bingwow/recipe-site-markup-coverage-2026

Recipe Site Markup Coverage and Extraction Observations 2026

Open dataset that pairs a hand-maintained inventory of 137 recipe sites with anonymized domain-level extraction observations from 441 real extraction attempts, capturing which recipe sites expose machine-readable Schema.org Recipe markup, where AI fallback parsing is needed, and where bot protection blocks automated readers.

Published by RecipeStripper — a free web app that turns any public recipe URL into a clean, mobile-first cooking view with ingredient quantities embedded inline into each instruction step.

This is the Hugging Face mirror of the data on GitHub (forrestmill-cmd/recipestripper-public-data) and Kaggle.

Contents

File	Rows	Description
data/site-inventory.csv	137	The recipe sites RecipeStripper lists as supported, with category, monthly visitor estimate, status (listed_supported / limited / blocked), and example path.
data/site-inventory.json	137	Same data, JSON.
data/domain-extraction-observations.csv	122	Per-domain aggregate from 441 strip_log rows: attempts, successes, failures, success rate, AI-fallback uses, primary parser source (json-ld / microdata / heuristic / openai), common error code, average step + ingredient counts.
data/domain-extraction-observations.json	122	Same data, JSON.
data/summary.json	1	Top-level totals + category counts + privacy/methodology metadata.

Headline numbers (2026-05-12 snapshot)

• 137 listed sites across 6 categories (major: 26, food-blog: 43, healthy: 20, international: 20, niche: 20, baking: 8).
• 4 sites are explicitly limited or blocked by PerimeterX bot protection (Serious Eats, The Kitchn, J. Kenji López-Alt at Serious Eats, Serious Eats Drinks).
• 441 real extraction attempts aggregated across 122 observed domains.
• 216 successes / 225 failures (49% success rate over the full sample, including known-blocked sites and roundup-listicle URLs).
• cooking.nytimes.com: 21/21 successes via JSON-LD, 0 AI-fallback uses, avg 4.48 steps + 11.24 ingredients per recipe.

How the data was collected

• site-inventory.* — hand-curated from RecipeStripper's lib/data/recipe-sites.ts, the source of truth for the /works-with/<slug> directory.
• domain-extraction-observations.* — aggregated from RecipeStripper's strip_log table, with no submitted URLs, user identifiers, IP addresses, or saved recipe content included. Only domain-level counts, parser source, error types, and average step/ingredient counts.

Why this matters

Recipe blogs, food media, and AI cooking assistants increasingly rely on parsing third-party recipe pages. This dataset gives builders, researchers, and AI-tool developers a concrete picture of where Schema.org Recipe markup is healthy, where heuristic or LLM fallback is needed, and where bot protection makes automation impossible.

If you're building a recipe extractor, a cooking assistant, a JSON-LD coverage report, or evaluating which recipe sites are friendliest to automated reading — this dataset is the empirical baseline.

Citation

@dataset{miller2026recipesitemarkup,
  author       = {Miller, Forrest},
  title        = {Recipe Site Markup Coverage and Extraction Observations 2026},
  year         = {2026},
  publisher    = {RecipeStripper Research},
  url          = {https://huggingface.co/datasets/bingwow/recipe-site-markup-coverage-2026},
  note         = {Mirrors https://recipestripper.com/research/recipe-site-markup-coverage-2026 and https://github.com/forrestmill-cmd/recipestripper-public-data}
}

Related links

• Web app: recipestripper.com
• Research page (canonical): https://recipestripper.com/research/recipe-site-markup-coverage-2026
• GitHub mirror: https://github.com/forrestmill-cmd/recipestripper-public-data
• Kaggle mirror: https://www.kaggle.com/datasets/forrestmill/recipe-site-markup-coverage-2026
• Wikidata Q-item: Q139885422

License

Creative Commons Attribution 4.0 International (CC BY 4.0). Reuse, redistribute, and build on the data — including commercially — with attribution to RecipeStripper (recipestripper.com).