Selling custom GPTs, AI agents and agent tools

The research was done by AI agents that open web pages. Some sites refuse them; where that happened we say so. The small numbers point to the source list at the end. Where a figure is our own sum or guess, it is marked “our estimate” and the basis is given.

The claim

Our collection for this scheme holds only 4 tweets, from 3 accounts¹. That is a small collection, so this page also rests on how the method is promoted generally: on the platforms’ own pages and on what sellers of courses and tools publish. View counts are as collected. Dates are worked out from each tweet’s ID number. The page judges the scheme, not the people who tweeted.

“here’s how to make money selling to ai agents on autopilot…

agents are the ones using software now, and they only use what they can call, so build that and you get paid every time it runs

• scrape what businesses already pay to have done
• ask claude which of those an agent would call
• claude builds the endpoint and deploys it
• add docs, a key, a cap and a receipt
• list it where agents search for tools
• the business behind the agent pays per call

just wrote a whole guide on how you can do this:”

@everestchris6, 15 September 2026, 75,706 views¹⁶⁷. The guide is an article on X in another post by the same account. We did not read it.

“This made me quit my 9-5 job for GOOD. CUSTOM GPT’S. You can make THOUSANDS selling them, and the best part is literally anyone can learn it. Like + reply ‘GPT’ and I will DM you the full guide!”

@Zephyr_hg, 25 November 2024, 38,145 views¹

“No-code AI agents you can actually build, publish, and earn from? This move from AITOPIA is pretty interesting. Turn your idea into an AI agent, put it to work, and potentially make money from it.”

@choyamymuna, 24 August 2026, 4,097 views¹. This tweet quotes an announcement by AITOPIA, a subscription service for AI apps, of its “Agent Builder”. The announcement says: “Earn 70% revenue share from your agents”⁶⁷.

The fourth tweet is a near copy of the second, from the same account in December 2024, with 16,029 views¹.

The only income figure in the four tweets is the word “THOUSANDS”¹. The only rate is the 70% share in the quoted AITOPIA announcement⁶⁷. In the text we read, no tweet gives a price per GPT or per call, a number of buyers, the hours needed or the running costs¹⁶⁷. We did not view the tweets’ images or videos, the guides, or the threads under them, so some of that may be there. The first tweet does list building steps and says a business pays per call; it does not say the work needs no skill⁶⁷.

What the scheme is

Three different things are sold under this heading. They need to be kept apart, because the rules and the evidence differ for each.

Custom GPTs. A custom GPT is a version of ChatGPT that someone has set up with their own instructions and files, for one job, without writing code. The pitch is to build these and sell them to businesses, or to publish them in OpenAI’s GPT Store and be paid for use.

AI agents. An AI agent is a program built on an AI model that carries out a task in steps, such as answering customer emails or finding sales leads. “No-code” means it is assembled in a visual tool, with no programming. The pitch is to build agents and sell them to clients, or publish them on a platform that pays.

Agent-callable tools. An AI agent cannot do much alone. It “calls” outside tools: small services that fetch data or perform an action. Many are published through MCP (Model Context Protocol), a standard way for an AI assistant to connect to outside tools. The pitch is to publish a tool on a marketplace and collect a small fee each time it runs.

All three mechanisms exist or existed. The questions are who can still use them, what they pay, and how many sellers get any buyers.

The arithmetic

The tweets give almost nothing to multiply, so the figures below come from the platforms and from people who published results.

Step 1: can you still make and sell a custom GPT?

Mostly not, and not for long.

OpenAI’s help page says: “Personal ChatGPT accounts, including Free, Go, Plus, and Pro, cannot create or publish new GPTs.”² Only Business, Enterprise and Edu workspaces can²⁴. The page was updated around 28 September 2026².

OpenAI also plans to close the product. Its help centre says: “Custom GPTs are scheduled to retire on Dec 11, 2026.” Enterprise workspaces with an approved delay get until 11 February 2027³. On retirement “custom GPTs and their GPT pages become inaccessible”³. These are plans: the Enterprise section of the same page says “the dates are subject to change”, and the help page on building GPTs says other plans “are expected to follow the same timeline”²³. The Enterprise section also gives 26 October 2026 as the planned end of creating new GPTs in affected Enterprise workspaces³.

Creators are told to move to “plugins”, OpenAI’s replacement format³. The move does not carry the audience: “Migration doesn’t carry over your GPT’s sharing settings or give its existing users access to the replacement. A migrated personal plugin starts private.”³ Connections to outside services (called custom actions) do not transfer automatically either³.

Counting from today, 10 October 2026, a custom GPT has about two months left on that schedule (our sum from the dates in³). A GPT sold to a client stops working on the retirement date unless it is rebuilt³. The two “THOUSANDS” tweets date from late 2024, before any of this¹.

Step 2: what did the GPT Store pay?

For almost every builder, nothing that we could find.

In January 2024 OpenAI wrote: “In Q1 we will launch a GPT builder revenue program. As a first step, US builders will be paid based on user engagement with their GPTs.”⁵ This is where the “earn from your GPT” claim comes from. The same post said over 3 million custom GPTs had been created⁵.

What followed was a small test. A news report from March 2024 said OpenAI “has partnered with a small group of developers for this initial test”⁷. A member of OpenAI’s own forum quoted the company’s help text in June 2024: “Selection is currently limited to a select group of US-based builders who have created popular and engaging GPTs in the GPT Store. We are not currently accepting additional builders.”⁶ We did not see that help text ourselves; its address now redirects elsewhere⁴⁶. In December 2025 another forum member wrote: “Nothing. They’re still testing with US based users. You need your own way to monetize GPTs, otherwise nothing.”⁶

We found no published rate per conversation or per user, and no figure for what the test group was paid⁵⁶. So this part of the arithmetic has no price to multiply.

Step 3: what a custom GPT sells for when built for a client

The only GPT income we found documented came from building one for a paying client. Asking prices are low:

• One freelancer’s listing on Contra, a freelance marketplace, has three tiers at $40, $100 and $250, with two-day delivery on the cheapest¹⁷.
• An Upwork listing was shown in search results with tiers at $80, $300 and $800 and delivery in 7 to 20 days. Upwork refused our tools, so we did not read the page itself¹⁸.
• A forum member wrote in November 2024: “Just sold 2 CustomGpts each für 800 EUR.” This is that person’s own statement, and the price is in euros¹³.

These are asking prices. No page we read showed how many orders the listings get¹⁷¹⁸.

Agents built for businesses are priced differently, and our page on AI automation agencies covers that route in full. It found that 42 of 43 established firms in the partner directory of n8n, an automation tool, say they accept no project under $1,000, while one open job board’s average bid was $102⁷⁵. The last row of the table uses that $1,000 floor.

