Selling a Digital Product Made With AI: What Nobody Tells You First

Page 8 of the PDF I sell on Gumroad contains this sentence: “Spend an hour a day and you will have a live income stream and, very likely, your first sale by day 30.”

I put that on sale. I have no evidence for it, and neither would anyone else: it is a promise about strangers’ results, attached to a 30-day plan I had never followed myself. It sat in a file with this site’s name on it for eleven weeks before I opened it again for this post and read it the way a buyer would.

That is the thing nobody tells you first about digital products made with AI. Making one takes an evening. Making one you can stand behind needs the part a model cannot supply, which is having done the thing the product describes.

What this is based on: two products of mine, a twelve-page PDF on Gumroad and a book on Amazon KDP. The details of the PDF come from the file itself. Platform rules are quoted from each platform’s own pages, read on 21 September 2026. Sales figures come from the Gumroad and KDP dashboards on the day this post was published. Nothing here is a projection.

What the file actually is

Twelve pages, about 25 KB, no images, every line set in Helvetica. The metadata names the ReportLab PDF library as the producer, which is a Python library, so the layout came from a script. It gives the creation time as 20:40 on 4 July 2026. The author field says “Vireltify”, the name of this site, rather than the name of a person. Twenty-three minutes earlier, the same process had produced a free seven-page companion.

In plain terms: I asked a model for a guide, it wrote one, and a script laid it out. The result has an introduction and eight chapters: an “AI income mindset”, “the 5 proven models”, a toolkit, a 30-day launch plan, getting your first clients, a chapter on “scaling past” a monthly income figure, “The 10 mistakes that keep beginners broke”, and next steps.

The uncomfortable part is that it isn’t badly written. The structure is sensible, and the main point of the first chapter, that you sell outcomes and not “AI”, is advice I would still give. It reads like exactly what it is: the average of every guide on the subject. That is why it is worth so little. A buyer can get the average for free, from the same kind of tool, in less time than it takes to reach the checkout.

Three things in it I would take out today

The day-30 promise. Quoted above. It predicts a result for a reader I know nothing about, on a timetable nobody measured. Any sentence of the form “you will have X by day Y” belongs in the bin unless you can show whose X, and when.

A wrong fact, stated calmly. Page 6, describing freelancing: “You keep 100% of what you charge and can start today on Upwork, Fiverr, or by direct outreach.” For the two marketplaces it names, that is wrong. Both charge freelancers a fee, and each states it on its own site. The model wrote the sentence with total confidence and I didn’t check it. One wrong fact in a paid product does more damage than a hundred vague ones, because it is the one a buyer can verify.

The money in the titles. The cover promises a first four-figure month “online with AI”, and the title of chapter 6 is about scaling past that same figure. Page 2 carries a disclaimer saying the guide “does not guarantee any specific income”. A disclaimer on page 2 does not cancel a promise on page 1. Google’s publisher policies name get-rich-quick schemes when they describe deceptive claims, and for weeks this blog applied for AdSense while linking to that product from every post.

The Gumroad listing has since been retitled so it no longer names an income figure, and the price is $5 everywhere it appears. That fixes the shop window. It doesn’t fix the file, whose own cover and page 10 still say what they said in July. So the PDF is not linked from anywhere on this site, and it won’t be until the file matches the listing.

The question Amazon asks, and why the wording matters

KDP makes you declare AI content, and its definition is narrower than most people assume. From the KDP content guidelines, read on 21 September 2026: “We require you to inform us of AI-generated content (text, images, or translations) when you publish a new book or make edits to and republish an existing book through KDP.”

The next part decides which box you tick. Content counts as AI-generated if an AI tool created the actual text, images or translation, “even if you applied substantial edits afterwards”. It only counts as AI-assisted, which you don’t have to declare, if “you created the content yourself” and then used tools to “edit, refine, error-check, or otherwise improve” it.

Most people who have rewritten a model’s draft heavily believe they belong in the second group. By Amazon’s wording they are in the first: authorship decides it, not the amount of editing. The same page adds that you are responsible for making sure all AI-generated and AI-assisted content meets its content guidelines, intellectual property rights included. And the declaration applies again every time you edit and republish, which is easy to forget by the third revision.

