AI Side Hustles Half-true — works only if you do the unspoken work

By the Income Reality Check Editorial Team · Updated 2026-05-22 · 7 min read

Cloning Gumroad ebooks with Claude AI: the math behind the $64 profit

Verdict: Half-true — works only if you do the unspoken work. The workflow runs cleanly, but the niche, the ad rules, and the steady-state economics are not what the demo shows.

A YouTube video from a channel called Mr. AI CASH, sitting at roughly 57,000 views, walks viewers through a four-step plan: scrape the Facebook Ads Library for a proven digital product, rebuild it with Claude AI plus a Higgsfield AI connector, publish it on Gumroad at $49.99, and push traffic with $15-a-day Facebook ads. The headline proof is a single test run: $60 in ad spend produced $124 in revenue, or about $64 of “profit.” Is the strategy real? The mechanics work. The story around them quietly falls apart in three different places.

What the video actually claims

The creator opens with a Gumroad dashboard he attributes to “the last 3 months,” then pivots to a clean rebuild. Step one is research: open the Facebook Ads Library, filter Germany for ads that have run more than five days, and log ten digital products into a Google Sheet. Claude AI then picks the “winner” — in the demo, a “40-page hormone mapping protocol” aimed at women over 40.

Step two is production. The video shows how to connect Higgsfield AI’s MCP connector to Claude, then run a chain of prompts: extract the original product’s positioning, generate a table of contents, draft chapters with two images each, make three cover variations, then merge everything into a final PDF. About 25 minutes of generation time, per the demo.

Step three is the storefront. Set up a fresh Gumroad account, paste in the AI-written title and description, list the ebook at $49.99, attach a 25%-off coupon so the “effective” price reads as $37. Step four is traffic: Claude writes the ad copy and audience brief, Higgsfield generates short-form video ads, and Facebook runs a $15/day test split across two ad sets (interest-targeted and broad). The headline result: three sales in three days, $124 in revenue, $60 in ad cost, an interest-targeted ad set that worked and a broad set that didn’t.

What the method actually requires

Start with the fees, because the demo doesn’t. Gumroad charges 10% plus $0.50 per transaction on direct sales through your own links, climbing to 30% if buyers find you through Gumroad’s Discover marketplace (Gumroad). On a $49.99 sale with the 25% coupon applied, you net about $37.49 in revenue; after the 10% + $0.50 cut, you’re at roughly $33.25 before card processing. Three sales like that gross $112 to the seller — close to the creator’s “$124” figure once you account for any buyer who skipped the coupon, but materially lower than the sticker price.

Next, the tool stack. Claude Pro runs $20/month for light use, and Claude Max sits at $100 or $200/month depending on tier (Anthropic). Higgsfield AI’s published plans start at $15/month for 200 credits and run to $99/month at the Ultra tier (Higgsfield AI). The “monster” content engine the video pitches is real, but it isn’t free, and rate limits are a recurring interruption — the creator himself admits Claude will “stop you because of usage limits” and recommends waiting four hours between bursts.

Then there’s the niche problem, which is the biggest gap in the pitch. The “hormone mapping protocol for women over 40” is exactly the category Meta has spent years tightening. Meta’s Health and Wellness advertising standards specifically flag claims about “boosting metabolism,” “cleansing toxins,” and “balancing hormones” as the kind of physiological-effect language that demands medical substantiation, and the same policy restricts before/after weight-loss imagery and any creative that implies negative self-perception (Meta Transparency Center). A landing page on Gumroad that promises a hormone “protocol” can get the whole ad rejected even when the creative itself looks clean. Meta now evaluates ad copy, images, landing pages, and comments holistically.

The FTC sits one layer behind that. In 2015 the agency sued Lunada Biomedical over Amberen, a supplement marketed to women over 40 as causing “significant weight loss and loss of belly fat” through hormonal effects — the suit ended in a $4.5 million settlement (Federal Trade Commission, 2015). The FTC has continued the same line of enforcement: in July 2025 it moved against the telehealth firm NextMed for deceptive weight-loss claims and fake reviews (Federal Trade Commission, 2025). U.S. readers, in particular, should know that the agency reads “deceptive health claim” loosely — and an AI-drafted ebook with no clinical backing fits the pattern. (Outside the U.S., similar regimes apply: the ASA in the U.K., ASIC’s ACCC counterpart on consumer claims in Australia, and most EU member states under the Unfair Commercial Practices Directive.)