Jobs needed each month (our sums: the target divided by the price, rounded up):

At $800 to $1,000 a job, $10,000 a month means finding a new client about every two or three working days. That is freelance work with a sales task attached, not income “on autopilot”. We found no measured source for how long each job takes or how many pitches lead to a sale.

Step 4: what a tool marketplace pays per run

The clearest working example of “get paid every time it runs” is Apify Store. Apify is a platform where developers publish small programs, which it calls Actors, and users pay to run them. Most are scrapers: programs that collect data from websites.

Apify’s terms say: “Your payout for a monetized Actor will be calculated as 80% of the fees paid by Users for your Actor, minus Platform usage costs for the corresponding Actor runs”²⁵. Its documents put it as a formula: “profit = (0.8 * revenue) - platform costs”, with a worked example of $31 charged, $4 of costs and $20.80 kept²⁴. Runs by users on the free plan earn nothing, and the cost of any outside AI model the tool uses is the developer’s own²⁵.

So a $1,000 monthly payout needs more than $1,250 of charges to users, and $10,000 needs more than $12,500 (our sums: the payout divided by 0.8, before running costs).

AI agents can pay for these tools without a human account. Apify’s documents say: “Agentic payments let AI agents run and pay for your Actor without an Apify account.”²⁷ The minimum payment is $1 in USDC, a digital dollar, and the developer must have passed an identity check²⁷. In an earlier download of the store’s public data, 8,175 of 10,162 listings were marked as eligible for agent payments³⁰. So the mechanism in the most-viewed tweet exists.

Step 5: how many tools get users

This is the step the tweets leave out, and it decides the result.

Apify’s store lists 83,526 Actors²⁸. We downloaded its public listing data on 10 October 2026. The data service would not return every listing, so our download covers 63,617 of about 83,000. The method is described in the source entry so that it can be repeated³⁰. The figures move by about a percentage point from one download to the next, so read them as “about”.

Among 61,900 paid listings (our count from³⁰):

The median listing had 1 user in 30 days. “Median” means the middle one: half had more and half had fewer. The top 1% of listings had about 79% of all the users counted, and the top 10% had about 93%³⁰.

The per-listing figure is pulled down by mass publishing. Some accounts publish hundreds of listings each, and Apify’s contest (see “The upside”) entered “the first five Actors published by January 31, 2026” for each person who registered³⁰⁷². A fairer measure of “what happens to a person who tries” is the count per developer (our count from³⁰):

The “about 450” row matters because it is roughly the user count that would go with $1,000 a month, if payouts follow users. That is our estimate, worked out under “The upside”. The top 10% of developers had about 95% of the users counted³⁰.

More tools go with more users (our count from³⁰; listing numbers here are those in our download, which misses some):

Listings tagged as MCP servers and as agents did no better. In both groups the median was 1 user in 30 days, and 42% of the MCP group had no users at all³⁰.

Limits: a user count is not income. It may include a developer’s own test runs. One person using two of a developer’s tools is counted twice. Apify’s own account is in the totals and holds about 27% of all the users counted³⁰. Apify itself told its developers, in an announcement shown as two days old when we read it, that it had “seen a rise in attempts to manipulate Apify Store”, including “Artificially inflated metrics”, so some counts near the top may be inflated⁷³.

Custom GPTs showed the same shape when they could be published. A study that crawled about 334,000 GPTs in early 2024 found: “A small 28.5% of custom GPTs have solicited over 10 conversations.”⁸ So 71.5% had 10 or fewer. A small group of featured GPTs, 0.36% of those recorded, held about 33.7% of all conversations, and 90.78% of GPTs had no ratings⁸.

Step 6: what a “paid per call” price needs

Agents can also pay a tool directly, with no marketplace. The best-known method is x402, a way for software to pay per request using digital dollars. Its site says: “x402 is free for the customer and the merchant—just pay nominal payment network fees”³⁶.

Sources disagree on the average payment, so the table uses several (our sums: the target divided by the price; each average is a source’s volume divided by its transactions):

Step 7: how much agents actually spend

The headline numbers for agent payments do not agree with each other, and independent checks cut them sharply.

• The x402 site shows, for the “Last 30 Days”, “75.41M Transactions $24.24M Volume 94.06K Buyers 22K Sellers”³⁶. An article from September 2026 reports that the same numbers had been on display for months, which would mean the counter is not live. We confirmed only that today’s numbers match the ones in that article³⁶⁴¹.
• A public explorer of x402 payments shows, for the past 30 days, 2.69 million transactions, $1.08 million of volume, 36,150 buyers and 40,000 sellers³⁷. That is about $27 per seller a month on average (our estimate: $1.08 million divided by 40,000).
• TRM Labs, a firm that analyses blockchain payments, traced $52.7 million of x402 payments since May 2025. It removed payments from sellers to themselves, bulk flows from one or two payers, and sellers with fewer than ten different buyers. That left $25.62 million. It estimated that 0.6% to 7.5% of that looked like AI agents paying, and wrote that “the volume they move has stayed small, roughly USD 5,000–11,000 a month”. TRM says its test may miss some agents³⁸.
• A report by Visa and the data firm Artemis, paid for by Visa, found raw volume of $135.7 million shrank after removing test payments and sellers paying themselves: “it has processed roughly $15.0 million in adjusted volume across 109.6 million transactions”, with data to 21 April 2026³⁹. A news report on the same work says: “The top 1% of x402 buyers, roughly 4,000 wallets, drive about 90% of adjusted volume”⁴⁰. A wallet is a payment account.

These sources conflict on totals. They agree on the direction: real spending is far below the raw counters, and it comes from a few thousand buyers. We found no figures on how the money divides between sellers. The explorer has a list of services it has picked out, ordered by number of transactions. We read the first page, 15 services out of 24 pages. On that page the highest 30-day volume was $54,760, and one service showed 420,000 transactions for $936.99³⁷.

Step 8: the running costs

Costs in money are small. An automation tool such as n8n or Make costs tens of dollars or euros a month; we saw those prices only in a summary, not word for word⁶⁶. If an agent is built on OpenAI’s paid service for programmers, the builder pays the model fees²³. The real cost is unpaid time: building, finding buyers and keeping tools working. Apify’s documents say to “reserve approximately 2 hours per week for bug fixes, updates, and user support”⁷¹.

What people who tried it report

Custom GPTs. We found no income report from a custom GPT seller that was checked by anyone else.

