How to make a digital product with AI that people actually pay for

AI can make the product in a weekend. Finding five people who will pay for it is the real work, and AI can help with that too. Here are the real numbers, the order that works, and three prompts and a skill ready to paste.

Playbook

A digital product is a file someone pays to download: a template, a checklist, a short guide, a set of prompts. The five steps below run in the order that works: find the problem, pick the smallest version, get proof, build it, then price it and list it.

Paste each prompt into a new chat in Claude or ChatGPT, swap every line in square brackets for your own details, and send.

The numbers first

  • Gumroad's own figures, the last it published. In its report on 2020, "the median creator made only $70", meaning the middle seller, and the top 1% earned about 60% of the money. It is the most recent report of its kind I could find.
  • An outside estimate. A May 2026 scan of 146,271 public Gumroad products estimated that 44% had made nothing. It is an estimate from public ratings, and Gumroad has not confirmed it.
  • A creator with four shops. One blogger's August 2026 income report shows $9 from her Payhip downloads and under $12 from two Etsy shops. Her own verdict: "Too many shops, not enough marketing."
  • A first sale. Daniil Kornilov (March 2026) needed 54 days and 16 products to make one sale of about $6.

The pattern in all four: the product existed before a buyer did. So the order on this page puts the people before the product.

Step 1. Find a problem people already pay to fix

Go where the people you want to help complain in their own words: reviews of similar products, forum threads, comments under popular posts, messages people have sent you. Copy 20 or more into one document, word for word, with where each came from. Ten is the floor, and remove names and private details from anything sent to you personally.

Kornilov's lesson from those 54 days is the one to carry: "Nobody wants to buy a Notion template. They want to stop feeling overwhelmed by their tasks." (Notion is a notes app.) You are looking for the feeling and the failed attempts, in their words.

AI is good at reading that pile and sorting it. It is also good at sounding sure. As Elham of Elham Growth Studio put it in August 2026, "AI can make uncertain demand sound extremely convincing." So this prompt makes it separate evidence from guesses.

PromptRead what buyers actually said (buyer-problem-research.md)
You are helping me find a problem worth building a small digital product around. Your job is to read what real people wrote, in their own words, and tell me what they are struggling with and what they have already paid for. You are a careful reader, not a cheerleader. Demand can sound convincing when it is not there, so you separate what the text shows from what you are guessing.

FILL THIS IN
- Who I want to help: [one kind of person, e.g. "freelance designers in their first two years"]
- What I know about from my own work or life: [the experience I could put into a product]
- The real text: [paste 20 or more reviews, forum posts, comments or messages from these people, unedited, with where each came from]

If I paste fewer than 10 pieces of real text, or the text is my own summary instead of their words, stop and ask me for more before you do anything else. Tell me two or three places I could collect it, such as reviews of similar products, forum threads, comments under popular posts, or replies in my own inbox.

STAGE 1: WHAT THEY SAID
List every distinct problem in the text. For each one give:
- the problem in their words, as a short direct quote with its source;
- how many separate people mentioned it;
- what they tried already, quoted, and why it did not work for them;
- anything they said they paid for, or would pay for, quoted.
Never write a quote that is not in the text I pasted. If a problem appears only once, say so. If most of the text comes from one source or one thread, say so: one loud thread is not several people.

STAGE 2: EVIDENCE OR GUESS
Sort the problems into two lists:
- EVIDENCE: several people describe it, and at least one mentions money, time lost, or a workaround they built.
- MY GUESS: it sounds plausible but the text does not show it. Say what would turn it into evidence.
If everything lands in MY GUESS, tell me plainly that this text does not show demand yet.

STAGE 3: WHERE I FIT
Compare the EVIDENCE list with what I know about. For each match, say in one sentence what I could make that fixes that one problem, and name the smallest form it could take (a template, a checklist, a short guide, a set of prompts, a short email course). If nothing matches my experience, say so and do not stretch.

STAGE 4: THE NEXT STEP
Pick the single strongest match and write:
- the problem, in one sentence a buyer would nod at;
- three questions to ask a real person about the last time they had this problem;
- what I should NOT assume yet.

NEVER
- Never invent a quote, a number, a market size or a price people pay.
- Never tell me an idea is good because it sounds good. Point to the lines that support it.
- Never suggest a product I could only make by pretending to have experience I did not list.

LAST CHECK
Re-read your answer. Every claim about what people want must point to a quote above. Mark any that do not as [MY GUESS].

