Writing recipes is an iterative process. You start with an idea, draft the ingredients, write out the steps, then realize you forgot to mention when to preheat the oven. You want to add a note about substitutions, simplify a wordy instruction, maybe convert everything to metric for a friend abroad. Each change is small, but together they add up to a lot of editing.
We built an assistant to help with this. It lives in the recipe editor, understands the fork.club recipe format, and can modify your recipe based on natural language requests. Ask it to add a vegetarian option, and it updates the ingredients and instructions. Ask it to translate to Spanish, and it does. Ask it to simplify a verbose recipe, and it strips away the unnecessary words while keeping the essential techniques.
We could have put the assistant everywhere: in the library view, on the recipe page, in search. But we started with the creation view because that's where it has the most impact. When you're writing or editing a recipe, you're already in the mindset of making changes. The assistant slots naturally into that workflow.
The creation view is also where precision matters most. A recipe isn't a casual document. Quantities, temperatures, and timing need to be exact. An assistant that makes invisible changes to your recipe would be dangerous. We needed to build something that keeps you in control.
When you send a message to the assistant, it receives three pieces of context:
The current recipe. Whatever is in the editor right now, including any changes you've made since you started editing.
Selected text. If you've highlighted a portion of the recipe, the assistant knows you want to focus on that section. Selecting 1 cup butter and asking what's a dairy-free substitute? gives much better results than asking about the whole recipe.
Your message. The natural language request explaining what you want.
This context is enough for the assistant to understand your intent and make targeted changes. It knows what you're working with, what you're focused on, and what you want to achieve.
The assistant can respond in two ways:
A clarifying question. If your request is ambiguous, the assistant asks for more information instead of guessing. Make this healthier could mean a dozen things. Lower fat? Less sugar? More vegetables? Smaller portions? The assistant asks what you have in mind.
A modified recipe. When the request is clear, the assistant returns an updated version of your recipe with the changes applied. Not a description of what to change, but the actual recipe with the modifications made.
This binary output keeps interactions focused. You either get a question that helps refine your intent, or you get a concrete result you can evaluate.
Here's where the design gets important.
When the assistant returns a modified recipe, you don't just see the new text. You see a diff: a side-by-side comparison showing exactly what changed. Added text is highlighted. Removed text is marked. Unchanged sections are clearly unchanged.
# Chocolate Chip Cookies
- 2 1/4 cups all-purpose flour
- - 1 cup butter, softened
+ - 1 cup vegan butter, softened
- 3/4 cup sugar
- 2 eggs
- - 2 cups chocolate chips
+ - 2 cups dairy-free chocolate chips
This matters because details matter. If you asked the assistant to make a recipe dairy-free and it changed butter to margarine but missed the chocolate chips, you need to catch that. Scanning a diff is faster than reading two complete recipes and mentally comparing them.
The diff view also builds trust. You can see exactly what the assistant did, no hidden changes, no surprises. If something looks wrong, you reject the changes and try again with a clearer request.
Recipes rarely come out perfect on the first try. You might accept the assistant's dairy-free changes, then realize you want to adjust the baking time, then decide to add a note about which vegan butter brand works best.
Each subsequent message includes the full conversation history:
This lets the assistant understand the trajectory of your edits. It knows what you started with, what you've changed, and what direction you're heading. Asking actually, make it gluten-free too after already making it dairy-free gives intelligent results because the assistant remembers the context.
The assistant knows the fork.club recipe format intimately. It understands that # creates titles, - creates ingredient lists, numbered items create instruction steps, and > creates notes. It knows about definition lists for recipe facts, scalable amounts in brackets, and the semantic meaning of combining ingredients with instructions into steps.
This isn't just pattern matching. The assistant generates recipes that follow our format correctly, with proper structure and syntax. When you ask it to Add a tips section, it creates a proper blockquote at the end. When you ask for a sauce sub-recipe, it creates a new step with its own ingredients and instructions.
The result is that assistant-generated content looks native. It fits naturally into your recipe without manual cleanup.
An assistant that can read and modify recipes based on natural language opens up possibilities we're only beginning to explore. Here are some of the use cases that work well today:
You open the fridge and find chicken thighs, some wilting spinach, half a lemon, and a jar of olives. You don't have a recipe in mind, just ingredients that need to be used.
You: I have chicken thighs, spinach, lemon, and olives. What can I make?
Assistant: Returns a Mediterranean-style braised chicken recipe using those ingredients, with quantities estimated for the typical amounts those items come in.
Lemon Chicken Thighs with Spinach and Olives The assistant creates a complete recipe from scratch, structured and ready to cook.
Sometimes you just want a starting point. You know what you want to cook, but you don't have a recipe in front of you. Instead of searching the web, just ask.
You: I want a chilli con carne recipe
Assistant: Returns a complete recipe with ingredients, quantities, and step-by-step instructions. A solid foundation you can cook as-is or tweak to your taste.
Chilli con carne This is particularly useful when you want to quickly get cooking, or when you need a baseline to iterate on. The generated recipe follows fork.club format, so you can immediately start adjusting quantities, adding notes, or asking follow-up questions like make it spicier or add a vegetarian version.
You found a perfect cookie recipe, but it calls for buttermilk and you only have regular milk.
You: [selects "1 cup buttermilk"] What can I use instead?
Assistant: Suggests "1 cup milk + 1 tbsp lemon juice (let sit 5 minutes)" and offers to update the recipe.
-- 1 cup buttermilk
++ 1 cup milk mixed with 1 tbsp lemon juice (let sit 5 min)
Substitution knowledge built into the workflow, right where you need it.
