AI jobs

AI Data Annotation Platforms

The eight major AI data annotation and training platforms compared on payout method and cadence, degree floors, entry gates and current recruiting status — read from each operator's own published terms, with a practical verdict on where each fits.

Updated August 2026

Quick answer: The strongest starting points among AI data annotation platforms in 2026 are DataAnnotation for generalists and Mercor for credentialed specialists. DataAnnotation publishes the clearest tier-by-tier rates in the category — general projects from $25–30+ per hour, coding, STEM and licensed-professional tracks from $50–100+ — and pays by PayPal within days of a withdrawal request. Mercor publishes its contract terms in full and pays weekly on Wednesdays through Stripe or Wise. Outlier has the clearest payment mechanics of the open generalist platforms (weekly Tuesday payouts, rate shown before you accept a project), Handshake AI runs a US-only fellowship structure suited to students and academics, Appen is the most geographically open, on a monthly payment cycle, and Surge AI advertises the highest rates in the category for elite credentials, by application only. Every platform here engages you as an independent contractor with no guaranteed volume — the single most important fact about the category.

The comparison table

Every column below was verified on 25 August 2026 against the operator’s own page or, where the cell refers to past events, a named report. Recruiting status reflects what the public sign-up path said on that date; individual country availability can differ from the marketing page and only the sign-up form itself will tell you.

One relationship explains three rows of that table. Scale AI, Inc. is the contracting company. Outlier and Remotasks are both Scale-operated worker-facing platforms — Outlier is the higher-skill, Western-facing front end, operated through Smart Ecosystem, Inc., and Remotasks is the older, globally distributed microtask platform. Outlier’s own FAQ states that it is “a platform operated by Scale AI” and its site footer carries the copyright line “Smart Ecosystems” (the plural is the footer’s own). Any comparison that presents Scale AI, Outlier and Remotasks as three competing companies is wrong, and the practical consequence is that signing up to all three does not diversify your risk in the slightest.

How we judged these platforms

Most comparison pages in this category rank on advertised hourly rates. That is close to useless, because advertised rates are marketing, most platforms publish no rate card at all, and the number that decides whether the work was worth doing is the one that arrives in your account.

We ranked on four things instead, in this order.

1. Published payout terms. Does the operator publish a payment method and a stated cadence you can hold it to? A platform with a modest published rate and a stated weekly transfer beats a platform advertising a high rate with no published terms.

2. Entry gates, including the unpaid ones. An unpaid one-hour assessment with a single attempt is a real cost. So is a degree requirement that is only disclosed in an FAQ. Both belong in the decision before you start, not after.

3. Current recruiting status. Several platforms in this category advertise globally while quietly closing individual countries. A page that tells you a platform is “available worldwide” without checking is worse than no page.

4. What the work actually is, and what it is not. Every platform here engages contributors as independent contractors. None guarantee volume. That is the single most important fact about the category and it is usually buried.

For the mechanics of getting hired once you have picked a platform, see data annotation jobs. For the wider map of AI training work including the specialist marketplaces and research-participation platforms, see get paid to train AI. This page is about which platforms to trust with your unpaid hours.

The California settlement worth knowing about

One public court record is directly relevant to contributors in California. Scale AI has agreed to a $12.5 million class action settlement in McKinney, et al. v. Scale AI, Inc., et al. (S.F. Superior Court No. CGC-24-620481), which alleged that Contributors were misclassified as independent contractors. The defendants deny violating any law and deny that Contributors are not independent contractors (settlement website). The class covers people who worked as Contributors for Scale AI or Smart Ecosystem, or were placed by HireArt, while residing in California between 10 December 2020 and 28 February 2026 — including work done through Outlier and Remotasks. Payments are automatic with no claim form; the opt-out and objection deadline is 3 September 2026, and the final approval hearing is 30 October 2026. The court has granted preliminary approval only. If you worked through these platforms from California, check the settlement website for your position.

Payout terms, compared

This is the section to read before the rate card, because cadence and method decide when advertised money becomes real money. Everything below is from each operator’s own published documentation.

