Using AI to make money: what works, what does not, and where to start

The honest version first: most people who make money with AI make it by selling something to other people, not by asking a model for ideas. The model compresses work. It does not find customers, and it does not take responsibility for the output.

That distinction separates the approaches that survive from the ones that quietly stop earning after a month.

The four models that actually pay

1. Service work, delivered faster. Writing, design production, research, transcription cleanup, reporting, small automations. Clients pay for the outcome, not the tool. Your advantage is turnaround and consistency, and the pricing stays close to normal freelance rates, which is fine, because your cost per hour collapses.

2. Productised micro-services. The same service, packaged with a fixed scope and price: a monthly content package, a set of landing pages, a research brief. Easier to sell because clients know what they are buying and easier to deliver because repetition makes the workflow cheap.

3. Small tools for a narrow audience. A generator, calculator, or internal tool for one industry. The AI part is usually trivial. The value is in understanding the workflow well enough that the tool solves a real, specific problem.

4. Content and education. Explaining a niche workflow to that niche, whether as a newsletter, a course or a paid template. This is the slowest to start and the most durable once it works.

Anyone selling you a prompt pack as a business model is selling you the product, not the outcome.

What to set up before you take money

• A deliverable you can describe in one sentence. Clients buy clarity. "I write four product emails a month for SaaS companies" sells; "I do AI content" does not.
• A quality step that is yours. Every one of these models needs a human review pass. Skipping it is how people lose the second contract.
• Terms that cover the tooling. Say that you use AI assistance, and that you are responsible for accuracy. Hiding it creates the only risk that matters.
• A simple invoicing setup. A contract, a deposit, and a payment method that works across borders. This is more of a differentiator than it should be.

Where the real money is thin

Fully automated content farms still earn occasionally, but platform detection, search algorithms and buyer scepticism have eroded the margin. Bulk work competes on price with people in cheaper markets and with the platforms themselves. Anything where the customer cannot tell whether a human reviewed the output is a race to the bottom.

Conversely, the strongest niches are the unglamorous ones: regulated industries, internal processes, and languages or jurisdictions where few people can produce a competent output. A working automation for a freight forwarder is worth more than another newsletter about prompts.

A realistic first month

Pick one service you can already do manually. Define a fixed deliverable. Build a repeatable workflow with a review step. Sell it to three clients at a normal rate. Then raise the scope, not the effort.

That path is unexciting and it works. If you would rather build a product, start by solving the workflow you already understand, and read the engineering roles in this directory to see what companies are paying for the same skills full time.