Step 2. Pick the smallest version you can make

There are six forms a non-developer can make with AI in a weekend, and each has a real one on sale today, with the price read from its page on 17 September 2026. To choose, ask what the buyer would do with it in the first ten minutes, and pick the form that fits.

  • A Notion template: a ready-made page or system someone copies into their own Notion. Easlo sells a budget tracker for $5 and an all-in-one notes system, on sale at $39, usually $79.
  • A spreadsheet: a Google Sheets or Excel file that does one job. This budget template is listed from $3 and has no ratings yet, which is what most listings look like.
  • A prompt or skill pack: tested prompts or Claude skill files for one kind of work. One pack sells at $39, while another is pay what you want, where the buyer picks the price, even nothing. Many are given away to grow a mailing list.
  • A short guide or checklist: ten to thirty pages that fix one problem. Anna Burgess Yang's freelance pricing guide is pay what you want.
  • A Canva template set: designs the buyer edits in Canva. This set of 300 sells at $39.80.
  • A short email course: five to seven emails, one lesson each, sent automatically. I found no paid example I could verify, so treat this one as unproven.

A Claude skill file counts as an ordinary file, so you can sell it on any platform that delivers downloads. Charging for access to a chatbot is different, and Gumroad's rules ban it.

Step 3. Get proof before you build

Proof means a real person commits to something: a reply saying "I want this", a sign-up, a deposit, a purchase. Kind words do not count. Elham's seven-day sprint (11 August 2026) is the method this prompt follows. Her test for validation is "collecting evidence", and encouragement is not evidence.

The short version: name five real people with the problem, ask each about the last time it happened, send one plain offer, and count who commits. Decide your pass mark before you start, the result you will count as a yes, so you cannot move it afterwards.

The people who wrote the complaints you collected in Step 1 are your first candidates, so reply to them. If you still cannot name five, start with people you have already helped: former clients, colleagues, members of a group you belong to.

Kind words or a commitment? Two made-up examples

Kind words: "Sounds great, let me know when it is out."

A commitment: "Yes, here is my email, send it on Friday." Only the second one counts.

If you take money before the product exists, say what exists today, what is still being made, the delivery date, and that you refund in full if you do not deliver. Some guides tell you to ask AI to play ten imaginary customers instead; made-up buyers prove nothing, and this prompt will say so if you ask.

PromptPlan a seven-day proof test (proof-before-product.md)
You are helping me test a digital product idea before I build it. The test is whether real people will commit to it, not whether it sounds good. Encouragement does not count. A reply, a sign-up, a deposit or a purchase counts.

FILL THIS IN
- The problem, in one sentence: [from my research, in the buyer's words]
- What I would make: [the smallest version, e.g. "a one-page pricing checklist for new freelance designers"]
- Five real people who have this problem: [names or handles, and how I can reach each one. They must be real and reachable this week]
- What I will ask them to commit to: [a reply saying "I want this", joining a waitlist, a deposit, or buying it at a launch price]
- The price I have in mind: [a number, or "not sure"]
- My pass mark, set now: [e.g. "3 of 5 commit", or "10 sign-ups in 7 days"]

If I have fewer than five real people, or the list is groups and audiences instead of individuals, stop and help me find five first. Ask me where these people already talk, and who I have helped with this before.
If I ask you to pretend to be customers, to simulate interviews, or to predict what people will say, refuse and explain in one line that made-up buyers prove nothing.
If my pass mark is blank or vague, ask me what result would make me build and what result would make me stop, and wait. Do not suggest a number for me.

STAGE 1: THE OFFER SENTENCE
Write three versions of one sentence that says who it is for, the problem it fixes, and what they get. Use the buyer's own words from my problem line. No hype words, no promises about results or money.

STAGE 2: THE QUESTIONS
Write five questions for a short conversation with each person. Every question must be about what they did the last time they had the problem: what they tried, what it cost them, what they paid for. No questions that start with "would you".

STAGE 3: THE ASK
Write the message I send after the conversation, under 80 words. It states the offer, the price if I have one, and the exact commitment I am asking for, with a date. It must be easy to say no to. If the commitment involves money, the message must also say what exists today, what is still being made, the delivery date, and that I refund in full if I do not deliver.

STAGE 4: THE SEVEN DAYS
Lay out a day-by-day plan for one week: who I contact on which day, when I send the ask, and when I count the results. Keep each day under 30 minutes. Include one follow-up message, under 40 words, for anyone who has not replied by day five. Send it once. Anyone silent on count day is a no.

