Selling on Gumroad with AI: 130 products, 0 sales. Greg is wrong

Originally published at tensorust.com.

TL;DR: In a post with 339,000 views today, Greg Isenberg said AI will "probably create 1000+ millionaires" and showed how he would build a one-person business "with $0 and 0 social following". I ran that experiment already. My AI agents built and listed 130 products on Gumroad in 13 days, 88 of them real working apps. They have been on sale for 36 days. At 7:11 PM ET tonight I pulled every number from the Gumroad API: 0 sales, $0 revenue, and 0 downloads of the 87 products that cost nothing. Not one stranger took a free app.

So, is selling on Gumroad worth it? Only if you bring the buyers. Gumroad is a checkout, not a crowd: its own marketplace, Discover, will not list you until you have made at least $100 from genuine sales somewhere else and passed a risk review that "can take on average 3 weeks". With 0 followers you never reach the $100. Below is the store, the code, the full funnel with real numbers, the mistakes that are mine, and a checklist that would have saved me 36 days.

Gumroad API v2 pull, 7:11 PM ET Oct 19 2026. Static frame: funnel.png.

"$0 and 0 social following." I have the receipt

This is the post, from @gregisenberg at 2:38 PM today. It is a 24-minute video about a one-person business powered by AI and Meta Muse. 10,000 people bookmarked it in four hours.

@gregisenberg on X, Oct 19 2026, captured 7:14 PM ET, real and unedited.

I am not going to argue about Meta Muse. I am going to argue about the sentence that sells the video: $0 and 0 social following. That is the part every "AI side hustle" thread repeats, and it is the part I can test, because I did the cheapest possible version of it in September. AI builds the product. AI writes the listing. AI uploads it. Price it low, then free. Wait.

Here is what that looks like when the waiting is over. Every number below came out of the Gumroad API tonight, not out of a screenshot of a dashboard.

What 13 days of AI building put on the shelf

Between September 13 and September 25 an AI coding agent (Claude Code) built and published the store at mattitude8861.gumroad.com. I wrote one-line specs and picked the ideas. It wrote the apps, the listings, the covers and the upload script. Today the API counts 131 products, 130 live:

• 88 single-file HTML apps: budget, habits, invoices, chores, pet records, meal plans, workout logs, a PDF tool, a vibe-coding spec writer, and a long tail of hobby calculators (espresso, reef dosing, 3D print quotes, bike gears).
• Spreadsheet templates, checklists, bundles, and a few $25 to $149 notebooks and services.

6 of the 88 single-file apps, captured by the build scripts Sep 21 to 25 2026.

"Single-file" means each app is one .html file you double-click. No install, no account, no server, nothing to pay for every month. I ran a fresh census over every shipped file tonight:

• 88 apps, 5,482,185 bytes, 25,912 lines in total. The median app is 28.6 KB and 216 lines.
• 0 load an external script, font or image. They work on a plane.
• 60 save your data with localStorage, the browser's built-in key-value store. None needed a database.
• The biggest is the PDF tool at 2.1 MB because it carries its PDF library inside the file.

This is the whole persistence layer of the envelope budget app, from apps8/budget/index.html. It is the pattern every one of the 60 uses:

const KEY = "envelope.v1";
let S = load();
function load() { try { const s = JSON.parse(localStorage.getItem(KEY)); if (s) return { ...DEF(), ...s }; } catch { } return DEF(); }
function save() { try { localStorage.setItem(KEY, JSON.stringify(S)); } catch { } }

// backup = download your state as a file, restore = read it back
$("#exportB").onclick = () => { const a = document.createElement("a");
  a.href = URL.createObjectURL(new Blob([JSON.stringify(S, null, 1)], { type: "application/json" }));
  a.download = "envelope-backup.json"; a.click(); };

Merge with defaults on load so old saves survive new fields, swallow storage errors (private windows throw), and give people a backup button built on URL.createObjectURL, because a browser can wipe localStorage. That is honestly good engineering for a $5 app. The product was not the problem. Hold that thought.

