How many sales is one Gumroad rating?
This estimator needs JavaScript. Every band it reads is in the tables below and they need none.
Which categories this covers. The listings behind every figure on this page, and the estimator above come from 15 of Gumroad's 15 top-level categories (3D, Audio, Business & Money, Comics & Graphic Novels, Design, Drawing & Painting, Education, Fiction Books, Films, Fitness & Health, Gaming, Music & Sound Design, Other, Photography, Recorded Music), so they are not an even draw across the platform. They are also unevenly weighted: 57% of them sit under 3D alone, against the 7% an even draw would give it. Weight them accordingly.
Why there is no single multiplier
The number sellers quote each other is “about thirty sales per rating”, sometimes a hundred. Neither has anything behind it. Here is what 229 products that publish both a unit-sales count and a rating count actually show.
| Group | Products | Median | Middle half | Full range |
|---|---|---|---|---|
| Paid products | 190 | ×25.5 | ×11.7 – ×54.2 | ×1.0 – ×335.0 |
| Free products | 39 | ×24.1 | ×7.1 – ×90.0 | ×1.2 – ×1418.0 |
| Everything together | 229 | ×25.2 | ×11.0 – ×56.5 | ×1.0 – ×1418.0 |
Read the fourth column before the third. For paid products the middle half of listings sit between ×11.7 and ×54.2 — a factor of 4.6 between the 25th and 75th percentile, before counting the quarter of listings above and the quarter below. Applying one multiplier to one listing is guessing inside that range and calling it measurement.
Said the way a seller experiences it: the median paid listing converts about 3.9% of its buyers into raters. On free products it is 4.1% — people who paid nothing say less, which is the opposite of the common assumption that free products farm reviews.
The multiplier is not a constant. It rises with the size of the listing.
This is the part that makes a fixed number actively misleading, and it is visible two independent ways.
Cut by how many ratings a listing has:
| Group | Products | Median | Middle half | Full range |
|---|---|---|---|---|
| 1–2 ratings | 50 | ×22.0 | ×6.8 – ×64.0 | ×1.0 – ×1418.0 |
| 3–9 | 73 | ×31.2 | ×12.5 – ×63.9 | ×1.3 – ×407.8 |
| 10–49 | 58 | ×27.0 | ×14.0 – ×49.8 | ×1.2 – ×334.5 |
| 50 or more | 48 | ×26.4 | ×14.3 – ×53.5 | ×1.3 – ×224.9 |
A listing with 1–2 ratings is a ×22.0 product at the median. One with 50 or more is ×26.4. The same “×30 rule” applied to both overstates the small one by roughly 1.4× and understates the large one.
Cut by how many sales a listing has, which points the same way:
| Group | Products | Median | Middle half | Full range |
|---|---|---|---|---|
| 1–9 sales | 16 | ×3.5 | ×2.0 – ×6.8 | ×1.0 – ×9.0 |
| 10–49 | 30 | ×11.2 | ×7.8 – ×21.8 | ×1.2 – ×36.0 |
| 50–199 | 53 | ×18.0 | ×10.8 – ×46.9 | ×1.3 – ×151.0 |
| 200–499 | 55 | ×30.3 | ×16.1 – ×62.3 | ×2.1 – ×205.0 |
| 500–1,999 | 40 | ×54.1 | ×24.5 – ×87.0 | ×1.7 – ×1418.0 |
| 2,000+ | 35 | ×49.0 | ×28.8 – ×103.7 | ×9.6 – ×334.5 |
Read that second table with a correction in mind. When a listing has at least one rating, its ratio can never exceed its sales count — a product with six sales cannot show a multiplier above six. So the low bands there are censored by construction and the rise across that table is partly mechanical. The ratings-cut table above it is not censored that way, and it shows the same climb from ×22.0 to ×26.4. That is why both are printed: one of them would have been a nicer chart and a worse answer.
Price barely matters
| Group | Products | Median | Middle half | Full range |
|---|---|---|---|---|
| under $10 | 70 | ×17.9 | ×8.6 – ×40.7 | ×1.0 – ×220.0 |
| $10–25 | 71 | ×39.9 | ×20.4 – ×58.8 | ×1.0 – ×203.0 |
| $25–50 | 36 | ×20.3 | ×9.9 – ×49.4 | ×1.7 – ×205.0 |
| $50+ | 43 | ×31.2 | ×11.9 – ×73.6 | ×2.1 – ×335.0 |
Across paid listings the ratio does not move with price in any direction worth acting on. Whether a buyer leaves a rating appears to be a function of how many buyers there are, not of what they paid.
Does the proxy this whole site runs on actually work?
Every other page here uses ratings to compare demand, and every one of them says so. This sample is the only place that can test it. Across the 316 products that publish a sales count, the rank correlation between ratings and units sold is 0.831 (0.784 among the ones with at least one rating).
That is a strong relationship, and it licenses exactly one use: ratings rank demand reliably, and measure it badly. If listing A has four times the ratings of listing B, A almost certainly outsells B. If you want to know by how much, the tables above are the honest answer and they are wide.
An unrated listing has not necessarily sold nothing
87 of the 316 products publishing a sales count have zero ratings. Their median is 8 units; 37 have ten or more, 8 have over a hundred, and the largest has 1,320 sales and not one rating.
They are excluded from every median above, because sales divided by zero ratings is not a number. That exclusion is also why those medians are a lower bound: the products where buyers rate least are precisely the ones the ratio cannot see.
What this sample is
1,359 Gumroad product pages fetched individually from 824 sellers. 316 of them — 23.3% — publish a unit-sales count (23.7% of paid listings, 20.7% of free ones), covering 450,651 units sold in total across 214 sellers. The ratio is measured on the 229 of those that also have at least one rating.
The category limit comes first and it is stated in full at the top of this page: the crawl behind these products has not finished and its sample is uneven, so it leans on one branch of the marketplace rather than being a cross-section of it.
Two further biases, and which way each cuts. First, displaying a sales counter is opt-in, and a seller with nothing to show is likelier to leave it off — so this is not a random draw of Gumroad products, and the listings in it are larger than typical. Second, the ratio needs a rating to exist at all, which drops the 87 zero-rating listings and biases the medians down. Neither is fixable from public data; both are stated wherever these figures appear.
Which sample, and why not the other one. These products come from the category walk, re-fetched one page at a time — the same population as the seller analysis, not the 42-search sample of 1,344 products every other guide here is built on. The two samples disagree on price by a wide margin and are published as measured. Nothing on this page is averaged with anything from the other one.
This page tells you what a rating is worth. The report tells you where the ratings are: all 42 categories read together and each classified as an opening, a crowded room or thin — which is the question you hit the moment you can convert a competitor's rating count into a plausible unit figure. You are paying for the interpretation, not for the rows. The rows are free, above and in the repository. If the data is all you wanted, take it and skip this.
Read the report — $249 Read three sections free first Or take all four CSVs free
The sample is the method sections lifted unedited out of the report — what was measured, the background rate, and the limits in full. Every section that says what to do is in the paid one. The free mirror is $0 with a $0 minimum and asks for an email at checkout. The same files without one: the data folder. The report carries a money-back guarantee, stated on the checkout page: full refund, no reason needed, and you keep the files.
Check it yourself
Every figure above comes from the sales CSV — one row per product fetched, including the 1,043 that publish no sales count, so the opt-in rate is re-derivable and not just asserted — with the derivation and the collector beside it. CC BY 4.0, no signup, DOI-archived.