Pay for AI Training Work: Rates by Platform and Field, Dated

AI training work pays whatever each platform states on its own pages, and on 2026-11-09 those stated rates ran from USD 6 per hour for first-person video collection to single listings of USD 200 to 400 per hour. Platforms state entry rates of about USD 15 to 50 per hour for general tasks and expert rates of about USD 60 to 150 per hour. Pay arrives weekly, bi-weekly or on request, mostly through PayPal or bank transfer.

Every figure here is a platform's claim with its check date, never a confirmed earning. The page sits within AI training data licensing as the contributor side of the market: the people paid to create, rank, rate and test AI training data. Stated rates differ from take-home pay because of unpaid assessment time, task-based pay, contractor status and tax, and three court records, Boggs, Cavalier and McKinney, concern pay.

Every rate comes from the named platform's own page, re-read on the check date; see how we source, date and check every figure.

How much does AI training work pay?

Platforms state entry rates of about USD 15 to 50 per hour for general tasks and expert rates of about USD 60 to 150 per hour on 2026-11-09, with a few single listings above that. DataAnnotation states general projects from USD 25 to 50 per hour and expert rates of up to USD 75 to 125+ (checked 2026-11-09). Mindrift states entry-level work at USD 15 to 30 per hour and domain experts at USD 60 to 100+ per hour. Alignerr states an average of USD 80 per hour (checked 2026-11-09), and one Surge AI workforce listing states USD 200 to 400 per hour.

"Up to" and "average" are the platform's chosen framing. A ceiling describes the best-paid listing, not the typical task, and an average has no stated population. No platform in the record publishes how many contributors earn a stated rate, so no figure here is an earning, a median or a salary.

Finding work, passing assessments and building a reputation are covered in the comparison of AI training jobs and platforms compared, which cites the rates below instead of restating them.

Stated pay rates by platform, checked 2026-11-09

The table lists 11 contributor platforms alphabetically with the rate each states, the rate unit, the payment schedule and method it states, and the page the figure comes from. Four rules govern how to read it:

• Claim: the platform's own figure on the check date, not an earning.
• No rate on record: no figure was found in the record, which is not proof that none exists.
• Up to: a ceiling chosen by the platform.
• Mercor: the home-page figures are not shown because they could not be read reliably; only the referral formula is.

The table compares stated rates, units, schedules and referral terms across platforms, as of 2026-11-09; every cell is the named company's claim unless another date is given.

Outlier publishes a payment structure but no rate, and Toloka's expert program states 6,000+ active expert contributors without a contributor rate in the record.

Stated pay rates by field: generalist, coding, medicine, law, languages and video

Stated rates rise with the field: general evaluation and language roles sit lowest, and coding, medicine, law and finance highest. DataAnnotation's job board lists 133 roles on 2026-11-09, led by 43 language roles, 19 medical roles and 14 finance roles, and states a maximum for each job card. Handshake's listings span from up to USD 17 per hour for an AI Evaluation Specialist to up to USD 150 per hour for a Software Engineer. Mindrift states monthly maxima by field and USD 6 to 15 per hour for egocentric video collection (blog post of 2026-09-17).

The table compares stated rates by field with the roles DataAnnotation lists, as of 2026-11-09; every rate is a platform's claim.

Monthly caps are maxima, not salaries, and the DataAnnotation counts sum to 133 only with its single Design role included.

How AI training pay is set: hourly rates, per-task prices, monthly caps and bounties

AI training pay is set in four units: an hourly rate, a price per completed task, a monthly cap and a bounty per item.

• Hourly rate: a rate per hour worked, as DataAnnotation and Alignerr state.
• Per-task price: a price per completed task; Mindrift states that contributors earn per completed task and see the amount before starting.
• Monthly cap: a maximum per month for a field, as Mindrift states for STEM, coding and consulting.
• Bounty: a fixed price per accepted item; on 2026-11-09 Shipd shows bounties of USD 80 to 220 per item.

Outlier states that it pays either hourly or per task, and that rates vary by expertise, project complexity and location (checked 2026-11-09). The location variable matters because Outlier's Flexible Working Guidelines of 2026-07-15 list 117 locations, and Kenya, Nigeria and Pakistan are not among them. Handshake's Fellowship is open to US-based fellows only.

