If you publish recipes and want them to appear in AI answers, rich results, and voice responses, Recipe schema is the single most important thing you can add. It turns a wall of prose into a machine-readable object with named ingredients, ordered steps, times, yield, and ratings — exactly the structured facts an answer engine needs to lift when someone asks "how do I make focaccia?" This guide shows what Recipe schema requires, which optional fields punch above their weight, and the mistakes that get recipes quietly dropped from results.
What Recipe schema does for AI and search
Answer engines do not want to read your headnote about your grandmother's kitchen. When a user asks a cooking question, the assistant needs the ingredient list, the steps, and the timings as clean data it can restate. Recipe schema hands it that data on a plate. Google uses it for recipe rich results and the recipe carousel; Google Assistant and other voice surfaces read steps aloud; and AI assistants that summarise recipes lean on the structured fields rather than guessing from paragraphs.
This is the same principle behind all JSON-LD structured data: you are removing ambiguity. Prose says "bake until golden"; the cookTime field says PT25M. Machines trust the second one. The clearer your data, the more confidently an engine will surface your recipe over a competitor's unstructured page.
The fields that matter
A valid Recipe object needs a name and an image at minimum, but a recipe that actually competes includes far more. The high-value fields are:
- recipeIngredient — one entry per ingredient, written the way a human reads it ("2 cups flour"). This is what gets matched to "does this need eggs?" style questions.
- recipeInstructions — use an ordered list of HowToStep objects, not one blob of text. Each step should be a discrete action so assistants can read them one at a time.
- prepTime, cookTime, totalTime — in ISO 8601 duration format (PT15M, PT1H). These power "quick recipe" and time-based filtering.
- recipeYield — servings or quantity.
- nutrition — at least calories; increasingly used for dietary queries.
- aggregateRating — real ratings from real users. Stars dramatically increase click-through, but they must reflect genuine reviews on the page.
Add recipeCuisine, recipeCategory, and keywords to help engines understand context. If you have a video demonstration, nest a VideoObject so it can appear alongside the recipe.
Match the markup to what's on the page
The cardinal rule of structured data applies with force here: everything in your schema must be visible on the page. If your aggregateRating claims 4.8 stars from 200 reviews, those reviews must exist on the page for a human to see. Inventing ratings, padding review counts, or marking up ingredients that aren't shown is a structured data violation that can trigger manual action and remove all your rich results — not just for that page.
The same goes for instructions. If your steps in schema say something different from the steps in your visible content, engines detect the mismatch and lose trust in your markup. Keep them identical. This mirrors the guidance in our piece on FAQ and How-To schema: markup describes what is on the page, it never adds phantom content.
Common mistakes that kill recipe results
Several avoidable errors keep recipes out of rich results. Putting all instructions in a single text string instead of stepped HowToStep objects means assistants can't read steps sequentially. Using free-text times like "about 25 minutes" instead of ISO 8601 durations means the time fields are ignored. Omitting a high-quality image — recipe rich results are visual, and a missing or tiny image disqualifies you. Marking up a category or roundup page ("15 best pasta recipes") with a single Recipe object when it should be an ItemList of recipes. And forgetting to validate: always run the markup through a testing tool, as covered in our guide to validating structured data, because a warning you ignore today is a lost rich result tomorrow.
One more: write your steps to be self-contained. An assistant may lift a single step out of context, so "add it to the pan" is weaker than "add the garlic to the pan." Recipes that read cleanly step by step get restated more accurately — the same self-contained principle behind how AI chunks content for retrieval.
Recipe schema is one of the highest-return markups you can add, because cooking is one of the most common things people ask assistants to walk them through. Get the ingredients, stepped instructions, and honest ratings into clean JSON-LD, keep them matched to the page, and validate before you ship. To confirm your recipe pages expose the right structured data and that AI crawlers can actually read them, run the CheckMy.site scanner and address what it surfaces.