AI Side Hustles Misleading — the headline number is real but unrepresentativ

By the Income Reality Check Editorial Team · Updated 2026-05-15 · 7 min read

AI training side gigs promise $55–$400 an hour — what they actually pay

Verdict: Misleading — the headline number is real but unrepresentative. The eye-popping hourly rates on Mercor and Scale AI exist, but they describe a starting offer, not a steady income.

More Perfect Union’s investigation I Tracked Down the Hidden Workers Secretly Powering ChatGPT has crossed 1.24 million views by following a recently graduated Ivy League PhD named “Jen” through the AI data-work pipeline. She answers a $55-an-hour LinkedIn ad from a contractor called Mercor, gets the gig, then watches the rate slide to $45, then $35, then disappear. Reporter Karen Hao stitches Jen’s story to a research finding that gets quoted everywhere: 86% of U.S. data workers struggle to pay their bills and median earnings sit below $23,000 a year. The video isn’t a side-hustle pitch — it’s the reality check on one. For a generation of 22-to-45-year-old job hunters who keep seeing “Get paid $40+/hr to train AI, no experience” on TikTok and LinkedIn, that gap between the ad copy and Jen’s pay history is the story.

What the video actually claims

The pitch the video is exposing is the one being aimed at viewers every day on social platforms: data-annotation and AI-training work pays $40, $55, $101, even “up to $400” an hour for people with college-level expertise, and you can do it from your laptop on your own schedule. The platforms named on screen — Mercor, Scale AI, Surge AI, Outlier — really do post those rates. Jen’s first Mercor contract is listed at $55/hr for a “philosophy intelligence analyst” role. A later one offers her $101/hr. A voiceover from one platform’s ad in the video brags about pay reaching $400 an hour.

The counter-claim from the reporting is also specific. Tim Newman, the labor researcher featured in the piece, cites his own study: 86% of U.S. data workers struggled to meet financial responsibilities, roughly a quarter rely on Medicaid or food stamps, more than one in five have experienced homelessness, and median earnings sit under $23,000 a year. Jen’s $55/hr role ends two weeks in. The next offer drops to $45, then $35, then $101 for a single 40-hour sprint that vanishes the day she completes it. The video doesn’t argue the ad rates are fake. It argues that the rate, on its own, doesn’t describe the job.

What the method actually requires

The structure of the offer is the missing piece. These are 1099 independent-contractor gigs in the United States — no benefits, no overtime, no unemployment insurance, no employer-paid payroll taxes. The U.S. Bureau of Labor Statistics counted 11.9 million Americans whose main job was independent contracting in 2023, about 7.4% of total employment, and AI data work is the fastest-growing slice of that segment. The hourly rate the contractor sees has to absorb roughly 15.3% self-employment tax (Social Security and Medicare combined, the portion an employer would normally cover), plus federal and state income tax, plus any health insurance the worker buys on the marketplace. A $55/hr advertised rate is closer to $35–$40/hr in equivalent W-2 take-home for U.S. workers earning at that level. That’s before factoring in the bigger problem: the rate only applies while a contract is open.

Mercor was valued at $10 billion in October 2025 after a $350 million funding round, with CNBC reporting the platform pays its 30,000-plus contractors a combined $1.5 million per day. Divide that out: that’s an average of about $50 per contractor per day, or $18,000 a year if work were steady. It isn’t steady. Projects end mid-contract, like Jen’s did. Scale AI, the bigger rival, cut 14% of its staff after a Meta investment in July 2025 — and contractor workloads tend to track that kind of corporate re-org. Scale and Mercor have also both faced wage-related class-action lawsuits in 2025 alleging that workers were misclassified as contractors when they were functioning as employees, and that effective hourly pay sometimes fell below California’s $16.50 minimum wage once unpaid time was factored in. The U.S. Department of Labor opened a Fair Labor Standards Act inquiry into Scale AI that’s been active since at least August 2024.

The “race against the clock” Jen describes in the video — taking 40 hours of $101/hr work in a single week because the project ends the next day — is the operational reality of the rate. The hourly number is real for the hours the work exists. The week-to-week dollar number is a different question entirely.

Who actually wins this game

The people clearing meaningful money on these platforms tend to share two traits: an unusually narrow domain expertise that AI labs are willing to pay scarcity-pricing for, and the personal financial cushion to wait between contracts. CNBC’s reporting on Mercor calls out medical experts billing $130–$180/hr and senior software engineers and coding experts at $60–$140/hr. Those rates fund the $10 billion valuation. The generalist “philosophy analyst” or “humanities grader” tier — where most viewers of a recruiting TikTok will actually land — sits much lower and churns much faster.

The U.S. Federal Trade Commission has been treating this gap as enforceable consumer harm, not just bad marketing. In July 2024 the FTC announced a $7 million settlement with the gig-work platform Arise Virtual Solutions for advertising “up to $18/hour” jobs when 99.9% of workers earned less between 2019 and 2022. A few months later the agency hit Lyft with a $2.1 million civil penalty for earnings claims it said overstated typical driver pay by as much as 30%, by basing the figures on the top 20% of earners. Neither case names an AI data-work platform yet, but the playbook the FTC is enforcing against — quoting top-decile pay as the typical experience — is exactly what’s happening in the Mercor and Scale ad copy. (U.S. readers should note that this enforcement is FTC-specific; the U.K.’s ASA and Australia’s ACCC police similar earnings-claim issues under their own ad codes.)

What you’d realistically earn

If you’re a U.S.-based generalist with a bachelor’s or master’s degree, expect listed rates in the $20–$45/hr range, contracts that last anywhere from a single afternoon to a few months, and unpredictable gaps in between. Newman’s study and the Communications Workers of America’s 2025 survey of U.S. AI data workers both put median earnings under $23,000 a year for people doing this as primary or significant income. That’s lower than the U.S. federal poverty line for a household of three.

If you bring a credentialed specialty — clinical medicine, frontier-math reasoning, securities law, a less-common second language — the math gets better. Real $80–$180/hr work exists in those lanes. Even then, you’re paid only when a project is running and rated highly, you carry the self-employment tax burden, and a single rejected submission can throttle the algorithm that hands out future tasks. The closest honest framing of the side hustle is: tutoring work with no schedule, no employer obligations, and a pay rate that includes a premium for accepting all of that uncertainty. For more on how AI-adjacent income claims tend to behave, see our breakdowns of 5 ways to make money in 2026 with AI and the best online work for students using AI side hustles.

Who this is (and isn’t) for

This work fits someone who already has a stable primary income, a verifiable credential AI labs are willing to pay scarcity rates for, and the patience to treat the gig as variable supplemental income rather than a paycheck. It also requires a willingness to be exposed to content moderation tasks that may include graphic violence or abuse — the video’s interview with Ozzy, a Surge AI grader who reviewed AI-generated gore for a project codenamed “Arsenic,” is not an outlier. It does not fit someone who needs steady weekly income, doesn’t have a safety net to bridge contract gaps, or is being recruited on the basis of an advertised top-of-band rate they were told was typical. The 86% financial-stress figure in Newman’s study is largely drawn from people in the second category trying to do the job of the first.

What to remember

The hourly numbers in those LinkedIn and TikTok ads are not invented. They are also not the typical experience, the steady-state experience, or the take-home experience. More Perfect Union’s video is one of the few mainstream investigations to do the math on that gap, and the math is uncomfortable: $55 an hour for two weeks, then nothing, is not a $55-an-hour job.

View counts and on-platform pay listings change continuously; figures above reflect the state of public reporting at time of review (May 2026).