Prolific works the other way around. The researcher who runs a study sets the pay, subject to a stated minimum of GBP 6 or USD 8 per hour (help page of 2026-07-17). Mindrift also states that there is no employment contract and no fixed hours, so a per-task price does not translate into an hourly rate without a measured time per task, which no platform in the record publishes.

When and how AI training pay arrives: schedules, payment methods and referral bonuses

Platforms state weekly, bi-weekly or on-request payment, mostly through PayPal or bank transfer. Outlier states weekly payment on Tuesdays through PayPal, Airtm or ACH. Mindrift states that payments are processed bi-weekly. DataAnnotation pays through PayPal, delivers deposits within a few days after a contributor requests them, and sets no minimum hours (checked 2026-11-09). Shipd publishes no payment terms, and Alignerr claims USD 1M paid weekly, a platform total with no stated scope. Shipd bounties are the only figures in the table priced per item.

Referral bonuses follow stated formulas on two platforms. Handshake's referral terms of 2026-09-24 set the bonus at 10 % of the referred fellow's hourly rate per paid hour, capped at 40 times the bonus rate, equal to 4 times the hourly rate, and ending at the cap or 6 months after account creation. Mercor states 20 % of the referral's earnings, capped at 4 times the hourly rate on hourly contracts (checked 2026-11-10).

Outlier accepts referrals from the contributor dashboard for candidates the contributor can personally vouch for, with no bonus amount published. None of these formulas affects the order of this page, which is alphabetical.

What lowers AI training pay in practice: unpaid assessments, task-based pay and claims in court

Three things separate a stated rate from take-home pay: time spent on assessments and onboarding, work paid per task rather than per hour, and the claims contributors have taken to court.

DataAnnotation's Starter Assessment takes about an hour and is taken once, its specialized assessments take 1 to 2 hours, and Outlier's onboarding takes 30 to 90 minutes with skill screenings (checked 2026-11-09). DataAnnotation's and Outlier's pages do not state whether this time is paid, and Alignerr's Terms of Use state that it has no obligation to pay for the interview. Alignerr tracks time with Hubstaff and periodic screenshots, and its FAQ says both that a contributor whose deactivation appeal is rejected is not eligible for payment for completed work and that all Alignerrs are paid for completed work except in extreme fraud.

Two complaints concern pay as an allegation. Boggs v. Stryder Corp. (N.D. Cal. 3:26-cv-07498-TSH), a misclassification class action filed on 2026-07-20, alleges pay of about USD 4 per chat interaction on a first project; the case is stayed pending mediation by the Dkt. 22 order of 2026-10-22, with a status report due 2027-02-05. Cavalier v. Surge Labs, Inc. dba DataAnnotation, filed 2025-05-20 in the Superior Court of California, County of San Francisco, alleges misclassification and unpaid training and qualification time and is pending as of 2026-11-09.

McKinney v. Scale AI (Superior Court of California, County of San Francisco, CGC-24-620481) reached final approval on 2026-10-30 at a gross settlement of USD 12,500,000 for a class period of 2020-12-10 to 2026-02-28, with payments allocated 100 % to non-wages on Form 1099; the distribution status conference is set for 2027-09-13. Outlier's own pages give two paid-out totals, USD 500M+ on the home page and USD 600M+ on the about page, and the contributor lawsuits against human data platforms page holds the full record.

What AI training pay is worth after US and UK tax rules

Outlier, Alignerr and Mindrift engage AI trainers as contractors rather than employees, so a stated rate is a gross figure. Outlier's terms of use, section 3, state that it withholds no tax except as required by law, and the McKinney settlement reports contributor payments on Form 1099.

• Form 1099-NEC: USD 2,000 for payments made after 2025-12-31 (26 U.S.C. 6041(a) as amended by Pub. L. 119-21 s.70433), adjusted for inflation from 2027.
• Form 1099-K: payments over USD 20,000 and more than 200 transactions (IRS, checked 2026-11-09).
• Self-employment tax: 15.3 % (12.4 % plus 2.9 %, 26 U.S.C. 1401) on net earnings of USD 400 or more (IRS, checked 2026-11-09).
• UK trading allowance: GBP 1,000 per tax year (ITTOIA 2005 Part 6A Chapter 1); Self Assessment registration is due by 5 October after the tax year when gross trading income exceeds it.