One more thing the pitch glosses: copying. The whole research step is “find a product that’s already running ads, then rebuild it.” Most of what gets rebuilt — a structure, a list of chapter ideas, a price point — isn’t itself copyrightable. But cover art, distinctive phrasing, screenshots, and direct text are, and Meta’s DMCA process for ads and landing pages is fast (Meta / Facebook Help Center). A complaint from the original seller can knock a copycat’s ads offline overnight.

Does the demo profit survive scaling?

Here’s the unit economics most viewers won’t run. The video reports three sales at an average implied price of around $41 (because at least one buyer didn’t use the coupon), against $60 in ad cost. That’s a cost per acquisition of about $20. Across digital products on Meta in 2026, broader benchmarks put average CPAs at roughly $20–$39 depending on category, with the overall cross-objective average closer to $7.50 — but health and weight-loss niches sit on the higher end, both because the ads themselves are restricted and because the audiences are heavily fought over.

Now add the fees back. Sticker $49.99, average effective price $41, Gumroad’s cut about $4.50, processing about $1.50, ad cost about $20, leaving roughly $15 of contribution margin per sale before any tool subscriptions or refunds. Three sales over three days isn’t a business; it’s a noisy sample. Two refunds or one bad creative iteration and the same campaign is underwater.

Who actually wins this game

Gumroad publishes enough data to know who clears real money on the platform. An independent analysis of roughly 146,000 Gumroad products found that the median product earns about 28 sales at a $13 median price — around $364 in lifetime revenue. About 34% of products with a price tag make zero sales, another 34% sell fewer than ten copies, and the top 1% of products earn more than 2,000 times what the bottom half earns combined. The winners aren’t randos using new accounts. They’re creators with prior audiences (newsletter lists, YouTube subscribers, Twitter followings), agencies that have built up paid-traffic playbooks across dozens of products, or specialists with topical authority — design templates, Notion systems, programming guides — that buyers actively search for.

The Mr. AI CASH demo is closer to the agency/paid-traffic profile than the audience-led one. If you have $5,000 to test offers across ten niches with Meta ads, the workflow described in the video is a credible way to spin up creative cheaply. If you have $200 and a single hormone ebook, the math doesn’t lean your way.

What you’d realistically earn

Set expectations against the platform’s own distribution. A beginner without an existing audience, posting one or two AI-built ebooks per quarter and running a small ad budget, should plan for $0–$100/month in the first six months, with a real chance of zero. A more deliberate operator who tests four or five offers, irons out a non-restricted niche (productivity, design assets, niche hobbies — not health), and rebuilds creative repeatedly can plausibly clear $500–$2,000/month after six to twelve months of work — call it 10–20 hours a week.

What you almost certainly will not do is replicate the Gumroad dashboard the creator flashes at the start of the video on your first try, in a restricted niche, with $60 in ad spend. NerdWallet’s own roundup of realistic side hustles is blunt about why: online platforms saturate, paid promotion creeps up as a cost of being visible, and “popular online platforms are becoming saturated” eats margin from anyone without a moat (NerdWallet).

Who this is (and isn’t) for

If you’re a paid-traffic operator who already knows how to read a Meta dashboard, this approach can be a faster way to ship creative and test offers — Claude and Higgsfield do compress production. If you’re new to all of it, the niche selection matters far more than the AI prompts: a Notion template, a printable bundle, a code-review checklist, a Photoshop preset pack will face none of the ad-policy friction that a hormone ebook will. Avoid this method outright if you’re considering weight-loss, supplement-adjacent, financial, or get-rich offers — those are the categories where Meta rejection rates climb and where the FTC, FCA, or ASIC actually shows up.

How much capital do you really need to test this cleanly? A defensible figure is around $500–$1,000: $35–$120 in monthly tools, $300–$700 for a real two- to three-week ad test, and a small buffer for refunds and creative iteration.

What to remember

The video isn’t selling a scam. It’s selling a fair production workflow attached to a niche choice that quietly hands you the hardest version of the problem. Strip out the hormone angle, ship a non-restricted product, plan for the fees and tool costs, and the same four steps become a reasonable, if slow, side hustle. Keep the hormone angle and you’re running into Meta’s ad rules and the FTC’s enforcement priorities at the same time.

For more on what tends to work and what doesn’t when Claude is at the center of a digital-product side hustle, see our reviews of the laziest way to make money with Claude and how selling ebooks with Claude actually plays out.

View counts and the creator’s claims may have changed since this review was published.