• On OpenAI’s forum, a builder with “over 80” GPTs in the store wrote in September 2024: “Nope I make 0$ so far but I do get users.” Six weeks later he reported about 90 GPTs and “6000+ users in 6 months”, without mentioning income¹³. This and the two sales at €800 in Step 3 are people’s own statements¹³.
• The day after the store opened, a forum member guessed at about $12,000 a year for a successful GPT and concluded that “revenue share will not be the main source of income for successful creators”¹⁴.
• A prediction market, where people bet on outcomes, asked whether anyone would be reported to have made $1 million or more in 2024 from a custom GPT. It was resolved as “no”. Only 7 people bet, so this is weak evidence¹⁵.
• One published figure exists: a post titled “How I Made $6,248 With Custom GPTs (No Code, No Audience)” from April 2025. We saw only the title in a search result. The page would not open, so we do not know whether the figure is sales or profit, or what proof it shows¹⁶.
• Two studies describe the store itself. One found that “some GPTs exhibited abnormally high dialogue counts, reaching thousands or even tens of thousands, while the corresponding number of ratings remained significantly lower”, which its authors treat as suspected manipulation of rankings⁹. Another counted copies across six AI app stores, “identifying 3,509 squatting apps, and 9,575 cloning apps”¹⁰. Squatting means taking a name close to a popular app’s name.

Tools on Apify. People are paid, slowly and in small amounts at first.

• A developer quoted on Apify’s own page, in a testimonial dated 1 January 2026, said he had published 37 tools over 15 months and that, if he keeps going, “it could turn into something bigger”. The page shows no amount²⁸. An account that appears to be his now lists 93 tools, with about 2,900 users in the last 30 days. That is roughly 45th among developers by users. The match between the testimonial and the account is ours, from the handle, and is not confirmed²⁸³⁰.
• A developer wrote on Apify’s blog in July 2026 about his first six months. Working “15 to 20 hours per week”, he reached “98 public Actors, 2,500 total users, 855 monthly active users”. A full week passed before anyone ran his first tool. He declined to give revenue, saying only that in months 3 to 4 “Revenue became meaningful but inconsistent”, and concluded: “Three reliable Actors out-earn ten unmaintained ones”⁷¹. The article was written for a programme run by Apify. The account it links to now shows 167 listings and about 790 users in 30 days³⁰.
• Another testimonial on an Apify page reads: “I was making $500 a month on other side projects, but now Apify is bringing in more than $2,000 from Apify Store.” The text we received did not name the speaker. The statement was chosen by the platform³².
• A two-tool Apify business launched in 2025 was put up for sale in March 2026 at an asking price of $5,000. The seller’s listing says it had 4 customers and “March 2026 Net Profit: $157.58 (Highest recorded month)”. The seller’s figures do not agree with each other: the headline shows “$240 ARR” (yearly revenue) while the text projects $2,400 a year³³.
• Apify’s documents say promotion is what separates tools “with many paid users and those with few to none”²⁷.

Paid tools sold directly.

• The founder of Ref, a paid tool that searches programming documents for AI coding agents, wrote that three months after launch at $9 a month, “Ref has thousands of weekly users and hundreds of subscribers”. The same post says “MCP is new and nearly everything else is free”. This is the founder’s statement, with no revenue figure⁴².
• A post about a tool that AI agents pay showed, when opened, “six settlements and counting” at $0.005 to $0.01 a call, on a test network that uses play money⁴⁴.
• A search result described a developer who launched a paid tool at $0.07 a use and had no paying users two weeks later. We could not check it: the site has closed⁴³.

Agents built for businesses. A founder who built agents for manufacturers wrote in May 2025 that after one paying customer, “getting our second and third customers has been way harder than we expected”⁴⁵. The difficulty reported was selling, not building.

What the rules allow now

OpenAI

• Custom GPTs are closed to personal accounts and are scheduled to retire on 11 December 2026²³. See Step 1.
• The replacement cannot be sold inside ChatGPT. OpenAI’s rules for plugins say: “Currently, plugins may conduct commerce only for physical goods. Selling digital products or services—including subscriptions, digital content, tokens, or credits—is not allowed, whether offered directly or indirectly (for example, through freemium upsells).”¹⁹ OpenAI says it is “actively working to support a wider range of commerce use cases”. What remains allowed is selling access on your own website and letting paying customers sign in²⁰.
• Publishing needs checks. “All plugin submissions must come from verified individuals or organizations”, and “Previously approved plugins that are later found in violation may be removed.”¹⁹ The submission page describes no payment to developers²¹.
• Removal happens. A research group collected 755,297 GPTs, flagged 2,051 as misused, and reported that “1,316 out of 1,804 GPTs reported in the first disclosure were removed” within about two weeks¹².
• Some popular niches are barred. OpenAI’s usage rules exclude “provision of tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional”, and breaking the rules “may mean you lose access”²².
• A GPT is easy to copy. A 2023 study that tested over 200 GPTs found: “Through prompt injection, an adversary can not only extract the customized system prompts but also access the uploaded files.” Prompt injection means typing instructions that trick the GPT into revealing its setup. We did not check whether this still works¹¹.
• Selling an agent built on OpenAI’s paid service puts the risk on the seller. The terms say: “Customer may not resell or lease access to its Account or any End User Account”, and “Customer is responsible for all activities that occur under its Account, including the activities of End Users”²³.

Tool marketplaces and payment services

• Apify payouts have conditions. An identity check is required. The minimum payout is $20 by PayPal or Wise, or $100 by other methods. Bank transfer fees of $10 to $50 come out of the payout²⁶. The terms add: “Any accrued payout that remains below the Minimum Payout for a continuous period of twelve (12) months shall be deemed abandoned and forfeited in favor of Apify.”²⁵ Payouts can be held during a review: “If Apify detects suspicious usage patterns related to your Actors, your payout might be withheld during the investigation.”²⁶
• Inflating numbers can end an account. Apify’s terms forbid developers to “use multiple accounts to influence ratings, review your own Actor(s), or otherwise artificially inflate or suppress ratings or reviews”²⁵. An announcement to its developer community, shown as two days old when we read it, says “a single confirmed violation can result in a permanent ban from the platform, removal of your Actors from Apify Store, and forfeiture of your payouts”⁷³. We read that announcement through a reader service and could not confirm its own address or exact date.
• Tools must be kept working. “If the Actor remains a Faulty Actor for 30 days or more, we are authorized to mark it as ‘deprecated’ and/or remove it from Apify Store.”²⁵
• Stripe, a payment company, offers pay-per-call payments for tools: “For stablecoin payments, the minimum amount is 0.01 USDC.” Stablecoins are digital dollars. Card payments have a minimum of $0.50. The service is open to businesses “in all US states, except New York”⁶⁴.
• Anthropic’s directory for tools that connect to its Claude assistant does not accept “Software that transfers money, cryptocurrency, or other financial assets” or software that serves adverts. It is a place to be found, not a place to be paid⁶³.
• AITOPIA, the platform in the third tweet, can end an account “for any or no reason” under its terms, with a refund “for any pre‑paid portion of your unused Subscription”⁴⁹. Clauses like this are common. We found nothing in those terms about a creator’s unpaid earnings when an account ends⁴⁹.