Filing steps, expenses and rules abroad are covered in the guide to taxes for AI training contributors.

Benefits, risks, downsides and alternatives of AI training work for pay

Paid AI training work offers flexible, remote, task-based income at stated expert rates, but the rate is a claim, the hours are not guaranteed and the contract terms favor the platform.

Benefits

• Mindrift shows the amount per task before the contributor starts.
• DataAnnotation sets no minimum hours (checked 2026-11-09).
• Expert fields state rates of USD 60 to 150 per hour on several platforms.

Risks

• A stated rate is a ceiling or average, and no platform publishes the share of contributors who reach it.
• A rejected appeal can end payment, as Alignerr's FAQ states.
• Allegations in Boggs and Cavalier concern per-interaction pay and unpaid qualification time; neither is decided.
• Task-based offers that look like platform work exist; the guide on is AI training work legit covers how to check a platform.

Downsides

• Assessments take 30 minutes to 2 hours; DataAnnotation and Outlier do not state whether they are paid, and Alignerr's terms exclude pay for the interview.
• Contractor terms put tax withholding and filing on the contributor.
• Eligibility limits apply: Handshake's Fellowship is US-based only, and Outlier excludes some countries.

Alternatives

• Participant panels with a stated minimum, such as Prolific's USD 8 per hour (page of 2026-07-17).
• Salaried employment, which carries benefits and withholding that contractor work lacks.
• Licensing company records to AI developers, a different route covered on the parent hub.
• Not doing the work, which is a valid choice.

How the pay figures on this page are checked

Every rate on this page is the named platform's own statement on its own page, re-read on the check date shown, and platforms appear alphabetically whether or not any pays referral fees. The method is set out in how we source, date and check every figure.

Salary aggregator and job-board figures are not used, because they are not primary sources. A figure that could not be re-read on its own page is not stated. The page is information, not tax or legal advice.

Where pay meets platforms, human data companies and payouts to data sellers

This page answers what AI training work pays; the pages below answer who runs the work, what the work involves, how contributors are classified and what companies are paid for their records.

• the human data companies behind these platforms
• what the ranking and rating work feeds: RLHF data
• what companies are told they will be paid for data, advertised vs confirmed
• how DataPacts is funded
• expert contributor work by field
• AI training work by language
• how AI trainer assessments work
• contractor or employee: how contributors are classified

Frequently asked questions

How much do people get paid to train AI?

Platforms state entry rates of about USD 15 to 50 per hour and expert rates of about USD 60 to 150 per hour. These are claims checked on 2026-11-09, not confirmed earnings; the table above lists each platform's figure with its source.

Is AI trainer pay a salary?

No; platforms publish contract rates per hour or per task, not salaries. Platforms such as Outlier and Alignerr engage contributors as independent contractors, and salary-style annual figures from aggregators are not used on this page.

Which field of AI training work pays the most?

Coding, medicine, law and finance state the highest rates. DataAnnotation states up to USD 150+ per hour for coding and up to USD 125+ for experts, and one Surge AI listing states USD 200 to 400 per hour (checked 2026-11-09).

How much does Outlier pay AI trainers?

Outlier publishes no rate; it states that it pays hourly or per task. Rates vary by expertise, project complexity and location, and payment is weekly on Tuesdays (checked 2026-11-09).

What do micro1 and Invisible Technologies pay AI trainers?

No contributor rate is on record for either. Invisible Technologies states 24,000+ Meridial experts, a scale claim and not a pay rate (checked 2026-11-09).

Are AI training assessments paid?

DataAnnotation and Outlier do not state it, and Alignerr's terms say it has no obligation to pay for the interview. DataAnnotation's starter assessment takes about an hour; Cavalier v. Surge Labs alleges unpaid training and qualification time, which is an allegation, pending as of 2026-11-09.