Every category Gumroad publishes, measured

What this sample says that the first one could not

• The market is more concentrated than the earlier sample suggested. The top 1% of products carry 52.1% of all 201,554 ratings measured, and the top 10% carry 87.9%.
• Sellers, not just products, are concentrated. The top 1% of sellers hold 53.1% of the ratings; the busiest single seller has 89 listings in this sample while the median seller has 1.
• Asking prices are lower than a search-term sample shows. Median paid listing $18.03, 75th percentile $41.53, 90th $100.18. Walking the whole taxonomy reaches parts of the catalogue that popular search terms never surface, and those parts are cheaper.
• 43% of products have no ratings at all — 3,562 of 8,311. That is the background rate you compete against.
• Subscriptions stay rare: 170 of 8,311 products bill recurring.

Method, and the one thing this cannot tell you

The sampling frame is Gumroad's own published category tree, read out of the category picker rather than invented here: 359 nodes were crawled, 255 returned listings and 98 returned none and are excluded rather than reported as zeroes — "we found nothing" and "there is nothing" are different claims and only the first is evidenced. A further 6 are excluded because Gumroad's discover endpoint does not recognise them and answers with the site-wide default feed instead of saying so; they were published as categories until 2026-08-11 and the six are named, with the test that catches them, in data/taxonomy-summary.json. Each node was taken up to three pages deep, which caps a node at 44 listings; 187 nodes hit that cap. So a category's listing count here is a crawl depth, not a category size, and no figure on this site should be read as "how many products are in this category". What the design does support is comparison between categories, because every one was sampled identically.

A row in the CSV is a listing observation: 9,617 observations cover 8,311 distinct products, because Gumroad files 1,306 of them under more than one category. Every market-wide figure above counts a product once, keyed on its product URL — an exact identity key, unlike the card-text key the earlier sample had to use. Prices are converted to USD at ECB reference rates for 2026-08-06; the raw price and its currency are both kept in the file.

Every category

Ranked by the share of listings carrying at least one rating — the cleanest available signal for whether a category transacts at all. Indented names are subcategories of the row above them. Categories with fewer than 30 listings sampled are shown here but have no page of their own; their quartiles would be noise.

Find yours:

The data — free, no signup

All 9,617 rows: data/gumroad-taxonomy.csv — category, product URL, seller, asking price, currency, USD price, rating count, star rating, subscription flag, title. Produced by collect_taxonomy.py and normalised by normalize_taxonomy.py, with every per-category figure on this page emitted as taxonomy-summary.json. Data CC BY 4.0, code MIT. Use it for anything.

The earlier, smaller sample built on 42 Discover searches is still published and still DOI-archived at 10.5281/zenodo.21830103; it is a different sampling frame, not a superseded file, and the two are kept side by side rather than silently merged.

The same data, one row per seller

Every listing above carries the storefront that published it, so the same file answers a question no other public Gumroad dataset can: the top 1% of 4,532 sellers hold 53.1% of every rating measured, and the median seller in that 1% has two products. Concentration here is not a catalogue effect.

All 4,532 sellers, ranked →  ·  the seller CSV (free, CC BY 4.0)