It is real, but smaller than the headlines. Two things are true at once.
First, the AI training platforms — DataAnnotation, Surge AI, Mercor — genuinely pay people to review and correct model outputs. DataAnnotation has over 1,500 employee reviews on Indeed, which is not something a scam has.
Second, general freelance marketplaces like Upwork and Fiverr have a growing category of AI-adjacent work: building small automations, writing with AI, cleaning up AI content, setting up chatbots.
What is not real: the screenshots of huge weekly payouts, the courses promising a fixed monthly income, the idea that any of this is passive. Every honest source describes work availability as inconsistent and the first month as mostly unpaid effort. If you go in expecting a salary, you will quit before the first payment clears.
Hourly ranges from cited sources; time-to-first-payment from independent 2026 reviews.
Path A: Train AI models. You apply to DataAnnotation, Surge, or Mercor, pass an unpaid assessment, and get matched to tasks — rating answers, correcting code, judging which of two responses is better. No client-hunting. The barrier is the assessment.
Path B: Sell AI-assisted services. You list a narrow offer on Upwork or Fiverr — for example, "I turn your messy meeting notes into a clean summary using AI" — and compete for buyers. No assessment. The barrier is being invisible until you have reviews.
Which fits you?
No sign-up, nothing to install, and it takes about a minute. This is the same thing the course has you do on day one.
You'll end up with: A proposal you can send to charge more.
Runs on bloom0's own AI. Nothing you type is stored or shared.
Read these as hourly ranges, not monthly income. Availability of hours is the part nobody can promise you.
Three hours at $20/hr is about $60. Whether you get three hours in a given week is not something you control. Most people earn little or nothing at first.
Every failure I have read about clusters into four shapes.
1. Applying broadly. Sending 40 Upwork proposals to "anything AI-related" gets zero replies. One narrow offer ("I clean up AI-generated blog drafts for coaches") gets replies.
2. Treating the assessment casually. DataAnnotation and Surge assessments are the most common failure point. They test written reasoning and instruction-following. Do them when you are rested, not at midnight.
3. Confusing hourly rate with income. A platform paying a high rate that only offers a few hours a week can pay less in total than a lower-rate platform offering many more hours (source). Ask about availability, not just rate.
4. Waiting to feel ready. The people who start earning are not the most skilled — they are the ones who shipped a mediocre first gig, got a review, and improved from there. The first order on Fiverr is priced low deliberately.
You do not need to code. You need to use AI tools competently enough that a client saves time hiring you instead of doing it themselves.
Realistic beginner offers:
Each of these takes one afternoon to learn with free tools (ChatGPT, Claude, a basic no-code automation tool). The value is not the tool — the client could use it too. The value is that you have already learned the annoying bits: what prompts fail, how to check for hallucinations, how to format output the client can actually use.
This is what bloom0 teaches from zero, in English and Spanish, without assuming any technical background.
Answer three questions honestly.
Do you have a professional domain already? If yes (nurse, lawyer, accountant, senior developer), Mercor is your best-paid option. Apply this week.
Can you commit to two focused hours in a quiet room? If yes, do the DataAnnotation or Surge assessment. It is the fastest path from nothing to variable income, with the honest caveat that acceptance is not guaranteed.
Would you rather build a small business than pass a test? If yes, spend a weekend making one narrow Fiverr gig. Price it low. Get your first review. Raise the price.
If you answered no to all three, be honest with yourself: freelancing may not be the right shape of work for you right now, and that is fine. A part-time job with predictable hours will pay more than an abandoned freelance profile.
Total time: about 435 minutes.
Three questions. No score is stored, nothing is sent anywhere.
1. You see a Reddit post claiming Surge pays $30/hr for general RLHF tasks. What is the safest read?
Independent 2026 reviews put verified general-task pay at $14–$20/hr. Higher numbers circulate widely online but usually reflect specialist work, not the default. Neither belief nor dismissal is right — calibration is.
2. You have no professional background yet. Which platform is a realistic month-1 target?
Mercor's top rates go to verified specialists — it is a month-3 target if you build expertise, not a week-1 one. Bidding broadly on Upwork is the classic beginner mistake. DataAnnotation and a narrow Fiverr gig are the two entry points that actually work from zero.
3. Platform A pays $40/hr but sends you two hours of work a week. Platform B pays $18/hr with steady availability. Which pays more?
Two hours at $40 is $80. Ten hours at $18 is $180. Independent reviews explicitly warn that hourly rate alone is misleading — the platform with more available hours often pays more in total.
Free forever · no account needed · nothing you make is shared