Legal risk

• Scraping. The tools with the most users on Apify collect data from large websites (see “What it takes to compete”). Many sites forbid this in their terms. LinkedIn’s user agreement, for example, tells users not to “Develop, support or use software, devices, scripts, robots or any other means or processes (such as crawlers, browser plugins and add-ons or any other technology) to scrape or copy the Services”⁷⁴. Apify’s terms allow a developer to publish a tool “of any kind, provided that you comply with these Publishing Terms, the General Terms (including the Acceptable Use Policy), and all applicable laws and regulations”²⁵. So the person who publishes the tool carries that duty. We did not check how courts in any country have treated the scraping of public pages, or the terms of other sites.
• Income promises. The US Federal Trade Commission (FTC) sued Air AI in August 2025. It alleged the company sold access to a “conversational AI” product and licences to resell it, at $9,800 to $100,000 for licences, with earnings promises, and that “most consumers are not able to make any money using Defendants’ Services and lose thousands of dollars”⁵⁷. These are allegations. In March 2026 the FTC announced a proposed settlement that would ban the company and its owners from marketing business opportunities, with “a monetary judgment of $18 million” that is largely suspended and $50,000 paid⁵⁸. The case page still showed it as pending when we read it. A settlement is not a court finding⁵⁸.
• In September 2024 the FTC announced several actions over AI claims. Some were against sellers of AI-branded online-shop packages. In one of them, against Ascend Ecom, the FTC said its complaint notes that “for nearly all consumers, the promised gains never materialize”. That is an allegation. The FTC’s Chair at the time said “there is no AI exemption from the laws on the books”⁵⁹. Those companies sold online shops, not GPTs or agent tools⁵⁹.
• These cases concern only the companies named in them. We found no regulator or court action about selling custom GPTs or agent tools on a marketplace.
• FTC guidance for anyone selling a business package in the US says: “it’s illegal to make an earnings claim unless you have written materials on hand that back up what you’re saying.”⁶⁰
• In the EU, a chatbot or agent that talks to people must say it is an AI. A law firm’s note says this duty has applied since 2 August 2026, with fines of up to €15 million or 3% of worldwide turnover⁶¹.
• In the US, an agent that makes phone calls falls under robocall rules. The telecoms regulator ruled in February 2024 that “calls made with AI-generated voices are ‘artificial’” under the relevant law⁶².
• US tax applies as to any self-employed income; marketplaces report payments that “total over $20,000 in more than 200 transactions” to the tax authority⁶⁵.

Who makes money from it

The platforms. Apify keeps 20% of what users pay and also bills the running costs²⁴²⁵. MCPize, a marketplace for paid MCP tools, also keeps 20% and pays out above $100. Its page gives income ranges for creators, which are its own projections, not payout figures⁵⁴. The payment services earn on each transaction⁶⁴. Poe, an AI chat app, says in its creator documents: “The creator monetization program lets Poe bot creators get paid”⁷⁰. We found no payout data for it.

The platform in the third tweet. AITOPIA’s home page presents a subscription shop: “Top AI Apps, Models, and Agents. One Subscription.” and says it is trusted by over 3 million users⁴⁸. It publishes two separate ways to earn:

• A share when people run your agent. Its Agent Builder page says: “Every run of your agent spends credits, and a share of that spend becomes the distributable net. You take 70% of it and the platform keeps 30%.” If someone else’s reworked copy of the agent is run, the split is 50% to the first creator and 20% to the person who reworked it. Joining needs “no applications, no follower thresholds”⁶⁹. So the tweet’s “build, publish, and earn” has a published mechanism behind it.
• A referral commission. Its “Earn” page reads: “Add your referral link to your profiles. Earn 40% commission for every subscription to AITOPIA plans.”⁴⁹

What is missing is any record of results. The page does not say how large “a share of that spend” is, and we found no payout figures, minimum payout or count of paid creators⁶⁹. Its calculator shows an “Estimated Annual Earning” of $16,800 for “10 Agent, 100 Credits”; that is the platform’s own projection, not a payout⁶⁹. The bio of the account that posted the tweet begins: “AI & Tech content expert. Ghostwriter, Passionately here to help promote your brand. Dm to collaborate”⁶⁷. We do not know whether this post was paid for.

The accounts in the other tweets. We write what we saw on their pages today, not what they offered when they tweeted.

• The @Zephyr_hg profile links to a free daily email newsletter that states “13,000+ readers”⁵⁰. The same site offers a paid community about working with the Claude AI assistant, at $147 a month for the first members, and a Claude course⁵¹⁵². None of these pages mentions custom GPTs or makes an income claim⁵⁰⁵¹⁵². We did not see the guide the 2024 tweets offer.
• The @everestchris6 profile links to atonomi.ai, a site that offers custom AI agents for finding sales leads to businesses through a booking call. It shows no prices; one of its questions is “Why isn’t pricing listed?”⁵³ That is a service built for clients, which is a different model from the pay-per-call tool in the tweet.

Teachers and communities. Teaching a neighbouring method is a business of its own. One paid community that teaches selling AI automation to clients, the nearest taught version of this scheme, charges $129 a month and shows about 3,400 members (displayed as “3.4k”)⁵⁵. Its refund line reads: “Annual members can request a refund within 7 days. For monthly members, all sales are final.”⁵⁵ If every member paid the monthly price, that would be roughly $440,000 a month (our estimate: 3,400 x $129). They do not all pay it, since there is an annual plan, so the true figure is unknown.

Tool vendors pay for referrals. Skool, which hosts paid communities, tells promoters they “earn 40% of their monthly subscription fee for life” for each community owner they refer. Among automation tools, GoHighLevel advertises a 40% recurring commission; Make advertises 35% and n8n 30%, each for the first 12 months⁵⁶. So a person who writes about building agents can be paid when a reader signs up for a tool, whether or not the reader ever sells an agent.

Builders with users. A few hundred developers on Apify, and some sellers of specialist tools, are paid. The next section covers them.

The upside

The largest payout pool we found, and its trend

Apify says: “$1.6M paid out last month. Many developers earn over $3k.”²⁸ These are the company’s own statements and are not audited. The same day its page for MCP developers still showed “$500k+ payouts to developers every month”³². Apify publishes no median and no count of how many developers reach each level.

The company’s own figures rise over time:

So the pool roughly tripled in a year on Apify’s telling (our sum: $1.6 million divided by $563,000 is about 2.8). We did not find how fast the number of developers and listings grew over the same period, so a bigger pool does not show a bigger share for each developer.

Apify also ran a “$1M Challenge” that closed on 31 January 2026. Apify’s announcement offered “Up to $2,000 per Actor based on monthly active users”, and the top three prizes of $30,000, $20,000 and $10,000 were awarded by a jury³¹⁷². The three winning accounts now have about 396, 83 and 99 listings³⁰³¹.