DataAnnotation: pays on request

DataAnnotation pays via PayPal, with deposits delivered within a few days of a withdrawal request — the only on-request payout in the category rather than a fixed cycle. Its FAQ states the platform has paid over $20 million to contractors since 2020 across a network it describes as more than 100,000 contractors.

Outlier: weekly Tuesdays, rate shown up front

Outlier processes payments weekly on Tuesdays, for work completed the previous Tuesday through Monday at midnight UTC, via PayPal, Airtm or ACH bank transfer. Its FAQ states that rates “vary by expertise, project complexity, and location” and that “you’ll always see the tasking rate before starting any project” — a genuinely useful commitment, since it means the number you accept is the number you are paid at. Task availability varies more than payment mechanics do; treat volume, not the payout, as the variable.

Mercor: weekly Wednesdays, with a verification pass

Mercor’s hourly contracts are paid weekly: hours logged from Saturday 00:00 IST to Friday 23:59 IST are locked, verified and paid the following Wednesday, through Stripe where available and Wise where it is not, with the first Stripe payout held for seven days (talent documentation). Work must be tracked through the Insightful desktop application, which captures screenshots, and Mercor reviews time logs weekly — its documentation lists the situations where hours are adjusted (inactivity, idle timers, out-of-scope work, prohibited tool use), with email notification and a dispute route. Know the tracking model before you start: hours not captured through Insightful are not payable, and deliverable-based contracts pay on approval of the full deliverable. Mercor states it reached $2 billion in annualised gross revenue by June 2026 — a company figure, unaudited — and TechCrunch reported in July 2026 it was in talks at a $20 billion valuation, the strongest commercial position in the category.

Appen: a monthly cycle, so plan the gap

Appen’s contributor platform is CrowdGen, and its payments page states that payments are credited by the 14th of each month, covering the previous month’s earnings, through Dots — offering Payoneer, PayPal, Airtm and local bank transfer. The practical consequence is that the wait depends on when in the month you work: work done on the 1st waits about six weeks, work done on the 30th about two. Two mechanics from the payments FAQ are worth knowing up front: payout methods cannot be changed during the processing window (the 6th to the 15th), and hours cannot be logged retroactively for work done before your project activation date. Appen’s business is growing — Q2 FY26 revenue of $65.1 million, up 26 per cent year on year, with FY26 guidance of $270 million to $300 million reaffirmed — and its crowd spans more than 200 countries, the widest reach here.

Handshake AI and Remotasks: confirm terms before you start

Handshake AI does not publish payout method or cadence on its programme page, so confirm both during onboarding — and note the eligibility rule, which is narrower than most people assume: fellows must be US-based with work authorisation Handshake can support (F-1 students on CPT or OPT may qualify; STEM OPT is not supported). Remotasks publishes no current rate card or payout terms on a public page, and its availability varies by country; the sign-up page is the only reliable check for your location.

Degree floors: the requirement nobody advertises

The phrase “no experience needed” is used across this category to mean “no prior employment in AI”. It does not mean “no requirements”, and two of the largest platforms have hard educational floors that appear only in their FAQs.

DataAnnotation’s tier structure is the most transparent published requirement set in the category, and it is worth reading in full before you choose an assessment track:

• General projects: bachelor’s degree or equivalent experience, strong writing and critical thinking. Starting at $25–30+ per hour.
• Multilingual projects: native fluency in more than one language beyond English. Starting at $20+ per hour.
• Coding projects: programming experience in Python, JavaScript or similar. Starting at $50–100+ per hour.
• STEM projects: master’s or PhD in maths, physics, biology or chemistry, or a bachelor’s plus 10 or more years of professional experience. Starting at $50–100+ per hour.
• Professional projects: licensed credentials in law, finance or medicine. Starting at $50–100+ per hour.

The equivalent-experience clause is genuine and it is the most useful sentence on DataAnnotation’s site for anyone without a degree. Its FAQ states that “someone with a bachelor’s degree and 10 years in mathematics could be as qualified as someone with a master’s”, and that real-world experience counts as much as formal education.

Nowhere is the degree floor a barrier to attempting the assessment — it is a barrier to passing it and to being matched to work. That distinction matters because the assessment is unpaid.

The unpaid hour: what each platform asks for before it pays anything

Onboarding costs are the most under-reported number in this category. Here is what each platform asks you to give up front.