STAGE 5: THE DECISION
Give me a table to fill in: person, what they said about the problem, did they commit (yes, no, or no reply), exact words. Then state the rule: at or above my pass mark, I build the smallest version. Below it, I choose one of three: change who it is for, change the form, or stop. Tell me what each result would suggest, without deciding for me before the results exist.

NEVER
- Never count "sounds great" or "I would buy that" as a commitment.
- Never raise or lower my pass mark after the results are in.
- Never invent a response, a statistic or a typical conversion rate.
- Never let me take money without a delivery date and a refund promise.

LAST CHECK
Confirm the plan contacts five named people, asks for one specific commitment, and has a pass mark written before day one. If any of the three is missing, say which.

Step 4. Build it: your parts and AI's parts

  • Yours: the problem you chose, your method, your examples and screenshots, your mistakes, and the final edit.
  • AI's: the outline, the first drafts, the formatting, keeping every page consistent, and a first pass at the sales page.

That split is what working creators do. In Kit's survey of 550 creators (August 2026), 89.2% said they always review and edit what AI writes, and none trusted it without changes.

The all-AI route is the crowded one. A Stony Brook study, not yet peer-reviewed and reported by Fortune on 14 September 2026, found substantial AI text in 20% of 14,419 self-published genre novels on Amazon. The number of books selling grew about twice as fast as the money they made.

This skill asks for your proof and your material before it writes, builds one page for you to approve, and marks every place your own example has to go.

Paste it into a chat as it is, or add it to Claude as a skill so it is there every time: save the download as SKILL.md inside a folder named small-product-builder, zip the folder, and upload the zip at Customize, then Skills.

SkillSmall product builder (small-product-builder.md)
---
name: small-product-builder
description: Builds one small digital product from the user's own knowledge and a proven buyer problem, flagging every gap. Use when the user wants to make a template, checklist, guide or prompt pack to sell.
---

# Small product builder

You help me build one small digital product that a specific person will pay for. The value is my experience and my examples. Your job is structure, drafting and consistency. A product that is all you and none of me is worth nothing to a buyer and may not be protected by copyright, so you keep asking for my material and you mark where it goes.

## Before you write anything, ask me for these and wait

1. The problem, in the buyer's words, and the proof: who committed, and to what. If I have no proof yet, say so, tell me to run a proof test first, and ask whether I want to continue anyway.
2. The form: a template (Notion, spreadsheet or Canva), a checklist, a short guide, a set of prompts or a skill file, or a short email course. If I am unsure, ask what the buyer would do with it in the first ten minutes, and suggest the form that fits.
3. My material: notes, examples, screenshots, a process I follow, mistakes I made. Ask me to paste it or talk it through. If I give you nothing of my own, stop and ask again. Do not fill the gap from general knowledge.
4. The promise, in one sentence: what the buyer can do after using it that they could not do before. No income or results promises.
5. Anything it must not include.

## Weigh the right shape

Say in three or four lines how you will structure it and why, and what you are leaving out to keep it small. The smallest version that fixes the one problem wins.

Say plainly what you can produce here for the form I chose: a finished file if you can create files on this surface, or the content plus an exact build plan for me to assemble in Notion, Canva or a spreadsheet tool, with every page, column, formula and setting in order. Do not claim a limit without checking what you can do. Wait for my yes.

## Build it in this order

1. Write the first section or the first page only, to the standard of something a person paid for. Show me. Ask what is wrong with it.
2. Once I approve that page, match its depth, tone and format for the rest.
3. Wherever my own example, number, screenshot or story belongs, write [NEED YOUR EXAMPLE: what kind] and keep going. Never invent one.
4. Wherever you state a fact, a figure, a law or a tool feature, add it to a list called CHECK THESE at the end, with where I could verify it.
5. Keep a second list called YOURS, naming every part that came from my material or my edits. This is the part I can say I made.

## Self-check before you hand it over

- Does every section help with the one problem, or is some of it padding? Name what to cut.
- Would a buyer get one useful result in the first ten minutes? Say what it is.
- Is there anything a free search or a chatbot would give them in one try? Flag it as weak.
- List every [NEED YOUR EXAMPLE] still open, and everything in CHECK THESE.
- Finish by asking me: would you pay for this, and would you be comfortable putting your name on it?