What the average and the top tenth get (our estimates)

• Apify states 4,500 community developers²⁸. $1.6 million divided by 4,500 is about $356 each a month.
• Averages hide the split. Our download found 3,337 developer accounts with paid listings, leaving out Apify’s own account³⁰. If payouts follow user counts, which is an untested assumption, the top 10% (about 330 developers) would share about 93% of the pool, or roughly $4,500 each a month. The other 90% would share about 7%, or roughly $37 each a month²⁸³⁰.
• On the same assumption, $1,000 a month goes with about 450 users in 30 days (our sum: $1.6 million divided by the 724,790 users counted outside Apify’s own account is about $2.20 a user). About 200 developers, 6%, had that many. 124 had 1,000 or more³⁰.

No source gives real payout figures by rank, so the top hundredth cannot be stated in dollars. Apify’s own “over $10,000” for the most successful creators is the nearest thing²⁹.

The best single cases

• A paid MCP tool with payment-checked revenue. “Advanced GSC MCP” connects an AI agent to Google’s search data for website owners. It was founded in April 2026. TrustMRR, a site that reads revenue from a seller’s Stripe payment account, shows “$4,632/month in verified Stripe revenue. $4,717 MRR.” with 143 subscribers and $14,841 in total since launch. MRR means monthly recurring revenue: what subscribers pay each month. Plans cost $12 to $100 a month³⁴. This is revenue, not profit, and sellers choose to be listed.
• How typical that is. We looked for TrustMRR listings with “mcp” in the name and found 8. Their revenue for the last 30 days was $4,632, $348, $277, $30, $0, and three with none shown³⁵. So 1 of 8 was above $1,000 a month. The sample is tiny.
• Ref. “Hundreds of subscribers” at $9 a month would be $1,800 to $8,100 a month if “hundreds” means 200 to 900 (our estimate; the founder gave no figure)⁴².
• An expert assistant sold to professionals. CustomGPT.ai, a company that sells a platform for building AI assistants, reports that a customer called TaxWorld, an Irish tax-law assistant built on expert material, reached €1 million a year with “740 paying customers, only 8 cancellations”, and says it was done “with zero engineering budget”⁴⁶. That is about €1,350 per customer a year (our sum). The figures are the vendor’s and were not checked. It was not a GPT Store product.
• Demand for hired work. Upwork reported in February 2026: “Skills that explicitly reference AI grew 109% year over year.”⁴⁷ That is demand across the site, not income.

A realistic good result in the first year or two (our estimate)

These ranges are our estimate from the usage data in Step 5 and the handful of cases above. They are not a measured income distribution.

• A custom GPT or a generic no-code agent listed on a store: about nothing. The GPT format is scheduled to close, and the middle listing in Apify’s agent and MCP groups has 1 user a month³³⁰.
• Building GPTs or agents for clients: a few hundred dollars a job for a GPT, and $1,000 or more for an agent project, for someone who can find clients¹³¹⁷⁷⁵.
• A programmer who builds and maintains tools on a store: tens of dollars a month at first. The median developer with 10 to 40 tools has 29 users, which suggests about $65 a month on the assumption above. About 1 in 12 of that group has enough users to suggest $1,000 a month; among developers with more than 40 tools it is about 1 in 3²⁸³⁰.
• A specialist paid tool sold by subscription: usually under $350 a month; about 1 in 8 public cases at $4,000 or more, from a very small sample³⁵.

What it takes to compete

What the people at the top have

• Programming skill, mostly scraping. The most-used Apify tools collect data from large sites such as TikTok, LinkedIn and Reddit³⁰. Those sites change, so the tools break and must be repaired; a tool left broken for 30 days can be removed²⁵. This is engineering work, not writing prompts. It also carries the legal question set out under “Legal risk”.
• A portfolio. The 50 most-used developer accounts publish a median of 30 tools each. 14 of them publish more than 100, and 4 do well with one to three (our count, 10 October 2026)³⁰. We left out only Apify’s own account. Some of the other 50 may be run by Apify or its staff, which we did not check; if so, the independent few-tool successes are rarer still.
• Time, and upkeep. The developer with the testimonial had about two years and 93 tools behind his place near 45th²⁸³⁰. The blog author put in 15 to 20 hours a week for six months, 5 to 7 of them on repairs⁷¹.
• A specific buyer. The best single cases sell one result to one profession: search data to website marketers, tax answers to tax advisers³⁴⁴⁶. In the tax case the asset is the expert material and the customer list, not the assistant⁴⁶.
• A way to be found. Apify’s own documents say promotion separates tools with many paid users from those with almost none²⁷. A store listing does not bring buyers by itself (Step 5).
• For client work, the ability to sell. The founder quoted earlier found building easier than finding a second customer⁴⁵.

What it costs

Little money: tool subscriptions of tens of dollars or euros a month, plus AI model fees if the tool uses a model²³²⁵⁶⁶. A lot of time, with no pay while it is spent. Small balances can be lost: on Apify, a developer whose balance never reaches the $20 or $100 minimum in 12 months forfeits it²⁵²⁶. Courses and communities are an extra cost that the method does not require⁵²⁵⁵.

How to tell early which side you are on

• Can you already program? If not, the tool-marketplace routes are closed to you for now, and the no-code GPT route in the 2024 tweets is scheduled to end on 11 December 2026³. What remains is building for clients, where the limit is finding buyers (see AI automation agencies), packaging expertise you already own, as in the tax case⁴⁶, or a no-code platform with a published share and no published results, such as the one in the third tweet⁶⁹.
• Is someone already paying for the thing your tool does? The tools with users replace work that businesses already buy³⁰.
• Check your user count after a month. Apify shows it. The middle paid listing has 1 user in 30 days and about 92% have 10 or fewer³⁰. A tool with a few dozen real paying users in its first months is already in the top tenth.
• Does each run earn more than it costs? When a tool’s running costs exceed its 80% share, “Apify automatically sets that Actor’s profit to $0 for the month”, and model fees are on top²⁵²⁶.
• For client work: can you name ten businesses you could pitch this week? If not, the limit is sales, not building⁴⁵.

Early, or unwanted?

Custom GPTs as a paid product look unwanted: in nearly three years no payout programme opened beyond a small invited test, and the product is scheduled to retire³⁵⁶.

Tools that agents pay for look early. Large payment companies are building the means to pay³⁹⁶⁴. Measured spending is small but the measures differ widely: TRM’s estimate of payments it could tie to AI agents is roughly $5,000 to $11,000 a month, while all x402 payments on the public explorer come to about $1 million a month³⁷³⁸. Neither figure covers agents that use tools through an ordinary paying account on a marketplace such as Apify, which we could not separate from use by people³⁰.