DataAnnotation’s single-attempt rule is the sharpest of these, and its own FAQ repeats it in more than one answer: “You can only take the Starter Assessment once, so review carefully before submitting.” Its guidance is that successful applicants spend longer crafting thorough answers rather than rushing. It also says plainly that, given application volume, it cannot provide direct responses to every applicant — so silence is a likely outcome and not a signal.

Identity verification is now standard rather than exceptional. DataAnnotation verifies through Persona, requiring photos of a government-issued ID plus a front-facing selfie with head movement; its FAQ confirms that your information, including biometric data, is stored by both DataAnnotation and Persona. Outlier requires a valid ID and mobile phone from your country of residence, a current résumé and a LinkedIn profile showing your education and work history. Handshake AI verifies identity as step one of its process. If handing over biometric data to a third-party processor is a problem for you, it is a problem at the front door of most of this category.

Current recruiting status, as of 25 August 2026

Open and actively recruiting: DataAnnotation (six countries — United States, United Kingdom, Canada, Australia, Ireland and New Zealand), Outlier, Mercor, Handshake AI (United States only), Appen through CrowdGen.

Regionally restricted: Remotasks. The platform is live, but availability varies by market and has changed over time. Remotasks advertises availability in over ninety countries but does not publish the list, so the sign-up page is the only reliable check for your location.

By application only: Surge AI. There is no open marketplace or self-serve sign-up. The workforce page lists named specialist roles — Constitutional Lawyer, Award-Winning Author/Journalist, Venture Capital Partner, Startup CEO, Management Consultant, Investment Banker, University Professor — and the only route in is to email talent@surgehq.ai with your background. The advertised rate on the journalist role is $200 to $400 per hour, and the stated requirement is a Pulitzer, a National Magazine Award or equivalent, or long-form contribution to The New Yorker, The Atlantic or NYT Magazine. These are the highest published rates in the category, for a small pool of elite credentials — useful calibration for reading any “up to $400 per hour” category headline.

Not a platform you can join: Scale AI. It contracts through Outlier and Remotasks. There is no direct contributor sign-up.

Handshake AI’s eligibility rule deserves stating in full because it is narrower than most people assume: to participate in the Handshake AI Fellowship you must be based in the United States and hold valid work authorisation that Handshake can support. F-1 students eligible for CPT or OPT may qualify; STEM OPT is not supported.

Verdicts by situation

Best for a generalist starting this month: DataAnnotation. It publishes the clearest tier-by-tier rates in the category, screens on an assessment rather than employment history, pays by PayPal within days of a request, and has no documented payment disputes. The bachelor’s-or-equivalent floor and the single-attempt assessment are the two things to plan for.

Best for anyone holding a professional credential: Mercor. Weekly Wednesday payouts, published contract terms, and by far the strongest commercial position of any operator here. Accept that you will be screen-recorded throughout and that hours can be deducted on review.

Best-documented payment mechanics on a generalist platform: Outlier. Weekly Tuesday payments with three payout methods, a stated pay period, and a stated commitment that you see the rate before accepting a project. Its weakness is inconsistent task volume, not payment.

The only platform built around a US student or academic schedule: Handshake AI. Its fixed-term fellowship structure fits study around work better than spot-market tasking does, and its live listings run from a generalist track up to specialist roles at $120–125 per hour. Payout terms are not published on the programme page, so confirm cadence and method during onboarding — and, as everywhere in this category, keep your own record of hours worked.

Best for elite specialists: Surge AI. If you are a Supreme Court clerk or appellate advocate, a Pulitzer winner, an IMO gold medallist or a professor at a top-ranked institution, the rates are the highest published anywhere and the application is one email. If you are not, this is not a route.

Best if you are outside the six-country DataAnnotation list: Appen through CrowdGen. Appen states its crowd spans more than 200 countries and territories, which makes it the most geographically open operator here. Budget for up to about six weeks between working and being paid on the monthly cycle.

Check availability before planning around it: Remotasks. It publishes no current rate card or payout terms on a public page, and country availability changes — the sign-up page is the reliable test.

Not applicable: Scale AI. Join Outlier instead.