## Never

- Never invent my experience, my results, a testimonial, a statistic or a price.
- Never promise the buyer income, savings or a guaranteed outcome.
- Never copy the structure or wording of another seller's product. If I paste one as reference, use it only to see what buyers complained about.

Before you list it, the draft has to become a file a buyer can open.

  1. Turn the draft into the thing you will sell. A guide or checklist: paste it into a document, add your examples, and export it as a PDF. A Notion, spreadsheet or Canva template: build it in that tool, then create the share link that lets a buyer make their own copy.
  2. Open the file or link in a private browser window, where you are logged out. If you cannot open it, neither can a buyer. A template link still set to private is the most common failure.
  3. Send it to one of your five people before anyone else. Ask what confused them in the first ten minutes, and fix that.

What you own, and what you must say

  • Pure AI output has no copyright in the US. The US Copyright Office said in January 2025: "Copyright does not extend to purely AI-generated material", and "prompts do not alone provide sufficient control". Other countries differ, so check your own.
  • Your own work is protected. The same report protects a human's "creative selection, coordination, or arrangement" and their edits, which is another reason your examples and structure matter.
  • The AI companies let you sell what you make. Anthropic assigns you its rights in Claude's output. OpenAI says you own ChatGPT's output, and that you may not "represent that Output was human-generated when it was not".
  • Some platforms make you declare it. Amazon's book platform requires you to tell it about AI-generated text and images. Say how you used AI on your sales page wherever you sell.
  • Canva has its own rule. You cannot sell an editable template that uses Canva's paid elements as a PDF or any other file, so share it as a Canva template link. A finished guide or printable you designed in Canva is allowed.

Step 5. Price it and choose where to sell

Price. Start from what your five people said yes to in Step 3. For reference only: Stan's 2026 report puts the average digital download sale on its platform at $51 in 2025. That is an average across 80,000 sellers, many with large followings.

Tax, in one idea. Some platforms act as the legal seller, known as a merchant of record. They collect and pay sales tax for you, including VAT, the sales tax used in Europe and elsewhere. On the others, you are the seller, and you register and file with the tax authority where you live.

Fees below were read from each platform's own page on 17 and 18 September 2026. Check the page again before you sign up.

Gumroad

Cost: 10% plus 50 cents per sale, or 30% if the buyer finds you through Gumroad's own marketplace. No monthly fee. (Pricing page)

Tax: Gumroad acts as the seller for tax, so it collects and pays sales tax for you.

Worth knowing: it bans selling access to AI tools and chatbots, and reselling products you bought the rights to. A file you made with AI is fine.

Lemon Squeezy

Cost: 5% plus 50 cents per sale. (Pricing page)

Tax: it also acts as the seller for tax and handles it for you.

Worth knowing: it said in January 2026 that it is moving sellers to a new payments system, and support has been slower, so read recent reviews first.

Payhip

Cost: a free plan that takes 5% per sale, or paid plans with a lower cut. Card processing fees are extra. (Pricing page)

Tax: it collects VAT for European buyers. Any other sales tax is yours to set up and file.

Worth knowing: the free plan is the cheapest way to open a shop with no monthly cost.

Kit

Cost: the free plan, up to 10,000 subscribers, lets you sell digital products for 3.5% plus 30 cents per sale. (Pricing page)

Tax: its pricing page does not say it handles tax, so assume it is yours.

Worth knowing: it is an email tool first, which suits you if your buyers will come from a mailing list.

Stan

Cost: $29 or $99 a month with no fee per sale, and no free plan after the trial. (Pricing page, July 2026)

Tax: it can calculate tax in some countries, and you still file it yourself.

Worth knowing: the monthly fee only makes sense once sales cover it.

Etsy

Cost: 20 cents per listing, which renews every four months, plus 6.5% per sale and a payment processing fee. (Fees policy, updated 13 February 2026)

Tax: Etsy collects US sales tax, and VAT on digital downloads in some countries. Everything else is yours.

Worth knowing: buyers browse Etsy, which the other five lack, and it is crowded with AI-made listings.

Which one, then

  • You want tax handled for you: Gumroad or Lemon Squeezy, the two that act as the seller.
  • Your buyers will come from a mailing list: Kit, since the list and the shop sit in one place.
  • You have nobody to tell yet: Etsy has browsing buyers, and a crowd of AI-made listings to compete with.
  • Whatever you pick: check that it pays out to your country, and mind the fixed fee at low prices. On a $5 sale, Gumroad's cut is $1.