Tool stores are growing on the one company’s figures we have: Apify’s stated payouts rose from $563,000 in September 2025 to $1.6 million a year later²⁸⁷². Signs that would settle the question: the cleaned figures for agent spending rising, Apify’s payouts continuing to rise faster than the number of developers, and any platform publishing a median payout.

What we did not verify

• The tweets’ attachments. We did not view the images, videos, guides or threads on the four tweets, so we do not know what the guides contain or lead to¹.
• What OpenAI’s test paid. We found no rate, total or per-builder amount for the 2024 GPT earnings test, and no notice that it formally ended. The help text about it was seen only as quoted on a forum⁶.
• OpenAI’s dates and plans. The help pages do not say when personal accounts lost the ability to create GPTs. The retirement dates are plans that OpenAI says may change³. OpenAI’s price page, read through a reader service, still listed custom GPTs under a personal plan, which conflicts with its help pages². Whether plugins will later allow paid digital products is unknown²⁰.
• Apify’s real payouts. No individual developer’s income was verified. The dollar split by rank and the users-per-$1,000 figure are our estimates from user counts. Our download covered about three quarters of listings. We did not separate use by AI agents from use by people, and we did not sample prices per run, so runs needed for $1,000 on Apify were not worked out³⁰. We did not check which of the most-used accounts are run by Apify or its staff. The address and date of Apify’s announcement about manipulation were not confirmed⁷³.
• AITOPIA’s results. The size of the “distributable net”, the prices, any payout figures and whether creators have been paid were not found⁴⁹⁶⁹.
• Agent payments. We found no figures on how payments divide between sellers. The totals from the x402 site, the explorer, TRM and Visa do not reconcile with each other^36373839. We read only the first of 24 pages of the explorer’s list³⁷. We confirmed only one point of the report that the site’s counter is frozen⁴¹.
• Marketplace prices and orders. Upwork and Fiverr refused our tools directly and through a reader service, so the Upwork prices come from search results¹⁸. No page we read showed how many orders GPT or agent listings get. One of our checks reported that the Contra listing showed 6 views and no likes; the text we received on a second reading did not include those counts, so they are not on this page¹⁷. The n8n and Make price pages were not read word for word⁶⁶.
• Pages read through a summarising tool. Several pages were read through a tool that returns a summary with quotes, not the raw text: some Apify help pages, the TrustMRR and MCPize pages, one OpenAI developer page and the post in⁴⁴.
• Other platforms. Poe’s programme exists on its own page, but we found no payout data; a 2024 news report about its price-per-message model was seen only in a search result⁶⁸⁷⁰. The creator terms of other agent marketplaces were not found.
• The single published GPT income figure ($6,248) and the report of a paid tool with no paying users were seen only in search results¹⁶⁴³.
• The Air AI order. We did not confirm that the court approved it⁵⁸. We did not check the outcomes of the 2024 FTC cases⁵⁹.
• Laws. Scraping law, laws outside the US and EU, and tax outside the US were not checked.
• Hours. The only figures for time are one developer’s own account and Apify’s two-hours-a-week guidance⁷¹. We found no measured source for time to first income or for how many sellers give up.

Sources

“Read at source” means we opened the page itself on 10 October 2026. “Secondary” means we read an article, summary or search result about the source, not the source itself. “Estimate” means our own sum. “Through a reader service” means a service (r.jina.ai) fetched the page for us because the site refused our tools directly.