What advertised rates are actually worth

Almost every figure in this article is labelled by what kind of number it is, because in this category the label matters more than the number.

The practical rule: judge a platform by the rate shown on the specific project you are offered, work out your true hourly rate including unpaid onboarding, and remember that project rates and availability can change. Mercor’s documentation, to its credit, commits that completed work is never repriced.

Before you sign up: five checks

Confirm the payout method works where you live. PayPal, Stripe, Wise, Payoneer and Airtm are not all available everywhere, and a platform that accepts your application may not be able to pay you. Check this before the assessment, not after.

Read what happens to a pending balance if your account is closed. Contractor agreements across this category give the operator wide discretion; know before you start how a pending balance is handled and what the dispute route is.

Work out your true hourly rate. Add unpaid onboarding, unpaid assessments, unpaid training and unpaid qualification tasks to your denominator. A $30 per hour project reached through three unpaid hours pays less than it looks like for the first week.

Keep your own records. Log your hours independently of the platform’s tracker. In any payment question, your own contemporaneous record is the thing that protects you.

Do not rely on one platform for essential income. Project volume is not guaranteed anywhere in this category, projects end without notice, accounts can be closed at the operator’s discretion and entire countries have been dropped. This is the reason the whole category is supplementary income rather than a job.

Where this market is going

Three well-documented shifts explain why generalist work is getting harder to find and specialist work is getting better paid.

The generalist tier is being cut. Scale AI cut 200 employees and 500 global contractors in July 2025, one month after Meta’s $14.3 billion investment for a 49 per cent stake, with interim chief executive Jason Droege saying the company “ramped up our GenAI capacity too quickly”. xAI reportedly cut 500 roles from its data-annotation team in September 2025 — roughly a third of a 1,500-person team — and said it would prioritise specialist AI tutors over generalists (TechCrunch, 13 September 2025).

Entry requirements are hardening at the platforms that used to be open. Outlier now requires a minimum of an associate degree. DataAnnotation’s baseline is a bachelor’s degree or equivalent experience. Both of those are recent enough that most of the “no experience needed” content written about these platforms is now wrong.

The money has moved to credentials. Surge AI advertises $200 to $400 per hour for award-winning journalists. Handshake AI’s top live listings are $120 to $125 per hour for philosophy PhDs and game designers. Mercor positions explicitly on medicine, law and PhD-level reasoning. Movement at the generalist end has been flatter, and project rates can be revised between projects.

The practical read: the market is rewarding credentials more each quarter. If you hold one, the specialist platforms are courting you; if you do not, treat annotation as supplementary income and pair it with building skills the market is moving towards. For routes that lead somewhere, see AI career paths; for what the wider market pays, see AI salaries.

Frequently asked questions

What are the best data annotation platforms in 2026?

DataAnnotation is the best entry point for generalists, because it publishes clear tier-by-tier rates starting at $25 to $30 per hour for general projects, screens on an unpaid assessment rather than employment history, and pays via PayPal within a few days of a withdrawal request. Mercor is the best option for anyone holding a professional credential, paying weekly on Wednesdays through Stripe or Wise with published contract terms. Outlier has the clearest payment mechanics of the generalist platforms, paying weekly on Tuesdays via PayPal, Airtm or ACH, but task availability is inconsistent. All three engage you as an independent contractor with no guaranteed volume.

Are AI data annotation platforms legit?

The major platforms are real businesses that pay real contributors, but “legit” and “reliable” are different questions. DataAnnotation states it has paid over $20 million to contractors since 2020, and Outlier and Mercor both publish specific weekly payment cycles. The structural caution is the same everywhere: every platform engages you as an independent contractor, volume is not guaranteed, and payout terms vary widely — so read the terms, and never depend on one platform for essential income.

Which AI training platform pays the most?

Surge AI publishes the highest rates in the category, advertising $200 to $400 per hour for award-winning authors and journalists, by application against elite credentials. Among open platforms, Handshake AI’s top live listings run to $125 per hour for specialist roles, and DataAnnotation advertises $50 to $100 or more per hour for coding, STEM and licensed professional tracks. For generalist work with no specialist credential, DataAnnotation’s $25 to $30 or more per hour is the highest published starting rate.

