Why a link becomes a card: structured recipe data explained

Paste a link, get a card, no AI involved. The trick is that most recipe pages already carry a machine-readable copy of themselves.

The recipe under the recipe

To earn rich results in search — the star ratings, cook times, and photos — recipe sites embed their recipes in a standard format called schema.org Recipe data, usually as JSON-LD inside the page. It includes the title, the yield, the complete ingredient list, and the ordered instructions. The generator reads exactly that: the same data the site voluntarily publishes for machines, with the source linked on every card.

From lists to a diagram

Structured data supplies lists; the card needs a tree. A matching pass ties each ingredient to the step where it first appears — smart enough to know that “the vanilla” means the vanilla extract, and that “butter a loaf pan” isn’t using your butter. Heuristics then detect separate bowls, the merge that unites them, layered assembly, and sequential finishing steps, producing the merged-cell layout.

When a page has no data

A minority of pages — older blogs, some paywalled sites — publish nothing structured. The generator falls back to reading the visible page itself, looking for an ingredients heading with its list and a directions section. And when even that fails, paste mode accepts the recipe as plain text and runs the identical pipeline, fully offline. If a favorite site consistently fails by link, paste is the reliable path, and the card comes out the same.

Why not AI?

Deterministic parsing is instant, free, private, and reproducible — the same recipe always yields the same card. Where it errs, it errs legibly, and you can see exactly why. For a tool meant to be trusted mid-recipe, that beats a clever guess. Try a link.