1. Does It Pay collection of 4 tweets on this scheme, collected 2026-10-06; individual tweets linked above. Read from our collected copy on 2026-10-10; attachments not viewed.
2. OpenAI Help, “Building and publishing a GPT”. https://help.openai.com/en/articles/8798878-building-and-publishing-a-gpt . Updated about 2026-09-28. Read at source on 2026-10-10 through a reader service. Also https://chatgpt.com/pricing , read through a reader service.
3. OpenAI Help, custom GPT retirement and migration FAQ. https://help.openai.com/en/articles/20001519 . Undated; lists milestones from 2026-09-11. Read at source on 2026-10-10 through a reader service.
4. OpenAI Help, “GPTs FAQ” and “Creating a GPT”. https://help.openai.com/en/articles/8554407-gpts-faq and https://help.openai.com/en/articles/8554397-creating-a-gpt . Updated about 2026-09-28. Read at source on 2026-10-10 through a reader service.
5. OpenAI, “Introducing the GPT Store”. https://openai.com/index/introducing-the-gpt-store/ . 2024-01-10. Read at source on 2026-10-10 through a reader service.
6. OpenAI Developer Community, “What is the status with GPT Store revenue share”. https://community.openai.com/t/what-is-the-status-with-gpt-store-revenue-share/839172 . 2024-06-25, latest reply 2025-12-04. Read at source on 2026-10-10; secondary for the wording of OpenAI’s help text quoted in it.
7. The Decoder, report on OpenAI’s test programme to pay GPT developers. https://the-decoder.com/openais-launches-program-to-pay-developers-based-on-usage-of-their-gpt-models/ . 2024-03-27. Secondary.
8. “GPTs Window Shopping” (research preprint). https://arxiv.org/html/2405.10547v1 . 2024-05-17, data from January to March 2024. Read at source on 2026-10-10.
9. “GPT Store Mining and Analysis” (research preprint). https://arxiv.org/html/2405.10210v1 . 2024-05-16. Read at source on 2026-10-10.
10. “LLM App Squatting and Cloning” (research preprint). https://arxiv.org/html/2411.07518v1 . 2024-11-12. Read at source on 2026-10-10.
11. Research preprint on extracting prompts and files from custom GPTs. https://arxiv.org/abs/2311.11538 . 2023-11-20. Read at source on 2026-10-10 (summary page).
12. CISPA, “GPTracker: A Large-Scale Measurement of Misused GPTs”. https://cispa.de/en/research/publications/84716-gptracker-a-large-scale-measurement-of-misused-gpts . 2025-05-12. Read at source on 2026-10-10.
13. OpenAI Developer Community, “Has anyone made money from GPTs”. https://community.openai.com/t/has-anyone-made-money-from-gpts/961791 . Posts dated 2024-09-30 to 2024-11-21. Read at source on 2026-10-10.
14. OpenAI Developer Community, “How do GPT builders make money”. https://community.openai.com/t/how-do-gpt-builders-make-money/581293 . 2024-01-11. Read at source on 2026-10-10.
15. Manifold, prediction market on custom GPT earnings in 2024. https://manifold.markets/HenriThunberg/will-anyone-be-reported-to-have-mad . Created 2023-11-10, resolved “no” on 2025-08-04. Read at source on 2026-10-10.
16. Indie Hackers profile listing the post “How I Made $6,248 With Custom GPTs (No Code, No Audience)”. https://www.indiehackers.com/siadam . 2025-04-20 per search result. Secondary (search result only; the page did not open directly or through a reader service).
17. Contra, freelance listing for custom GPT development. https://contra.com/p/3MegWvEr-i-will-develop-custom-gpt-applications-for-your-needs . Published 2024-10-28. Read at source on 2026-10-10 through a reader service; the view and like counts one of our checks reported were not in the text we received.
18. Upwork, listing for custom GPT development. https://www.upwork.com/services/product/development-it-custom-gpt-development-for-business-efficiency-1869776999803402620 . Undated. Secondary (search result only; the site refused our tools directly and through a reader service).
19. OpenAI developer documents, plugin guidelines. https://developers.openai.com/plugins/plugin-guidelines . Undated. Read at source on 2026-10-10.
20. OpenAI developer documents, plugin monetization. https://developers.openai.com/plugins/build/monetization and https://developers.openai.com/apps-sdk/build/monetization . Undated. Read at source on 2026-10-10, the second partly through a summarising tool.
21. OpenAI developer documents, plugin submission. https://developers.openai.com/plugins/deploy/submission . Undated. Read at source on 2026-10-10.
22. OpenAI, usage policies. https://openai.com/policies/usage-policies/ . Change log dated 2025-10-29. Read at source on 2026-10-10 through a reader service.
23. OpenAI, services agreement. https://openai.com/policies/services-agreement/ . Effective 2026-01-01. Read at source on 2026-10-10.
24. Apify documents, pricing and costs for monetised Actors. https://docs.apify.com/platform/actors/publishing/monetize/pricing-and-costs . Undated. Read at source on 2026-10-10.
25. Apify, Store publishing terms and conditions. https://docs.apify.com/legal/store-publishing-terms-and-conditions.md . Last updated 2026-09-15. Read at source on 2026-10-10.
26. Apify documents, monthly payouts. https://docs.apify.com/actors/publishing/monetize/monthly-payouts.md . Undated. Read at source on 2026-10-10.
27. Apify documents, monetising Actors and agent payments. https://docs.apify.com/platform/actors/publishing/monetize and https://docs.apify.com/platform/integrations/x402 . Undated. Read at source on 2026-10-10, partly through a summarising tool.
28. Apify, page for Actor developers. https://apify.com/partners/actor-developers . Undated; testimonials dated 2025-10-29 to 2026-03. Read at source on 2026-10-10, and again through a reader service.
29. Apify Help, “How does Actor monetization work”. https://help.apify.com/en/articles/8684010-how-does-actor-monetization-work . Last updated 2025-10-22. Read at source on 2026-10-10.
30. Apify Store public listing data. https://api.apify.com/v2/store?limit=1000&offset=0 . Live data, 2026-10-10. Read at source on 2026-10-10; the counts and shares are our own computation (estimate). The plain listing stops returning rows after about 16,000, so the main download asked for each category in three sort orders and removed duplicates, which gave 63,617 of about 83,000 listings. An earlier download of 10,162 listings is used only for the agent-payment flag. Listing totals per developer come from asking for one account at a time (https://api.apify.com/v2/store?username=NAME&limit=1 , the “total” field).
31. Apify, “$1M Challenge” results page. https://apify.com/challenge . Undated. Read at source on 2026-10-10 through a reader service.
32. Apify, page for MCP developers. https://apify.com/mcp/developers . Undated. Read at source on 2026-10-10 through a reader service.
33. Microns, listing of a two-Actor Apify business for sale. https://www.microns.io/startup-listings/apify-actor-suite . 2026-03-30. Read at source on 2026-10-10 through a summarising tool.
34. TrustMRR, “Advanced GSC MCP”. https://trustmrr.com/startup/advanced-gsc-mcp . Live page, 2026-10-10. Read at source on 2026-10-10 through a summarising tool.
35. TrustMRR, list of listings and individual listing pages. https://trustmrr.com/startup-sitemap.xml . Live pages, 2026-10-10. Read at source on 2026-10-10; the name search and count are our own (estimate).
36. x402, project home page. https://www.x402.org/ . Read at source on 2026-10-10; per-transaction and per-seller figures are our estimate.
37. x402scan, public explorer of x402 payments. https://www.x402scan.com/ . Live data, 2026-10-10. Read at source on 2026-10-10 through a reader service; per-seller figure is our estimate.
38. TRM Labs, “Who’s actually paying? Measuring AI agent payments onchain”. https://www.trmlabs.com/trm-tech-blog/whos-actually-paying-measuring-ai-agent-payments-onchain . 2026-09-09. Read at source on 2026-10-10.
39. Visa, “Agentic payments from the ground up”, and the linked report with Artemis. https://www.visa.com/en-us/thought-leadership/innovation/agentic-payments-from-the-ground-up and https://www.visa.com/vcom-assets/content/dam/visa/reimagine-visa/thought-leadership/documents/agentic-payments-report.pdf . About 2026-07-14, data to 2026-04-21. Read at source on 2026-10-10.