Do you need a degree to work on AI annotation platforms?

Yes at the two largest generalist platforms, and this has changed recently. Outlier’s FAQ states plainly that you need at least an associate degree, with some projects requiring a bachelor’s, master’s or PhD. DataAnnotation’s baseline is a bachelor’s degree or equivalent real-world experience, and its FAQ explicitly says someone with a bachelor’s and ten years in a field can be as qualified as someone with a master’s. Appen through CrowdGen and Remotasks publish no educational floor. Handshake AI’s lowest advertised tier lists an associate’s or bachelor’s.

Which platform pays fastest?

DataAnnotation is the fastest in practice, because it pays on request rather than on a cycle: deposits are delivered within a few days of a withdrawal request. Of the fixed-cycle platforms, Outlier is quickest, paying every Tuesday for work completed the previous Tuesday through Monday at midnight UTC, so the longest wait is about seven days. Mercor pays every Wednesday for hours logged Saturday through Friday, a wait of up to eleven days, and holds your first Stripe payout for a further seven. Appen through CrowdGen is by far the slowest, crediting earnings by the 14th of each month for the previous month’s work, so work done early in a month waits about six weeks for payment and work done late in a month waits about two.

What is the Scale AI settlement and am I eligible?

Scale AI agreed to a $12.5 million settlement in McKinney v. Scale AI, Inc. (S.F. Super. Ct. No. CGC-24-620481), a California class action alleging that Contributors were misclassified as independent contractors rather than employees. The class covers anyone who worked as a Contributor for Scale AI or Smart Ecosystem, or was placed by HireArt, while residing in California between 10 December 2020 and 28 February 2026 — which includes work done through Outlier and Remotasks. Payments are automatic and no claim form is required. The opt-out and objection deadline is 3 September 2026 and the final approval hearing is set for 30 October 2026. The defendants deny the allegations, and the court has granted preliminary approval only. Check the official settlement website for your own position.

Are Outlier, Remotasks and Scale AI the same company?

Effectively, yes. Scale AI, Inc. is the parent contracting company. Outlier is its higher-skill, Western-facing contributor platform, operated through Smart Ecosystem, Inc., and Outlier’s own FAQ describes it as “a platform operated by Scale AI”. Remotasks is the older, globally distributed microtask platform, also run by Scale. Signing up to all three does not spread your risk, because a decision made at Scale AI affects all of them at once — which is exactly what the McKinney settlement class definition reflects by covering Contributors across both platforms.

How much can you realistically earn on AI annotation platforms?

It depends far more on availability and credentials than on advertised rates. Credentialed specialists genuinely earn $100 or more per hour on the platforms that court them; generalist work starts around $20 to $30 per hour where rates are published, and volume is not guaranteed on any platform. The realistic framing across the category is supplementary income rather than a wage: treat any single platform as one stream, judge it by the rate on the specific project in front of you, and include unpaid onboarding time in your arithmetic.

Do AI annotation platforms monitor your screen?

Mercor does, mandatorily. All work on hourly Mercor contracts must be tracked through the Insightful desktop application, which captures screenshots, and Mercor’s own documentation states that hours are removed at weekly review when screenshots show no activity, unrelated tasks, an idle timer, or work outside the assigned project scope. Most platforms also require third-party identity verification, including biometric selfie checks stored by both the platform and the verification provider.

Is annotation work taxable?

Yes. Every platform on this page engages you as an independent contractor, which means no tax is withheld and you are responsible for declaring the income and paying tax on it yourself, in your own country, including any self-employment or National Insurance equivalent. Mercor, Outlier and Handshake AI all require tax forms as part of onboarding. The detailed country-by-country treatment, including United States, United Kingdom and Australian rules, is covered in our get paid to train AI guide.

Related pages

• AI jobs — the hub, and the two very different markets behind the phrase
• Get paid to train AI — the full landscape including specialist marketplaces, country eligibility and tax
• Data annotation jobs — what the work is and how to get hired
• AI training jobs — job types, expert marketplaces and salaried lab roles
• Best AI jobs for beginners — the routes ranked by realistic accessibility
• AI salaries — what AI work pays across the whole market
• AI career paths — the routes that lead somewhere other than annotation