Key findings
- 1
52 studies and investigations from 2016 to 2026 are recorded, with 130 figures.
- 2
A Data-Driven Analysis of Workers' Earnings on Amazon Mechanical Turk (2018): share earning above minimum wage, 111 workers (4%) earned more than $7.25/h, the U.S. federal minimum wage.
- 3
Quantifying the Invisible Labor in Crowd Work (2021): unpaid time share, 33% of their time daily on invisible labor.
- 4
Working Without Borders: The Promise and Peril of Online Gig Work (2023) estimates workforce size (upper bound) at as high as 435 million online gig workers.
Figure 1
What the studies found
What have researchers and reporters measured about pay and conditions in data work?
| Study | Year | Measure | Finding |
|---|---|---|---|
| Prolific's payment modelProlific | 2026 | minimum hourly pay rule | minimum hourly rate of £6 / $8 (recommended £9 / $12) |
| Surveillance at Scale: How AI Training Platforms Gather Personal Data from European WorkersDiPLab and Reversing.Works | 2026 | mean monthly earnings | €1,214 a month on average, 30% below the average cost of living in their countries |
| Surveillance at Scale: How AI Training Platforms Gather Personal Data from European WorkersDiPLab and Reversing.Works | 2026 | hourly pay rate by language (identical projects) | $10 an hour for Portuguese speakers to $50 for Swiss German speakers, on identical projects |
| Surveillance at Scale: How AI Training Platforms Gather Personal Data from European WorkersDiPLab and Reversing.Works | 2026 | share who earned nothing in the study period | 1 in 4 workers earned nothing during the study period |
| The Benchmark Ceiling: Human Judgment, Evaluation Scarcity, and the Political Economy of AI Capability MeasurementarXiv | 2026 | pay premium over primary job (mean) | approximately 15 to 25 percent above their reported primary role compensation, per hour |
| The Benchmark Ceiling: Human Judgment, Evaluation Scarcity, and the Political Economy of AI Capability MeasurementarXiv | 2026 | pay premium over primary job (hardest, scarcest tasks) | 35 to 50 percent in the upper quartile of task difficulty and credential scarcity; near zero in commodity task families |
| The Benchmark Ceiling: Human Judgment, Evaluation Scarcity, and the Political Economy of AI Capability MeasurementarXiv | 2026 | premium gap, hardest vs easiest task quartile | approximately 28 percentage points higher in the top difficulty quartile than in the bottom, after controlling for domain, credentials and experience |
| Fairwork Cloudwork Ratings 2025: Advancing Standards in Digital Labour and AI Supply Chain GovernanceFairwork | 2025 | platforms meeting minimum-wage principle | four platforms (ComeUp, Translated, Elharefa, Terawork) of 16 |
| Is tech 'taking over'? How the online workforce shifts towards tech work (Online Labour Index 2020)Online Labour Observatory | 2025 | share of countries where software development is the top online occupation | from 29% of countries in 2019 to 43% in 2022 |
| Scroll. Click. Suffer. The Hidden Human Cost of Content Moderation and Data LabellingEquidem | 2025 | maximum daily working hours reported | up to 18 or even 20 hours a day (the Philippines) |
| Scroll. Click. Suffer. The Hidden Human Cost of Content Moderation and Data LabellingEquidem | 2025 | workers interviewed | interviews with 113 workers across Colombia, Ghana, Kenya, and The Philippines |
| Scroll. Click. Suffer. The Hidden Human Cost of Content Moderation and Data LabellingEquidem | 2025 | bonus share of total pay | bonuses accounted for more than 70% of total remuneration in some cases |
| "If I Had Another Job, I Would Not Accept Data Annotation Tasks": How Syrian Refugees in Lebanon Train AI (Data Workers' Inquiry)Data Workers' Inquiry | 2024 | piece rate | 0.014 euros per image; 10,000 images for 140 euros |
| Data Workers' InquiryData Workers' Inquiry | 2024 | number of worker-researchers | 19 data workers in Venezuela, Kenya, Syria, Brazil, France and Germany |
| Fairwork Cloudwork Ratings 2023: Work in the Planetary Labour MarketFairwork | 2023 | platforms meeting minimum-wage principle | two platforms (ComeUp and Terawork) of 15 |
| Working Without Borders: The Promise and Peril of Online Gig WorkWorld Bank | 2023 | workforce size (lower bound) | 154 million unique registered online gig workers |
| Working Without Borders: The Promise and Peril of Online Gig WorkWorld Bank | 2023 | workforce size (upper bound) | as high as 435 million online gig workers |
| Working Without Borders: The Promise and Peril of Online Gig WorkWorld Bank | 2023 | workforce size (main gig workers) | 132.5 million main gig workers |
| Working Without Borders: The Promise and Peril of Online Gig WorkWorld Bank | 2023 | share of global labour force (lower bound) | 4.4 percent of the global labor force |
| Working Without Borders: The Promise and Peril of Online Gig WorkWorld Bank | 2023 | share of global labour force (upper bound) | 12.5 percent of the global labor force |
| Working Without Borders: The Promise and Peril of Online Gig WorkWorld Bank | 2023 | women's earnings as share of men's | women earning just 68 percent of men's wages |
| Work in the Planetary Labour Market: Fairwork Cloudwork Ratings 2022Fairwork | 2022 | platforms meeting minimum-wage principle | not awarded to any platform |
| How many online workers are there in the world? A data-driven assessmentOpen Research Europe | 2021 | workforce size (registered profiles) | 163 million freelancer profiles |
| How many online workers are there in the world? A data-driven assessmentOpen Research Europe | 2021 | workforce size (ever obtained work) | approximately 19 million obtained work at least once; 5 million completed at least 10 projects or earned at least $1000 |