The fresh pull: 0 sales, and 0 free downloads

Tonight's script makes read-only GET calls and writes counts, never buyer data. The part that matters:

prods = gr.all_products()                       # pages through GET /v2/products
summ = {
    'published': sum(1 for p in prods if p['published']),
    'sales_count_sum': sum(p['sales_count'] or 0 for p in prods),
    'sales_usd_cents_sum': sum(int(float(p['sales_usd_cents'] or 0)) for p in prods),
    'free_pwyw': sum(1 for p in prods if p['published'] and p['customizable_price'] and (p['price'] or 0) == 0),
}
sales = gr.call('GET', '/sales').get('sales', [])  # every transaction, any price

Fresh run, 7:11 PM ET Oct 19 2026. Read-only API calls, no buyer data.

Gumroad API, 7:11 PM ET Oct 19 2026	Value
Products / live	131 / 130
Free, pay what you want from $0	87
Pay what you want with a floor ($4+ to $9+)	15
Fixed price	28
sales_count, all products	0
Revenue	$0
Records in GET /v2/sales	0

The free number is the one that hurts, and it is the most useful number in this article. On Gumroad, a $0 checkout is recorded as a sale: it shows up in sales_count and in /v2/sales, same as a paid one. 87 products have been free since September 25. Zero claims in 24 days means nobody ever reached a checkout page. When free does not move, price is not your problem. You do not have a price problem, you have a nobody problem.

Where the traffic came from: 3 posts out of 1,383

On October 11 and 12 my agents made 33 tracked links through Gumroad's UTM feature, one per hero product per platform, so every click would be counted. Here is every click since:

Gumroad UTM link stats, pulled 7:11 PM ET Oct 19 2026; post count from our posts.json log.

10 unique clicks in 7 days. 8 came from a single dev.to article that mentioned one product in context. 1 came from a Moltbook thread, 1 from our own receipts page. The 16 links made for X and LinkedIn: 0.

Now the self-roast, because this one is mine, not Gumroad's. I searched our posting log. Our agent team keeps a log of everything it posts: 1,383 entries so far, posts, replies, articles and pitches. 3 of them carried a link to the store. Those 3 produced 9 of the 10 clicks. The agents were busy farming views for articles and a Steam game, and the store got 0.2% of the attention. You cannot get a sale from a page nobody is sent to, and I sent almost nobody.

So no, I did not prove Gumroad is dead. I proved something more boring and more useful: 130 products with no distribution is the same as 0 products. The AI made the shelf infinite. It did nothing for the foot traffic.

@shl asked how many agents you have. Here is what 3+ agents do on Gumroad

Two days ago Gumroad's founder and CEO Sahil Lavingia ran a poll: "How many agents do you have?" Of 880 votes, 54.5% said 3 or more.

@shl on X, Oct 17 2026, captured 7:14 PM ET Oct 19 2026, real and unedited.

Sahil, I am in your 54.5%. My agents shipped 130 products to your platform in 13 days, faster than a human could write the listings. And your platform did exactly what it should with a store like mine: nothing. Read the Discover rules again. To be shown to Gumroad's own buyers you need payout settings, a balance of at least $100 "from enough genuine sales (not self-purchases or asking your partner/housemates to buy your product)", and a risk review on top. Then each product needs one real sale of its own.

Here is my read, and I think it is where this is heading. That $100 wall used to be a fraud filter. In the agent era it is the most important sentence in Gumroad's help center, because it is the only thing standing between buyers and a flood of AI-built stores like mine. If 54.5% of Sahil's followers run 3+ agents, they can all do what I did in a week. Gumroad cannot show all of it, so it shows none of it until a human proves demand. The marketplace is not going to save the "0 following" crowd. It is specifically built to ignore them.

And when Discover does sell for you, it costs more. These are Gumroad's published fees, checked tonight:

Gumroad's fees and Discover help articles, fetched 7:15 PM ET Oct 19 2026.

A $5 sale	You keep	Fee
Buyer you sent (10% + $0.50, plus card 2.9% + $0.30)	$3.56	$1.45
Buyer Gumroad sent via Discover (flat 30%)	$3.50	$1.50

Source: Gumroad's fees page. On a $5 product the 50-cent flat fee is a tenth of the price, so cheap products pay the most. The fee was never the problem for me, though. 30% of $0 is $0.

Supply is infinite. Attention is the product

Gumroad's API has an endpoint that returns how many paid listings match a search, with price percentiles. I ran it tonight on the categories my apps compete in:

Gumroad API GET /v2/products/comps, 7:12 PM ET Oct 19 2026.

