How Much Do AI Training Jobs Actually Pay? (July 2026 Data)
2026-07-14 · Expert Match AI team
Ask this question on a forum and you'll get answers ranging from "$15 an hour, it's a scam" to "my friend makes $200 an hour reviewing legal documents." Both can be true, which is exactly the problem. So instead of anecdotes, here's what the market actually lists. We track live roles across ten platforms and recompute pay statistics every six hours from every listing that discloses a rate; as of this writing that's 218 priced roles out of 320 tracked. Current numbers are always on the live salary report, and this post explains how to read them.
The headline numbers, by field
Averages use the midpoint of each listing's posted range:
- Legal: about $137/hr on average
- Healthcare: about $119/hr
- Software and engineering: about $76/hr, with senior roles listed to $300/hr
- Science and research: about $74/hr
- Languages and linguistics: about $65/hr
- Finance: about $64/hr
- Generalist ML/data work: about $65/hr on average, but the entry tier runs $15-30/hr
The pattern is blunt: credentials get paid. A field where the work requires a license or an advanced degree averages two to four times the generalist rate.
The platform gap is bigger than the field gap
Here's the part most people miss. The same person, with the same skills, can be paid wildly differently depending on where they apply. Among platforms we track, current listing averages run from about $32/hr at the low end to over $400/hr at the top:
- Surge AI averages ~$406/hr across its listed expert tiers (physicians quoted $250-450/hr) but accepts very few applicants
- Mercor averages ~$116/hr across priced listings
- Braintrust averages ~$105/hr and charges talent zero fees
- micro1 spans the widest range, $20/hr generalist roles up to $300/hr senior engineering, averaging ~$57/hr
- Outlier averages ~$32/hr across its current tracks
Practical takeaway: before accepting a rate, check what your field is listed at elsewhere. Fifteen minutes of comparison on the jobs board is often worth a 2-3x difference.
What "listed rate" actually means
Three honest adjustments between the listed number and what hits your bank account:
Unpaid time. On several platforms, reading instructions, waiting for tasks, taking assessments, and redoing flagged work is unpaid. Worker reviews of task-based platforms consistently report that effective hourly rates land below the posted number once that overhead is counted. Project-based platforms (where you're staffed onto an engagement) suffer this much less.
Availability gaps. A $100/hr role you can work 3 hours a week earns less than a $40/hr role with steady volume. Every platform, including the best-reviewed, has dry spells. This is the single most reported frustration across worker communities in 2026, and it's why experienced people stack multiple platforms.
Tiers and negotiation. Many platforms route you into pay tiers based on your listed credentials. Stating your degree, license, or certification precisely (not modestly) is often worth more per hour than any negotiation you'll ever do. On Braintrust you set your own rate inside the posted band; elsewhere, tiering is mostly automatic.
What realistic earnings look like
Based on listing data plus what workers publicly report: a generalist doing rating and annotation part-time typically sees a few hundred dollars a month, with wide variance. A credentialed professional treating this as a serious side income, on 2-3 platforms, commonly works in the $50-150/hr band when work is available. The elite tier (board-certified physicians, senior attorneys, published researchers at the most selective platforms) is real but small: those $250-1,000/hr listings exist, and they reject most applicants.
Nobody should quit a job for this market's current shape. As project income that respects your expertise, it's among the better remote options available.
Check your field
The numbers in this post age; the salary report doesn't, since it recalculates every six hours. Look up your field there, then browse the live roles behind the averages, or upload a resume on the homepage and see every role scored against your background - free, no account, resume never stored.
Published by the Expert Match AI team. Pay figures are listed rates from live postings, not a survey of accepted offers. Some outbound application links on this site carry disclosed referral codes; rankings and recommendations are never influenced by them.