Contract work on AI lab projects — writing evaluation tasks in your professional field, grading model output, and expert interviews for clients like OpenAI and Anthropic.
The best-paying AI training platform here. Real listings run roughly $35–120/hr, with senior domain experts higher; Mercor advertises a $112/hr average. Generalist projects sit far lower — a Meta video review project paid $21/hr before being cut to $16/hr.
A résumé plus a recorded AI voice interview, reused across roles. Credentials genuinely gate the high rates: the strong money assumes real professional experience in law, medicine, finance or engineering.
Hard weekly hour caps, and projects can end abruptly — one Meta project of 5,000+ contractors was cancelled overnight. When a project ends there is no automatic transfer; you reapply from scratch.
Worth knowing: A March 2026 breach exposed data for 40,000+ contractors, including AI interview video, facial biometrics, Social Security numbers, government IDs and banking details. Multiple class actions are pending. Worth weighing before uploading ID.
Ease of work2
Pay5
Consistency2
#2
micro1
the company behind the Zara AI recruiter
Residents of 19 countries excluded, including China, Russia and Iran
Expert AI training for frontier labs — building reinforcement-learning environments, evaluating agent output and domain reasoning across engineering, medicine, law, finance and linguistics, plus a large robotics video-labelling push. micro1 sells an AI recruiting agent first and human data second, and you meet the recruiter before you meet any work.
Among the highest reported here, with the usual caveat. Glassdoor puts Expert AI Trainer at roughly $49–90/hr across 560 submissions, averaging about $66 — a larger sample than anything else on this page. Everyday domain listings are more often $20–50/hr. In September 2026 micro1 announced 10,000 robotics trainers in seven days at an advertised $50–90/hr for reviewing and labelling robot videos with no AI experience required; that is several times the going rate for labelling work, and we would treat it as an advertised figure until workers confirm it.
Apply to a specific listing, then sit an interview with Zara, micro1’s own AI recruiter, and verify ID before certification. No formal prerequisites are published and the rates plainly assume real domain credentials. micro1 states outright that certification "does not guarantee immediate placement".
The weak point, and the same structural one Mercor has: certification puts you in a pool, not on a project. Being passed and then never matched is the most repeated complaint, and the certified pool is far larger than the number of live projects at any moment. Projects also pause or fill without notice.
Worth knowing: Read its 4.7/5 Trustpilot carefully — those 10,000+ reviews are overwhelmingly job candidates rating the Zara interview as a pleasant experience, not trainers reporting on paid work, so the score says nothing about whether work or payment arrives. The recurring substantive complaint is being asked for a passport or driver’s licence up front, with some reviewers alleging the interviews mainly feed micro1’s own recruiting AI. Weigh that before uploading ID.
Ease of work3
Pay4
Consistency2
#3
User Interviews
paid research studies, not AI training
Get paid to be a research participant — product feedback sessions, interviews, focus groups and diary studies run by companies studying their own users. The odd one out on this page: you are the subject, not the trainer.
The best rate-per-hour here by a distance. User Interviews puts the average study above $45, with 1-on-1 interviews $20–300+, focus groups $40–200+ and diary studies $55–400+. Sessions are usually 30–60 minutes, so an accepted study is genuinely well paid for the time.
Effectively none. No degree, no credentials, no exam — you sign up, fill in a profile and apply to studies. Being an interesting fit for a study matters more than any qualification.
The catch, and a severe one. You are competing for a limited number of studies and most applications end in a screener rejection. Reviewers describe a month passing with a single study offered, and screeners themselves can run 15 minutes unpaid.
Worth knowing: Trustpilot sits at 3.5/5 across 1,243 reviews — the healthiest spread on this page (68% five-star), but 15% are one-star and the recurring theme is payment. One reviewer reports a $25 study from April 2026 still unpaid in September after six emails. Treat it as unpredictable pocket money, never as income you can plan around.
Ease of work5
Pay5
Consistency1
Writing and evaluating AI chatbot responses, plus coding and subject-matter tasks. The closest thing here to straightforward, accessible work.
Around $20/hr for general writing and evaluation, rising to $50–100/hr for coding, STEM and professional specialisms. Payment reliability is its strongest suit — weekly via PayPal, and even critical reviewers agree it pays as promised.
A starter assessment, no degree required. The common frustration is silence: many applicants report never hearing back at all, sometimes for months.
The single biggest complaint. Work arrives in bursts, and contributors describe weeks or occasionally months with nothing available. Those who score well on projects report a steadier pool.
Ease of work3
Pay4
Consistency2
#5
Handshake AI
the AI Fellowship
US only — no visa sponsorship
The college recruiting company's pivot into AI training: reviewing model output, writing step-by-step solutions to hard problems and flagging where the model is wrong. Runs two tiers — frontier-lab expert work and plainer evaluation tasks.
Advertised as up to $100/hr, with role-level rates listed from $40/hr to $125/hr — but the actual job posting quotes a far wider $15–150/hr band, and some tasks carry a maximum handling time that caps what you are paid however long the work takes. Worker reports split hard: some say the pay is good, one reviewer alleges about seven dollars an hour before tax.
Open to associate through postdoc level and it states no AI experience is required, so the academic bar is lower than the "PhD work" framing suggests. You verify identity and phone, pass a background check, then wait for a project invitation on no fixed timeline. Some projects ask for a portfolio, and reviewers mention a lot of quizzes.
