Smart parsing
The engine underneath everything else. We read messy recipe pages and figure out which words are ingredients, which are steps, which are filler, and which are ads — even when the original page makes no effort to mark them up.
Try it now- Author
- No Story Please
- Category
- Engine
- Coverage
- 1,000+ recipe sites
- Features
- Structure-aware · Site-resilient
- Share
Start with the messy original
A typical recipe page is 2,400 words: a memoir paragraph, three pop-ups, a sponsored block, a video player, an ingredient list, a step list, comments. The recipe is maybe 300 of those words, scattered.
We pull the page in raw, without trusting the publisher's structure — because most pages don't have one.
Detect what's recipe and what isn't
The parser tags every block: ingredient, step, time, yield, headnote, filler, ad. It uses structure (lists, headings), language patterns ("½ cup", "preheat to 350°F"), and visual signal (ad slots, sticky elements) — not a hardcoded site list.
When the parser is unsure, it keeps the block and dims it rather than dropping content silently.
Output a clean, structured recipe
The result is the recipe with no surrounding noise, restructured for cooking: a single ingredient list, a clean step list, time and yield surfaced at the top.
Because we worked from structure rather than a site allowlist, the same engine handles a recipe blog you've never heard of and last week's NYT Cooking post the same way.
Start with the messy original
A typical recipe page is 2,400 words: a memoir paragraph, three pop-ups, a sponsored block, a video player, an ingredient list, a step list, comments. The recipe is maybe 300 of those words, scattered.
We pull the page in raw, without trusting the publisher's structure — because most pages don't have one.
Detect what's recipe and what isn't
The parser tags every block: ingredient, step, time, yield, headnote, filler, ad. It uses structure (lists, headings), language patterns ("½ cup", "preheat to 350°F"), and visual signal (ad slots, sticky elements) — not a hardcoded site list.
When the parser is unsure, it keeps the block and dims it rather than dropping content silently.
Output a clean, structured recipe
The result is the recipe with no surrounding noise, restructured for cooking: a single ingredient list, a clean step list, time and yield surfaced at the top.
Because we worked from structure rather than a site allowlist, the same engine handles a recipe blog you've never heard of and last week's NYT Cooking post the same way.
Paste a URL. Get a clean recipe.
The engine runs on every page in the app — no setup required.