• 10,388 paid Notion templates, median $10.
• 3,151 ChatGPT prompt packs, median $15.02.
• 1,988 habit trackers, 1,569 budget trackers, 1,027 meal planners.
• 456 "offline app" and 399 "html app" listings, medians $20 and $23.99.

My budget app is listing number 1,570. Being better than the other 1,569 does not matter if the buyer never sees any of them, and on Gumroad the buyer mostly arrives from the seller's own audience. This is why @HussainIbarra's post today is the honest version of Greg's: "Most creators will never make a full-time living from it", because "if nobody knows who you are... you'll end up being a starving artist." 108,000 views. He is right, and he is right for the reason my store proves.

What I think is really going on with "$0 and 0 following"

This part is my opinion, not a measurement. The 1-person AI business pitch is true for exactly one kind of person: someone who already has the audience. Greg has a huge audience and a podcast. When he says "0 social following", he is describing a starting line he left years ago. The video is the product. 339,900 views and 10,000 bookmarks in four hours is the business.

I don't think that makes him a scammer. I think it makes the advice survivorship bias with a ring light. The people it works for are visible by definition. The people it failed for have 130 products and nobody to tell. Here is one of them telling you.

And there is a bigger thing hiding in the numbers. AI dropped the cost of making a product to about zero: 88 apps in 11 days, 25,912 lines, 0 servers. When making is free, every shelf fills up, and the only scarce thing left is people who trust you enough to click. My prediction: within a year the digital-products crowd stops bragging about how many products they shipped and starts bragging about list size, because that is the only number that turns into money. I already tested the other number. It is zero.

The mistakes that are mine (so you skip them)

1. I built 130 before selling 1. Ten a day is Gumroad's own cap on new products, and I hit it on purpose. Every product after the first should have waited for one stranger to take the first one.
2. I treated "free" as marketing. Pay-what-you-want from $0 on 94 products changed nothing, because nobody saw the price. Free is a conversion tool, not a traffic source.
3. The covers fought the price. The AI-made covers printed "$8 ONE-TIME" while the button said $0+. A visitor saw two prices. We only noticed in an audit on October 11.
4. A silent API trap. GET /v2/products now returns 10 products per page with a next_page_key. My original upload script made one call, so its "does this product exist" check saw 8% of the store. Always page.
5. The agents never promoted it. 3 posts in 1,383. That is the real cause of death, and it is an instruction problem, not a platform problem.

Do this in your project before you list product number two

• Run the $0 test. Make your first product free for a week. If nobody takes it, stop building and fix traffic. My 87 free products would have told me this on day 3.
• Track every link. Gumroad's UTM links count clicks per source. Count clicks before you count sales; 10 clicks in a week is a traffic number, not a conversion number.
• Count your own promotion. Search your sent posts for your store URL. If it is under 10%, you are not selling, you are hoping.
• Know the Discover gate. $100 of genuine sales plus a review of about 3 weeks before Gumroad shows you to anyone. Plan your first $100 from your own channels.
• Put the work where AI is weak. Building is now the cheap part. Spend the hours on the page a stranger lands on and the places that send strangers there. I measured what actually predicts a page's results in a landing page grader run on 2,077 Show HN pages, and what 500,000 views turned into in 500K views, 3 checkouts, $0.
• One hero, not 130. Pick the product that fits the audience you already have and point every post at it.

Building cheap with AI is real, and I have the other receipts too: 20 iPhone apps for $99 in fees and a vibe-coded game that launches on Steam tomorrow. Building was never the hard part. Being seen is.

The numbers, in one place

• Products: 131, live 130, built Sep 13 to 25, 2026
• Free ($0+): 87, free since Sep 25
• Sales, including $0 downloads: 0. Revenue: $0
• Single-file apps: 88, 25,912 lines, median 28.6 KB, external scripts 0
• Tracked links: 33, unique clicks in 7 days: 10 (8 from one dev.to article)
• Posts that linked the store: 3 of 1,383
• Discover entry: $100 genuine sales + about 3 weeks of review
• Competing paid listings: 1,569 budget trackers, 10,388 Notion templates

If your page is the thing nobody clicks, run it through the free landing page teardown before you build product number two.

One question for @gregisenberg and @shl, and for anyone who has actually tried it: name one person who went from 0 followers to $1 million on AI-built digital products, with receipts. I will put their numbers next to mine in the sequel.