40. The Defiant, report on Visa’s adjusted x402 figures. https://thedefiant.io/news/infrastructure/visa-s-sheffield-pegs-adjusted-x402-volume-at-19m . 2026-07-16. Secondary; read through a reader service.
41. Spoted Crypto, article on the TRM Labs x402 report. https://www.spotedcrypto.com/trm-labs-x402-ai-agent-payments-report-2026/ . 2026-09-15. Secondary.
42. PulseMCP, guest post by the founder of Ref, “Pricing the unknown: a paid MCP server”. https://www.pulsemcp.com/posts/pricing-the-unknown-a-paid-mcp-server . About 2025-09-17. Read at source on 2026-10-10 through a reader service.
43. Godberry Studios, closure notice at the address of a 2026 guide to monetising MCP servers. https://godberrystudios.com/posts/how-to-monetize-mcp-servers-2026/ . Read at source on 2026-10-10 through a reader service (closure notice); secondary for the original article, seen only in a search summary.
44. The AI Career Lab, post about an MCP server paid through x402. https://theaicareerlab.com/blog/x402-mcp-server-agents-pay . 2026-07-02. Read at source on 2026-10-10 through a summarising tool.
45. Hacker News, comment by a founder building agents for manufacturers. https://news.ycombinator.com/item?id=44062543 . 2025-05-22. Read at source on 2026-10-10.
46. CustomGPT.ai, case study including TaxWorld. https://customgpt.ai/turn-your-expertise-into-revenue-with-expert-ai/ . 2026-02-12, modified 2026-03-17. Read at source on 2026-10-10 through a summarising tool.
47. Upwork, “In-Demand Skills 2026” news release. https://investors.upwork.com/news-releases/news-release-details/upworks-demand-skills-2026-demand-top-ai-skills-more-doubles-ai . 2026-02-04. Read at source on 2026-10-10 through a reader service.
48. AITOPIA, home page. https://www.aitopia.ai/ . Undated. Read at source on 2026-10-10.
49. AITOPIA, “Earn” page and terms. https://aitopia.ai/earn and https://aitopia.ai/terms-conditions . Undated. Read at source on 2026-10-10, the Earn page through a reader service.
50. Zephyr, newsletter sign-up page. https://zephyr-hq.com/subscribe . Undated. Read at source on 2026-10-10.
51. Zephyr, community page. https://zephyr-hq.com/run-on-claude . Undated. Read at source on 2026-10-10.
52. Gumroad, “Claude Mastery” course page. https://zephyrhq.gumroad.com/l/ClaudeMastery . Undated. Read at source on 2026-10-10.
53. Atonomi, home page. https://atonomi.ai . Undated. Read at source on 2026-10-10.
54. MCPize, page on monetising MCP servers. https://mcpize.com/developers/monetize-mcp-servers . Undated. Read at source on 2026-10-10 through a summarising tool.
55. Skool, “AI Automation Society Plus” about page. https://www.skool.com/ai-automation-society-plus/about . Undated. Read at source on 2026-10-10; the revenue figure is our estimate.
56. Affiliate and pricing pages: https://www.skool.com/affiliate-program , https://www.gohighlevel.com/affiliate , https://www.make.com/en/affiliate , https://n8n.io/affiliates . Undated. Read at source on 2026-10-10; the Make and n8n pages through a reader service.
57. US Federal Trade Commission, complaint in FTC v. Air AI Technologies and news release. https://www.ftc.gov/system/files/ftc_gov/pdf/AiraiComplaint.pdf and https://www.ftc.gov/news-events/news/press-releases/2025/08/ftc-sues-stop-air-ai-using-deceptive-claims-about-business-growth-earnings-potential-refund . 2025-08-25. Read at source on 2026-10-10.
58. US Federal Trade Commission, news release on the proposed Air AI order, and case page. https://www.ftc.gov/news-events/news/press-releases/2026/03/air-ai-its-owners-will-be-banned-marketing-business-opportunities-settle-ftc-charges-company-misled . 2026-03-24. Read at source on 2026-10-10 through a reader service.
59. US Federal Trade Commission, “FTC Announces Crackdown on Deceptive AI Claims and Schemes”. https://www.ftc.gov/news-events/news/press-releases/2024/09/ftc-announces-crackdown-deceptive-ai-claims-schemes . 2024-09-25. Read at source on 2026-10-10 through a reader service.
60. US Federal Trade Commission, guidance on the Business Opportunity Rule. https://www.ftc.gov/business-guidance/resources/selling-work-home-or-other-business-opportunity-revised-rule-may-apply-you . 2022-11-16. Read at source on 2026-10-10.
61. Goodwin, note on EU AI Act transparency duties, and the text of Article 50. https://www.goodwinlaw.com/en/insights/publications/2026/08/alerts-technology-dpc-eu-ai-act-transparency-obligations-now-in-force and https://artificialintelligenceact.eu/article/50/ . 2026-08-03. Secondary for the note; the article text was read on 2026-10-10.
62. US Federal Communications Commission, ruling on AI-generated voices in robocalls. https://www.fcc.gov/document/fcc-makes-ai-generated-voices-robocalls-illegal . 2024-02-08. Read at source on 2026-10-10 through a reader service.
63. Anthropic, software directory policy. https://support.claude.com/en/articles/13145358-anthropic-software-directory-policy . Undated. Read at source on 2026-10-10.
64. Stripe documents, machine payments. https://docs.stripe.com/payments/machine . Undated. Read at source on 2026-10-10.
65. US Internal Revenue Service, “Understanding your Form 1099-K”. https://www.irs.gov/businesses/understanding-your-form-1099-k . Undated. Read at source on 2026-10-10.
66. n8n and Make price pages. https://n8n.io/pricing/ and https://www.make.com/en/pricing . Secondary: read on 2026-10-10 only as a summary; the prices did not appear in the page text we could fetch.
67. Public profile and tweet data for the account in the third tweet, read through a tweet-reading service because X refused our tools. https://api.fxtwitter.com/choyamymuna/status/2091890381541822723 . 2026-08-24. Read on 2026-10-10. The same record holds the AITOPIA announcement the tweet quotes ( https://x.com/AITOPIAai/status/2091886655749681581 ). The full text of the first tweet was read the same way at https://api.fxtwitter.com/everestchris6/status/2099864376840958185 .
68. TechCrunch, report on Poe’s price-per-message model for bot creators. https://techcrunch.com/2024/04/09/poe-introduces-a-price-per-message-revenue-model-for-ai-bot-creators . 2024-04-09. Secondary (search result only; article not opened).
69. AITOPIA, Agent Builder page. https://aitopia.ai/agent-builder-landing . Undated. Read at source on 2026-10-10 through a reader service.
70. Poe, creator documents, “Creator Monetization”. https://creator.poe.com/docs/resources/creator-monetization . Undated. Read at source on 2026-10-10 through a reader service, which returned a stored copy.
71. Apify blog, “How I built 98 Actors on Apify Store in 6 months” (written by a developer for Apify’s paid writing programme), and Apify documents, “Why publish Actors”. https://blog.apify.com/building-98-actors-on-apify-store/ and https://docs.apify.com/academy/build-and-publish/why.md . 2026-07-22; the second is undated. Read at source on 2026-10-10, the first through a reader service.
72. Apify’s post “The Apify $1M Challenge” on Indie Hackers, and Apify blog, “Programmer passive income: get customers for web scrapers”. https://www.indiehackers.com/post/the-apify-1m-challenge-mBIQoARFDO6JZKiZFU8M and https://blog.apify.com/programmer-passive-income-web-scraping/ . 2025-11-05 and 2023-09-05. Read at source on 2026-10-10 through a reader service.
73. Apify, announcement to its developer community headed “stricter enforcement against Apify Store manipulation”. https://discord.apify.com/m/1215263302406504499 . Shown as posted two days before we read it. Read on 2026-10-10 through a reader service, which returned the announcement under the title of an unrelated discussion; its own address and exact date were not confirmed.
74. LinkedIn, User Agreement. https://www.linkedin.com/legal/user-agreement . Effective 2025-11-03. Read at source on 2026-10-10.
75. Does It Pay, “AI automation agencies”. /ai-automation-agency/ . Researched 2026-10-07. Our own page; the figures quoted from it carry their sources there.

Corrections

If you are quoted or named on this page and think something is wrong, want your reply shown beside it, or want to be removed, write to [email protected]. We aim to reply within 14 days, and always within one month. Factual errors are corrected with a dated note here. A sentence that is seriously disputed comes down while we check it.

Change log: (empty)