My book, Make Money With AI, was drafted the same way as the PDF. When it went on sale, Amazon asked whether the text and the images were AI-generated. The cover is plain type on a blue gradient with no generated artwork, so the image half of the question never came up. The subtitle is another matter. I won’t repeat it here, but it promises beginners side hustles and income that keeps arriving without further work, which is exactly the kind of claim this site stopped making. The book needs the same treatment as the PDF before I point anyone at it.

What a $5 price does to the fee

Gumroad’s pricing page, checked 21 September 2026, lists no monthly fee, 10% plus $0.50 per transaction for sales through your own profile or links, and 30% when a new customer finds you through its Discover marketplace.

At $5, a sale through my own link costs $1.00 in fees, which is 20% of the price. Through Discover it costs $1.50. The same fee on a $20 product, through your own link, is $2.50, or 12.5%. The fixed fifty cents is what makes a cheap product expensive to sell, and it is a sum worth doing before you set a price rather than after.

One thing in Gumroad’s favour, if you sell from outside the US: the same page says that since 1 January 2025 it handles sales tax collection and remittance worldwide as merchant of record. For a seller in Spain with buyers anywhere, that is real work you don’t have to do. How a product like this compares with display ads and affiliate programmes, as ways for a small site to earn anything, is in how a small blog actually makes money.

What actually sold

I am not putting sales figures for either one in this post. The editorial policy on this site rules out publishing earnings numbers, and that applies to my own products before it applies to anyone else’s.

Before you read anything into that, here is all the promotion either product ever had: a box at the end of every post on this blog. In the three months to 19 September, Google sent this blog fifteen clicks. That box also advertised the PDF at $12 while the checkout charged $5, a mismatch I only found when I exported the whole site in September. The full list of what those boxes got wrong is in my write-up of the affiliate mistakes here.

So the numbers above mostly measure distribution, and there wasn’t any. I’m not going to project from them in either direction.

If your product is images rather than text

The five older posts that now redirect here covered Kindle ebooks, prompt packs, AI art for print-on-demand, AI stock photos and Etsy shops, and they now point here. Kindle is covered above. For the rest, I have not sold images or prompts anywhere, so this part is what the platforms say, not what happened to me.

  • Shutterstock does not take AI-generated work from contributors. Its contributor help page, read on 21 September 2026, says it “will not allow AI-generated content to be submitted by contributors for licensing”, because ownership of that content “cannot be assigned to an individual”.
  • Adobe Stock does accept generative AI content, as long as you tick its “Created using generative AI tools” box and hold the rights to submit it. Its help page refused my request on 21 September, so that summary comes from Adobe’s own page as it appears in search results. Read Adobe’s generative AI guidelines in full before uploading anything.
  • Etsy sets rules for AI-assisted items in its Creativity Standards. Its pages returned an error to me on 21 September, and I am not going to paraphrase rules I could not read. Open Etsy’s Creativity Standards yourself before you list.
  • Prompt packs I have never sold. My only view is a structural one: a buyer sees the whole prompt the first time they use it, so what you are really selling is the time it saved them, and that is easy to copy and hard to protect.

If generated images go into anything you sell, a cover included, the commercial-use terms of the tool that made them matter as much as the platform’s rules. The terms I could and couldn’t verify are in my notes on AI image generators for site and blog images.

What I would do before listing anything again

Start from something I have actually done. If I made another PDF tomorrow, it would be the launch routine I run in Search Console, because I can vouch for every line of it and I know which step I used to skip. That is a smaller product than a “complete guide” to earning money, and it is the only kind I would put my name to now.

Put that name on it, too. A file whose author is a website and whose tone is certain is exactly the profile buyers have learned to distrust, and they are right to.

Do the fee sum at the real price, on every sales route. Answer each platform’s AI question by its wording, not by how much editing it felt like. Search your own site for the price before launch day, so it is the same everywhere. And read every factual sentence in the product against a source, because the one you skip will be the one a buyer checks.

The order is the real lesson. Do something, write down what it took, and only then decide whether the write-up is worth money. In one week in July I did it the other way round twice, and the other small projects I launched that summer, with what each one took, are in the list of small AI projects I actually launched.