Your sister is coming for dinner and she's recently gone vegan. You have a family lasagna recipe you want to make.
You: Make this vegan
Assistant: Replaces ricotta with cashew cream, swaps beef for lentils, suggests nutritional yeast for the cheesy flavor, updates all relevant instructions.
-- 500g ground beef
++ 400g cooked lentils
-- 250g ricotta cheese
++ 250g cashew cream
-- 100g parmesan, grated
++ 100g nutritional yeast
The diff shows every change clearly. You can accept the cashew cream but decide to use store-bought vegan beef crumbles instead of lentils.
Your grandmother's recipe is in Spanish, but your friends in the UK would love to have it.
You: Translate to English
Assistant: Returns the complete recipe in English, including ingredient names, instructions, and any notes. Keeps the structure intact.
++# Grandma's Apple Pie
--# Tarta de Manzana de la Abuela
++ 6 medium apples, peeled and sliced
-- 6 manzanas medianas, peladas y en rodajas
++ 1 cup sugar
-- 1 taza de azucar
The recipe stays in your library in Spanish. You can copy the English version to share.
Many recipes call for store-bought components. Add 1 cup marinara sauce. But what if you want to make everything from scratch?
You: [selects "1 cup marinara sauce"] Replace with a homemade version
Assistant: Expands that single ingredient into a complete sub-recipe: tomatoes, garlic, basil, olive oil, with instructions for making a quick marinara that yields about a cup.
-- 1 cup marinara sauce
+## For the marinara
+
+- 400g canned San Marzano tomatoes
+- 2 cloves garlic, minced
+- 2 tbsp olive oil
+- Fresh basil leaves
+
+1. Heat olive oil, saute garlic until fragrant
+2. Add tomatoes, simmer 15 minutes
+3. Season and add basil before using
Your simple weeknight recipe becomes a from-scratch masterpiece when you have the time.
You imported a recipe from a blog and it came with paragraphs of explanation woven through the instructions. "Now, this is the part where my grandmother would always say..."
You: Simplify this, keep only the essential steps
Assistant: Strips the storytelling, reduces multi-sentence instructions to clear directives, removes redundant information while preserving every necessary technique and timing.
-1. Now, this is really important and I cannot stress this enough
- from my years of experience - you need to preheat your oven.
- I always set mine to 350°F, which is what my grandmother taught
- me back when I was just learning to bake in her tiny kitchen...
+1. Preheat oven to 350°F
The recipe goes from 800 words to 200 words without losing anything you need to cook it.
Your British friend sent you a recipe with weights in grams and temperatures in Celsius. Your American kitchen uses cups and Fahrenheit.
You: Convert to US measurements
Assistant: Changes 200g flour to approximately 1 2/3 cups, 180°C to 350°F, and so on throughout the recipe.
-- 200g flour
++ 1 2/3 cups flour
-- 100g sugar
++ 1/2 cup sugar
-- 225g butter
++ 1 cup (2 sticks) butter
-1. Preheat oven to 180°C
+1. Preheat oven to 350°F
Or the reverse: you want to share an American recipe with European friends.
A recipe from a professional cookbook assumes you know what "deglaze" means and how to julienne vegetables. You're cooking with your kids this weekend.
You: Rewrite for beginner cooks, explain techniques
Assistant: Expands "deglaze the pan" to "add the wine to the hot pan and scrape up the brown bits from the bottom with a wooden spoon." Adds notes explaining unfamiliar terms.
-3. Deglaze the pan with white wine
+3. Add the wine to the hot pan and use a wooden spoon to
+ scrape up all the brown bits stuck to the bottom - this
+ is where all the flavor is!
Or the opposite: you're an experienced cook and want to strip out the explanations you don't need.
You want to prep for a dinner party the day before. The recipe doesn't mention what can be done ahead of time.
You: Add make-ahead instructions
Assistant: Analyzes the recipe and adds notes about which components can be prepared in advance, how to store them, and how to finish the dish when ready to serve.
5. Let the pie cool completely before serving
+> Make ahead: The pie dough can be prepared up to 3 days in
+> advance and refrigerated, or frozen for up to 1 month. The
+> assembled pie can be frozen unbaked for up to 2 weeks - add
+> 10 minutes to the baking time when cooking from frozen.
Party planning built into the recipe itself.
You have a great slow-cooker recipe but want to make it in your Instant Pot. Or a stovetop recipe you want to adapt for the air fryer.
You: Convert this to Instant Pot
Assistant: Adjusts cooking times, liquid amounts, and technique for pressure cooking. Adds notes about natural release vs. quick release.
-- 4 cups chicken broth
++ 2 cups chicken broth (less liquid needed for pressure cooking)
-3. Cook on low for 8 hours or high for 4 hours
+3. Pressure cook on high for 25 minutes, then natural release
+ for 10 minutes before quick releasing remaining pressure
The same dish, adapted for the equipment you have.
The assistant today is focused on the creation view, but the patterns we've established (context awareness, diff-based review, iterative refinement) could extend further. Imagine asking questions while viewing a recipe: "Can I use olive oil instead of butter here?" without entering edit mode. Or batch operations across your library: Find all my recipes that use ingredients in season.
For now, we're learning from how people use the assistant in the editor. Every interaction teaches us what requests are common, where the assistant succeeds, and where it needs improvement. The foundation is solid: an AI that understands recipes, respects your control, and makes editing feel like a conversation.
The recipe is still yours. The assistant just helps you make it exactly what you want.
Author: Jorge Bastida
Published: January 8, 2026
RFD: #0010
If you'd like to discuss this RFD, share your thoughts, or simply chat about it, feel free to reach out to me - To stay up to date with the project's development, you can follow me on X