What a UK buyer is actually charged on Gumroad — 27 of 31 stores measured

Every product page below was loaded logged out from London, then its own “I want this” control was followed to Gumroad’s checkout and the checkout was read. 27 of the 31 stores we could read charged more at the pay step than the page advertised, a median of 21.2% more (range 16.6%–27.9%). The extra is VAT, added on top.

What each store advertised, and what the pay step totalled

The ones we could not read — 1 of 32

Listed rather than dropped. A percentage whose denominator quietly excludes the awkward cases is not a measurement.

Where the difference comes from, and it is not one thing

The headline figure looks like a tax rate and is not one. The pay step itemises a subtotal as well as a tax, and the subtotal is not the price the page advertised. Two effects stack:

• Tax on the subtotal: median 20.0%, range 18.7%–20.9% across 28 readings. Tight enough to read as a single statutory rate.
• A drift between the advertised price and the amount actually taxed: median 0.9%, range -3.6%–7.7% across 29 readings. The displayed price is converted for the visitor and the charged one is rounded separately, so the number being taxed is not the number that was quoted.

Both figures use only readings where the page and the pay step quoted the same currency — across a currency change the difference is an exchange rate, not a drift. Excluded from both: the one store published as our own misread — its page price was read wrong here, so its page-to-subtotal difference is our error and not a drift. Neither component is hidden and neither is large alone; the compounded number is the one a buyer meets.

Does it hold up on stores we did not draw at random? — 61 of 68

Not a random sample, and that is stated before the number and not after it. Alongside the draw above, 71 further stores were walked the same way between 2026-08-11 and 2026-08-12 — one product each, none of them in the sample, none of them ours. They were selected to be written to, so each passed filters a random draw does not apply: a ratings-ranked band of the seller index, a published contact address on the product page (about 17.5% of sellers have one, and that is the largest bias here), a top-rated product that is not free, and a published category page.

As an estimate of the platform that cohort is worth nothing. As a replication it is worth more than the draw: it was assembled for an unrelated purpose, by different criteria, from a different part of the index, on different days — and it lands 0.2 points from the sample’s median. 61 of the 68 readable stores charged more at the pay step, median 21.4% (range 17.1%–23.4%), against 21.2% across the 31 readable stores of the random draw.

Not read, 3 of 71, listed rather than dropped for the same reason as above.

Where a store was walked more than once, the most recent reading is the one counted and the store counts once — 32 readings were set aside that way, 1 for already being in the draw above and 1 for being ours. The records are in data/checkout-replication.json.

Method, in full

• 32 stores drawn and walked, 31 read, 1 not read. 27 charged more at the pay step, 1 matched the page, 1 came out lower — that last group is almost certainly us misreading the page price (a pay-what-you-want minimum or a struck-through price), not a discount, and it is counted in the denominator rather than thrown away. 2 switched currency between page and checkout.
• 32 of the stores were drawn from 4,255 sellers with a fixed seed (20260809), eight from each of four price bands, before anything was walked. The draw is reproducible from the same CSV, so nobody — including us — can re-roll it until the result looks better. Our own product was walked alongside them and is reported above rather than counted here — a page that reports a platform's behaviour and exempts its own author from it is worth less than nothing, and one that puts a self-chosen store in a random sample is worth less than that.
• Two page loads per store and nothing else. The product page, then the gumroad.com/checkout URL the page’s own buy control carries. No order was completed, no form was submitted, no email or card was typed into anybody’s checkout, and no account was touched. It is what a buyer who thinks twice and leaves does.
• From London, GB. A buyer somewhere without VAT sees a different number, and that is the point — the difference is a function of where the buyer is, which is precisely the thing the seller cannot vary from inside their own account.
• Read on 2026-08-09. Prices move; re-run it yourself. The reader is open source and the raw records for this page are in data/checkout-audit.json.

How many of them say so on the page? — 3 of 84

Everything above is a fact about Gumroad. This is the only part that is a fact about the sellers, and it is the sharper number. Of 84 third-party storefronts where we recorded a pay-step gap, 81 — 96.4% — say nothing about tax anywhere on the product page. Three tell the reader. None claims the price already includes it.

The distinction matters more than it looks. The 21.2% gap is not the seller’s doing and not the seller’s to fix — that is said plainly at the top of this page and it stays said. But whether the page mentions it is entirely theirs: one line in a description, no plugin, no platform change, about sixty seconds. It is the one part of this whole measurement that comes with an action attached.

The three that do it, and the words that earned it. The right-hand column is the span the classifier matched, not a tidied quotation:

Why 81 is not just “our search missed it”. A needle set can fail by being too narrow, and that failure looks exactly like a finding. So all 81 silent pages were re-read and searched for the bare words vat, gst, tax and taxes anywhere in their visible text, with no requirement that they sit near anything else. 0 of 81 contain any of them in any context. The silence is total, not a matter of phrasing. Matching is word-boundary anchored throughout, because pri​vat​e, inno​vat​ive and syn​tax would otherwise each read as a tax disclosure.

Our own store is excluded from both sides of that ratio, the same way it is excluded from the drawn sample above. It was measured too: all three paid pages state that tax is added, and did so before this count was taken rather than after it.

What this is not. It is a measurement of what page text says, and nothing else. It is not a claim that any seller has broken any rule — that depends on who the trader is, where their buyers are and how a regulator reads a platform-collected tax, none of which a reading of a web page can settle. If you want the rules themselves rather than our summary of them, the relevant ones are the definition of “selling price” in the Price Marking Order 2004, regulation 6 of the CPUTRs 2008, and the consumer provisions of the Digital Markets, Competition and Consumers Act 2024, in force since 6 April 2025. That last one is the one to read: it requires the total price up front, and where a mandatory charge cannot be quantified in advance — which VAT here genuinely cannot, since it depends on a buyer’s country — it asks that the headline price be accompanied by what a reader needs to work the total out. Which is the one line above.

What a seller can actually do

Not much, and it is better to say so than to invent a fix. You cannot turn it off; you are not supposed to. What you can do is stop being surprised by it:

• Know your own number. Open your product page in a private window from a VAT country and follow your own buy button. It takes a minute and it is the only way to see what your buyer sees. If you are not in a VAT country and cannot do that, this reads it for you — the same code that produced the table above, one file, MIT, no signup and no email. It loads two pages and stops.
• Say it on the page. A line saying tax is added at checkout for EU/UK buyers costs nothing and removes the surprise at the exact step where surprise loses the sale.
• Compare like with like. When you benchmark your price against a competitor, you are reading their pre-tax number and your buyer is paying a post-tax one.
• Or have it re-read for you. The first bullet is a thing you do once, and the number it gives you goes stale the next time you change a price, add a product or Gumroad changes what it collects. The checkout monitor, $99 a month, is that same walk, run against every product on your store on a schedule, and it writes to you only when a page and its pay step start disagreeing. It is the paid, recurring version of the free script above — the script is not a teaser for it, and if once is all you need, once is free. Your first month is the full storefront audit, included rather than added on. Cancel from your receipt.

What one storefront's full audit looks like

This page reads one product from each of many stores. The other half of the same measurement is every product on one store, walked the same way and placed against what comparable products in its own category charge. That report is a paid product, and a complete worked sample of it is published here — one real storefront, nine products, the seller's name removed and not one figure altered.