going Until…
Programmers call this a "while loop," but it's just like saying "keep adding salt until the soup tastes right" - continue until a condition is met.
In your prompt:
Help me create a slogan for my plant business following this process:
1) Show me an initial slogan draft.
2) Ask what specific aspects I like and don't like.
3) Create an improved version based on my feedback.
4) Repeat steps 2 and 3 until I respond with the exact phrase "this is perfect."This is ideal when you don't know in advance how many iterations will be needed to achieve the result you want - you're building in a refinement process.
Conversation Menu
In programming, this is called a "switch statement" or "case," but think of it as a restaurant menu where each selection leads to a completely different experience.
First, ask me what type of content I want to create (blog, video, or presentation). Then follow these instructions based on my choice:
If I choose blog: Ask me the ideal length (short, medium, or long). Then create an outline with introduction, at least 3 subtitles, and conclusion.
If I choose video: Ask me whether it's educational or promotional. Then create a script divided into scenes with approximate timing for each segment.
If I choose presentation: Ask me whether it's for clients or for my team. Then structure the slides with numbered sections, key content points, and speaker notes.This approach is perfect when you want to offer several distinct paths through a conversation, each with its own specialized process. It's like creating a choose-your-own-adventure story where each choice leads somewhere meaningful.
Give the AI a Memory
Programmers call these "variables," but it's simply asking the AI to remember something important throughout your conversation.
After asking about my industry, remember to use this information to customize all examples, analogies, and terminology throughout our entire conversation.This creates continuity and personalization throughout longer interactions, ensuring the AI doesn't "forget" crucial context that shapes everything that follows.
Create Reusable Conversation Shortcuts
In programming these are called "functions," but think of them as mini-recipes you can call upon whenever needed without rewriting the instructions.
When I type the command "SUMMARIZE:" followed by any text, always respond with:
- One sentence capturing the core idea in plain language
- The three most important takeaways in bullet points
- Any action items or next steps
This way, I can get consistent, structured summaries throughout our conversation without repeating these instructions.This technique is invaluable for creating consistent structures you'll use repeatedly in a conversation, saving you from typing the same instructions over and over.
The beauty of all these techniques is that you don't need to use technical programming terms like "if-then-else" or "while loops" in your actual prompts. Just use everyday language to explain the process you want the AI to follow, just as you would explain it to a person.
I promise that once you start using these patterns, your AI conversations will achieve a level of clarity and precision that previously seemed possible only for programmers. You'll have computational thinking superpowers without writing a single line of code (though knowing a bit of Python never hurts!).
A Powerful New Tool in Your AI Toolkit
You know what's most exciting about technology evolution? The moment when something that was once only accessible to specialists suddenly becomes available to everyone.
That's exactly what's happening with Recipe Prompting: you now have the power to direct AI with the precision of a programmer, but using everyday language that feels natural. No coding required!
Most people still use AI by tossing in vague requests and crossing their fingers, hoping the AI somehow reads their mind. Sometimes it works, often it doesn't. With Recipe Prompting, you're flipping that relationship entirely - you're designing intentional conversations from the start that lead precisely where you want them to go. It's the difference between hoping for the best and creating a reliable path to results.
The Results Speak for Themselves
Here's a quick challenge for you:
Pick something you need to accomplish with ChatGPT or Claude today
Before defaulting to your usual approach, invest just 2 minutes in designing a simple "recipe" prompt
Compare the results side-by-side with your usual approach
Feel slightly amazed, then make it a habit
Don't worry if your first attempts aren't perfect - just like learning to cook, your early "recipes" might be a bit messy. But I promise you'll improve rapidly, and the difference in results will make it worth the small learning curve.
Before You Go
I hope this technique helps you unlock the full potential of tools like ChatGPT and Claude. It's transformed how I work with AI, and I'm constantly finding new ways to apply it.
If you'd like to see Recipe Prompting in action, try out my custom GPT called "BandNameology" that creates band names following the exact pattern we discussed. While we're at it, stay tuned for my upcoming post about creating your own custom GPTs from scratch!
Here's a little homework assignment if you're up for it: Share your best recipe prompt in the comments section below. The most creative and original one wins bragging rights (and maybe a surprise!).
Until next time!
G
Hey! I'm Germán, and I write about AI in both English and Spanish. This article was first published in Spanish in my newsletter AprendiendoIA, and I've adapted it for my English-speaking friends at My AI Journey. My mission is simple: helping you understand and leverage AI, regardless of your technical background or preferred language. See you in the next one!