This last prompt checks your sales page for anything unsupported, writes your AI-use line, and gives you the questions to confirm on your platform's own pages. It will not quote fees or tax rules from memory.

PromptCheck my sales page and setup (sales-page-check.md)
You are checking my sales page and my selling setup before I publish. You are the last reader before a stranger decides whether to trust me with their money. Be blunt about anything unclear, unsupported or too good to be true.

FILL THIS IN
- The product and who it is for: [one sentence]
- What is inside: [pages, templates, files, lessons, and the format]
- The proof I have: [who committed in my test, in their words, or "none yet"]
- The price: [number and currency]
- Where I plan to sell it: [platform name, or "not decided"]
- Country I live in, and state or province if it has one: [fill in]
- How I used AI to make it: [e.g. "AI drafted the outline and first drafts, I wrote the examples and edited everything"]
- My refund position: [e.g. "refund within 14 days, no questions", or "no refunds on downloads", or "not decided"]
- My draft sales page, if I have one: [paste it, or write "none, please draft one"]

If the price, the platform or the refund position is missing, ask me about each before you continue, and give me two or three sensible options for each with the trade-off in one line. If my price differs from what the people in my test agreed to, ask me why before continuing.

JOB 1: THE PAGE
If I pasted a page, go through it line by line. If I did not, draft one under 300 words. Either way it must have: who it is for, the problem in their words, exactly what they get, what it does not do, the price, the refund line and the AI-use line. Mark every claim that has no support as [UNSUPPORTED] and tell me to cut it or prove it.

JOB 2: THE BANNED LIST
Remove or flag: any promise of income, savings or results; "guaranteed"; fake urgency; invented testimonials; numbers I did not give you; anything that implies more experience than I listed.

JOB 3: THE AI LINE
Write one honest sentence saying how AI was used, based only on what I told you. Remind me that some platforms require this, and that I must not present AI output as human-made.

JOB 4: THE PLATFORM QUESTIONS
Do not tell me fees, tax rules or platform policies from memory. Instead, give me a checklist to confirm on my chosen platform's own pricing and terms pages today:
- the fee per sale and any monthly cost;
- whether the platform acts as the seller for tax purposes and collects sales tax for me, or whether I must register and file with my own tax authority;
- whether it pays out to my country;
- its rules on AI-made products and on refunds.
For tax, tell me to check with the tax authority where I live, and mark it [VERIFY WITH YOUR TAX AUTHORITY].

JOB 5: THE FIRST TEN BUYERS
Write a short plan for telling the people who already know me: the five from my test first, then anyone who has asked me about this topic. Draft the message, under 80 words, no hype.

NEVER
- Never state a fee, a tax rate, a threshold or a legal right as fact. Mark it for me to verify.
- Never write a testimonial or a result I did not give you.
- Never add urgency that is not true.

LAST CHECK
Read the final page as the buyer. Can they tell in ten seconds what they get, what it costs and what happens if it is not for them? If not, say which of the three is unclear.

The honest bit

  • I cannot promise you a sale, and nobody honest can. Most of the platforms publish no figures on what a typical seller earns.
  • An audience is the biggest difference. In Gumroad's report, 60% of its higher earners kept an email list, against 31% of the rest. Five real people this week is how a list starts.
  • Be wary of who is teaching this. Plenty of the people earning from digital products are selling courses on how to sell digital products.

Start this week

Today, collect 20 complaints from the people you want to help and run the first prompt. Tomorrow, write down five names. By next week you will know whether to build.

If you are still choosing what kind of side income suits you, start with the six that are worth your time.

Questions people ask

Which AI is best for creating a digital product?

Claude or ChatGPT. The three prompts work in either, and the skill file is written for Claude. In ChatGPT, paste the skill's text at the top of a new chat, or put it in the instructions of a Project.

Can ChatGPT make me money?

It can help you make a small product to sell, and the money depends on finding buyers first. Most sellers earn little, so get five real people to commit before you build.

What if nobody commits?

That is the test working. Before you drop the idea, change one thing at a time: who it is for, or the form. If a second round of five also says no, the problem is real but people will not pay you to fix it, and you have lost a fortnight instead of a season.

What refund policy should I offer?

Decide it before you list, and print it on the page. A no-questions refund within a set number of days removes the buyer's main fear and costs you little on a cheap file. Some countries give buyers of digital goods a legal right to cancel, so check with your consumer regulator.