How-to

The best ChatGPT prompts for adapting a recipe you already trust

Published August 30, 2026 Last checked August 30, 2026

Start with a recipe, not a blank page

ChatGPT is more useful as a recipe editor than as an untested recipe author. Paste a recipe you already trust, ask for one well-defined change, and make the model preserve everything that does not need to move. That gives you a baseline you can compare against instead of a confident-looking recipe with no history.

This guide is based on analysis of prompt structure and authoritative cooking guidance, not on cooking every output. OpenAI’s own recipes and cooking examples include substitutions, scaling, conversions, and dietary changes. Those are sensible uses, but the output is still a draft. For safe temperatures, preservation, allergies, and narrow baking ratios, verify the answer against an appropriate trusted source before cooking.

Here is the reusable prompt shape:

Adapt the recipe below to [one goal]. Keep [what must not change]. My constraints are [equipment, ingredients, servings, diet]. Return a complete, self-contained ingredient list and numbered method, then a short change log. Include every required temperature, time, and doneness cue; if the source omits one, flag the gap instead of writing “as originally directed” or inventing it. Recipe: [paste recipe]

Asking for ingredients and method together makes inconsistencies easier to spot.

What happened when we used a precise base recipe

On August 30, 2026, we sent five fixed recipe prompts through the OpenRouter API to OpenAI GPT-5.4, Anthropic Claude Sonnet 5, and Google Gemini 3.7 Flash. This was an output-only comparison against stated checklists and trusted references. We did not cook the recipes, and the results describe these named model versions through the API — not the ChatGPT, Claude, or Gemini consumer-app interfaces.

The strongest pattern supported the template above. Given a complete four-serving rice recipe, a 10-serving target, and an explicit output format, all three models applied the 2.5× quantities, recommended larger cookware, and treated cooking time as something to reassess rather than multiply by 2.5. Given a trusted-style muffin formula and a dairy-free-but-not-vegan constraint, all three replaced the milk and butter, kept the eggs, and maintained the basic formula.

But one response exposed an important prompt gap. Claude Sonnet 5 returned a method ending with “bake as originally directed” even though the supplied base method contained no oven temperature or time. The ingredient adaptation was useful; the recipe was not self-contained. That is why the prompt above explicitly requires every time, temperature, and doneness cue, plus a missing-information check. A model can follow the main transformation well and still leave one operational hole.

10 prompts worth saving

1. Scale a dinner to a new serving count

Adapt this recipe from 4 servings to 7. Recalculate every quantity, but do not assume cooking time scales linearly. Keep the same cooking method. Flag any pan-capacity, batch-cooking, seasoning, or doneness changes I should handle separately. Return the full recipe and a change log. [Paste recipe]

For the underlying arithmetic and exceptions, see how to scale a recipe.

2. Replace one missing ingredient

Replace the [ingredient] in this recipe using only [ingredients I have]. Preserve the dish’s role for that ingredient: flavor, moisture, fat, acidity, binding, or leavening. Update every affected step. Give me up to two options and explain the tradeoff in one sentence each. [Paste recipe]

A substitute that matches flavor but not function can fail in baking. Our ingredient substitution guide explains how to identify that role.

3. Make a recipe dairy-free

Make this recipe dairy-free. Do not change ingredients that are already dairy-free. For each replacement, state whether it changes flavor, browning, texture, or cooking behavior. Check both ingredients and method for dairy references. Return a complete, self-contained recipe with explicit time, temperature, and doneness cues. If the source omits one, flag it rather than referring me back to the original. [Paste recipe]

4. Make a recipe vegetarian without flattening it

Make this recipe vegetarian while preserving its savory depth, texture, and approximate protein role. Replace meat-based stocks, sauces, garnishes, and cooking fats as well as the obvious main ingredient. Do not claim the result is nutritionally equivalent. Return the full recipe and list the flavor-building changes. [Paste recipe]

Our guide to making a recipe vegetarian treats it as a flavor-and-structure edit, not a deletion.