Handshake says outright that projects may start, pause or end on partner labs' needs, sometimes with short notice, and there are no minimum hours. The wait for a first invitation is open-ended, so approval is not the same as work.
Worth knowing: You must be based in the US with work authorisation Handshake can support; it is a 1099 contract with no sponsorship. F-1 students on CPT or OPT may qualify, but STEM OPT is not supported. Judge the ratings carefully too: Glassdoor's 3.2/5 is corporate staff rather than contractors, and the harshest Trustpilot page mixes in complaints about the student job board. The reviews that are actually from AI contractors are positive but number only about twenty.
Ease of work3
Pay3
Consistency2
Writing difficult prompts, building grading rubrics and ranking model answers, with separate coding, maths and domain-expert tracks. Historically the most generalist-friendly of the big platforms, though Scale is steering it toward expert work.
Outlier publishes no rate at all — its FAQ says only that rates vary by expertise, project and location, shown before you start. A detailed first-person account reports $15–40/hr, and a wage lawsuit alleged an effective ~$15/hr once unpaid time spent reading instructions was counted. The $50–150/hr figures circulating online trace to affiliate blogs and we could not verify any of them.
An associate degree at minimum, with bachelor's, master's or PhD expected on many projects, plus a résumé, LinkedIn, government ID and a mobile number from your country of residence. Contributors describe qualification assessments running two to five hours, and Outlier does not state whether that time is paid.
Feast or famine. Reviewers describe work arriving in waves with dry spells between. Scale cut 14% of its workforce after Meta's investment and closed its Dallas generalist contractor team in October 2025, so the generalist end is actively shrinking.
Worth knowing: Two things to weigh. Some projects involve violent or abusive content: six Outlier workers filed a class action in January 2025 alleging PTSD from exposure without adequate safeguards. Separately, sudden deactivation with an unpaid balance is the single most repeated grievance, with a 1,000-signature petition behind it. Scale also settled a $12.5M misclassification case covering Californian contributors, denying liability. On the other side, weekly payment is the most consistent praise in its reviews.
Ease of work3
Pay3
Consistency2
#7
Toloka
worker sign-up is now Mindrift
Rating and comparing AI responses, search relevance and red-teaming. Toloka is now the business-facing brand — as of mid-2026 its worker platform redirects to Mindrift, where accounts and balances were migrated.
Mindrift advertises $15–30/hr entry and $60–100+/hr for domain experts, but that is a marketing ceiling we could not corroborate from workers. Historic Toloka reports are far bleaker — cents per task, one Trustpilot reviewer citing under $4/hr. Rates are fixed per project and indexed to your country’s cost of living, so identical work pays much less outside high-income countries.
The easiest door into AI training work here: general tasks need no credentials, just sign-up and English. Specialist projects require a relevant degree, an unpaid assessment and ID verification.
Mindrift states plainly that task availability is not guaranteed and tracks demand in the AI industry. The most common complaint is passing unpaid onboarding and then waiting weeks for work that never appears.
Worth knowing: Restricted in a long list of countries, including China, Russia, Kenya and Taiwan. Check eligibility before investing time in onboarding.
Ease of work4
Pay2
Consistency2
#8
TELUS Digital
AI Community
Search and ads quality rating against Google’s evaluator guidelines, plus per-project AI Community gigs — data collection, transcription and annotation.
Modest and slow to move. A US Personalized Internet Assessor posting states $12/hr; raters commonly report $12–18/hr, less in lower-cost countries. Glassdoor rates compensation 2.7/5, and workers describe years at the same rate.
Application, language assessment, then a multi-part open-book qualification exam that is entirely unpaid — reviewers describe hours of unpaid study. Ongoing quality audits can end the work.
Weekly hours are capped at roughly 10–25, and supply often does not fill even that. Reviews describe stretches with little or no work, and applicants waiting months in review.
Worth knowing: Sentiment splits by audience: Glassdoor sits at 3.4/5 among people already working, while Trustpilot is 1.2/5, dominated by rejected applicants and deactivated accounts.
Ease of work2
Pay2
Consistency2
#9
Meridial
by Invisible Technologies
Expert-tier AI training — writing reasoning and judgment calls, evaluating model output and red-teaming across law, medicine, engineering and other specialist domains.
Not disclosed. Meridial states only that earnings are "set per project" and confirmed before you start, and we found no verified worker-reported rates anywhere. Treat unknown pay as a risk, not a blank cheque.
Aimed at credentialed specialists, with proctored assessments requiring camera, microphone and screen-sharing access.
The dominant complaint by a wide margin is passing the assessments and then receiving no work at all.
Worth knowing: The weakest reputation on this page: Trustpilot 1.7/5 across 25 reviews, 88% of them one-star. Several reviewers allege the unpaid assessments are themselves the product. Invisible Technologies behind it is an established company, but we would approach this one cautiously.
Ease of work2
Pay2
Consistency1
Scores are our editorial read of public information from official job postings, Trustpilot, Glassdoor and Indeed, last reviewed September 2026. Pay and availability on these platforms change often — verify current terms before signing up. RemoteFast is not affiliated with, and earns no commission from, any platform listed here.