Platforms are actively hiring specialists to evaluate AI outputs in coding, finance, law, STEM, and writing. The pay is real. The gotchas are too.

If you can write clearly, code competently, do math accurately, or apply professional judgment in a field like accounting, law, or medicine, there's a growing category of remote contract work that fits the way most of us prefer to operate: alone, on our own schedule, through a screen.

AI model evaluation. Not generic data labeling. The specialized kind, where your actual expertise is the product.

You apply to a platform that supplies human feedback to AI companies. DataAnnotation, Outlier, Mercor, and Alignerr all publicly advertise this kind of work. You pick a track based on your strongest skill. You pass a screening assessment. Then, if accepted, you see available projects in a dashboard and start completing tasks.

The tasks vary: writing prompts, comparing AI-generated responses, checking facts, reviewing code, debugging AI-generated files, rating outputs, translation, or applying subject-matter judgment to model answers. It's contractor work, task-based, with no long-term commitment required. DataAnnotation describes all of it as flexible, remote, and available 24/7/365.

No pitching. No calls. No audience-building. You log in, pick up work, and submit it.

DataAnnotation's public FAQ lists pay by track: multilingual projects start at $20+/hour, general evaluation at $25–$30+/hour, STEM at $40+/hour, professional tracks (law, finance, medicine) at $50+/hour, and coding at $50–$100+/hour.

Those are platform-reported figures. They're useful as benchmarks, but they don't tell you what you'll actually earn per week.

Outlier's listings show different and sometimes lower ranges. Their US-remote coding opportunities list $12.50–$50/hour depending on the role. Indeed's estimate for Outlier AI trainers, last updated March 2026, puts the average around $31.78/hour, though that number comes with inconsistent sample data and should be treated as directional only.

Deel reported in March 2026 that AI trainer roles grew 283% in cross-border hiring in 2025 and now span 70,000+ workers across 600+ organizations. That's a platform-reported labor-market signal, not a guarantee you'll get hired this week.

One anecdotal data point: a DataAnnotation contractor profiled by Business Insider said he started at $20/hour, later made about $25/hour, and saw rates up to $30/hour. He also said the work can be repetitive and that pauses and research time reduced his effective paid hours.

This is side income. Treat it that way.

Getting accepted and staying busy are separate problems.

DataAnnotation says the Starter Assessment can be taken only once. Specialized assessments can take one to two hours. Review typically takes a few days. And even after approval, project access depends on your expertise and performance. Workers on Reddit describe passing quickly but then facing additional unpaid qualification tests. Others describe dry spells with no available projects.

Outlier's onboarding runs 30 to 90 minutes and includes identity verification. After that, project matching is not guaranteed. Indeed reviews from February 2026 describe the work as potentially decent part-time income, but inconsistent.

A recent Guardian investigation interviewed Outlier workers who described unstable work, uncomfortable content, and monitoring concerns. Scale (which operates Outlier) responded that the work is flexible and project-based, that pay is transparent, and that contributors choose when and how they participate. Worth knowing before you apply.

If anyone asks you to pay for training, buy a verified account, or purchase a starter kit, walk away. DataAnnotation says it charges no signup fees and never asks workers to pay for anything. The FTC treats upfront-payment work-from-home offers as a major scam signal. Business Insider documented an active black market for AI training accounts. Don't be a customer.

Pick your strongest honest expertise lane: writing, coding, math, finance, law, medicine, or bilingual work. Write a short skills inventory listing your domains, tools, credentials, and examples of accurate work. Then apply to one specialized assessment on one platform. Not three platforms at once. One.

Have your PayPal account ready (DataAnnotation pays through PayPal) and a valid government ID for identity verification. Read the assessment instructions carefully before answering. There are no retakes.

After you finish, apply to a second platform as a backup. Track every hour and every payout in a spreadsheet from day one. US gig income is taxable even if it's part-time, temporary, or never reported on a 1099. If you expect to owe $1,000 or more in tax for 2026, estimated tax payments may apply.

The realistic timeline from application to first dollar: a few days to several weeks, and not guaranteed. But if you have genuine expertise and the patience for screening friction, this is one of the cleaner ways to turn knowledge into quiet, asynchronous side income.