5. Convert cups to grams without fake precision

Convert this recipe’s volume measurements to grams where a dependable ingredient-specific conversion is available. Keep teaspoons for very small seasonings if that is more practical. Do not use one generic cups-to-grams conversion for different ingredients. Mark any estimate clearly and keep the original amount beside it. [Paste recipe]

Conversions must be ingredient-specific; see our cups, grams, Fahrenheit, and Celsius guide.

6. Simplify a weeknight recipe

Reduce this recipe to no more than [number] active minutes and [number] pans. Preserve the main flavor profile. Prefer combining compatible steps and removing optional garnishes over replacing core ingredients. Tell me which quality tradeoffs you made. Return a complete revised recipe. [Paste recipe]

7. Adapt for the equipment you actually own

Adapt this recipe for [air fryer / convection oven / slow cooker / one skillet]. Keep the ingredient list as close as possible. Explain changes to batch size, airflow, liquid, temperature, and doneness checks. Do not invent a precise time if the source does not support one; give a checking window and observable cues. [Paste recipe]

Equipment conversions change heat transfer, not just a temperature number. Treat any time as a checkpoint, especially for meat and poultry.

8. Use what is already in the fridge

Adapt this recipe using only the ingredients below, plus salt, pepper, oil, and water. Keep the original technique where possible. If a missing ingredient has no reasonable substitute, say so instead of forcing one. Pantry: [list]. Recipe: [paste recipe]

9. Make a baking recipe less sweet

Analyze this baking recipe and propose the smallest reduction in sugar likely to meet my goal: [slightly less sweet / much less sweet]. Explain sugar’s role here in moisture, spread, browning, tenderness, and structure before changing it. Preserve the original pan size. Return one conservative revision, not several speculative versions. [Paste recipe]

Sugar does more than sweeten, so compare the suggestion with a tested recipe of the same type.

10. Audit a generated recipe before saving it

Audit this recipe against the trusted reference below. Check ingredient-to-step consistency, yield, units, pan capacity, cooking sequence, and food-safety temperatures. Separate definite inconsistencies from judgment calls. Then return a corrected recipe, changing only what the evidence supports. Generated recipe: [paste]. Trusted reference: [paste or link]

Verify temperatures against the USDA’s safe minimum internal temperature chart. Use tested USDA or extension recipes for canning and fermentation.

What a good answer should contain

Read the output beside the original. Fluency is not evidence that the adaptation was tested.

Move the useful result out of the chat

Copy the final recipe into Sous Chef’s text import on iPhone to turn it into structured ingredients and steps. Keep scaling, substitution, unit, and dietary edits on that recipe instead of creating disconnected chat versions.

With the Sous Chef connector, a generated recipe can render as a card in ChatGPT or Claude. Saving remains explicit; rendering alone does not add it to your library. Either route keeps the original available while edits can be reviewed or undone.

Sous Chef validates the structure of edits — that referenced ingredients and steps exist and the change is well formed. It does not certify taste, nutrition, allergy safety, or food safety. Keep using trusted sources for those judgments.

Common questions

Should I ask ChatGPT to create or adapt a recipe?

Adapt when you can. A trusted recipe gives you a tested baseline, recognizable ratios, and a source to compare against. Generation from scratch can still supply ideas, but it needs a more thorough audit.

Can ChatGPT safely adapt a recipe for an allergy?

It can suggest a draft, but it cannot inspect your product labels, kitchen, or cross-contact risk. Verify every ingredient and preparation surface yourself.

Can Sous Chef check whether the adapted recipe will taste good?

No. Sous Chef can keep the recipe structured and apply well-formed edits, but taste and culinary judgment still belong to the cook.

What is the easiest way to save a ChatGPT recipe on iPhone?

Copy the final recipe text, open Sous Chef, choose the text-import route, and paste it. Review the parsed ingredients and steps before keeping it in your library.