| Quantifying the Invisible Labor in Crowd WorkarXiv | 2021 | unpaid time share | 33% of their time daily on invisible labor |
| Quantifying the Invisible Labor in Crowd WorkarXiv | 2021 | median hourly earnings (including invisible labour) | $2.83 median hourly wage counting invisible labor ($3.76 if it is ignored) |
| Work in the Planetary Labour Market: Fairwork Cloudwork Ratings 2021Oxford Internet Institute | 2021 | platforms meeting minimum-wage principle | only 2 of the 11 platforms (Workana and Appen) |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of workInternational Labour Organization | 2021 | mean hourly earnings (paid and unpaid time) | US$3.4 average hourly earnings in a typical week |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of workInternational Labour Organization | 2021 | median hourly earnings | half of the workers earn less than US$2.1 per hour |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of workInternational Labour Organization | 2021 | mean hourly earnings (microtask platforms) | US$3.3 on microtask platforms (US$7.6 on freelance platforms) |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of workInternational Labour Organization | 2021 | mean hourly earnings (freelance platforms) | US$7.6 on freelance platforms |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of workInternational Labour Organization | 2021 | earnings gap, developing vs developed countries | 60 per cent less, after controlling for basic characteristics and tasks |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of workInternational Labour Organization | 2021 | unpaid time share | about one third of their time, or eight hours, of a 27-hour typical week spent on unpaid work |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of workInternational Labour Organization | 2021 | weekly hours | 27 hours on average in a typical week, including paid and unpaid work |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of workInternational Labour Organization | 2021 | unpaid minutes per paid hour | about 23 minutes (freelance) and 20 minutes (microtask) of unpaid work per hour of paid tasks |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of workInternational Labour Organization | 2021 | share working over 40 hours a week | half work 16 hours or less, 20 per cent work over 40 hours |
| A Data-Driven Analysis of Workers' Earnings on Amazon Mechanical TurkarXiv | 2018 | median hourly earnings | $1.77/h |
| A Data-Driven Analysis of Workers' Earnings on Amazon Mechanical TurkarXiv | 2018 | mean hourly earnings | $3.13/h |
| A Data-Driven Analysis of Workers' Earnings on Amazon Mechanical TurkarXiv | 2018 | share earning above minimum wage | 111 workers (4%) earned more than $7.25/h, the U.S. federal minimum wage |
| A Data-Driven Analysis of Workers' Earnings on Amazon Mechanical TurkarXiv | 2018 | median hourly earnings excluding unpaid work | $3.18/h (median) and $6.19/h (mean) if unpaid work is ignored |
| A Data-Driven Analysis of Workers' Earnings on Amazon Mechanical TurkarXiv | 2018 | mean hourly rate paid by requesters | $11.58/h on average |
| Demographics and Dynamics of Mechanical Turk WorkersPanos Ipeirotis (author's site) | 2018 | workforce size | more than 100K workers available; more than 2K active at any given time |
| Demographics and Dynamics of Mechanical Turk WorkersPanos Ipeirotis (author's site) | 2018 | worker half-life on platform | half-life of 12 to 18 months |
| Digital labour platforms and the future of workInternational Labour Organization | 2018 | mean hourly earnings (paid work only) | US$4.43 per hour |
| Digital labour platforms and the future of workInternational Labour Organization | 2018 | mean hourly earnings (paid and unpaid time) | US$3.31 per hour |
| Digital labour platforms and the future of workInternational Labour Organization | 2018 | median hourly earnings (paid and unpaid time) | US$2.16 per hour |
| Digital labour platforms and the future of workInternational Labour Organization | 2018 | share below minimum wage | about 48 per cent below US$7.25 counting paid work only, 64 per cent counting unpaid work |
| Digital labour platforms and the future of workInternational Labour Organization | 2018 | share above minimum wage | only 7 per cent above the German minimum wage of EUR 8.84 per hour (paid and unpaid hours) |
| Digital labour platforms and the future of workInternational Labour Organization | 2018 | unpaid minutes per paid hour | 20 minutes on unpaid activities for every hour of paid work |
| Digital labour platforms and the future of workInternational Labour Organization | 2018 | weekly hours | 24.5 hours per week (18.6 hours paid, 6.2 hours unpaid) |
| Digital labour platforms and the future of workInternational Labour Organization | 2018 | share wanting more work | 88 per cent would like to do more crowdwork, wanting 11.6 more hours per week on average |
| Digital labour platforms and the future of workInternational Labour Organization | 2018 | mean hourly earnings by region (paid and unpaid time) | US$1.33 per hour (Africa), against US$4.70 (Northern America) |
| Online Labour Index: Measuring the Online Gig Economy for Policy and ResearchMunich Personal RePEc Archive | 2018 | occupation share of vacancies (software development and technology) | roughly one third of the vacancies |
| Online Labour Index: Measuring the Online Gig Economy for Policy and ResearchMunich Personal RePEc Archive | 2018 | annualized growth of online gig work | an annualized rate of 14 percent |
| DataAnnotation: advertised expert ratesDataAnnotation | 2026 | advertised expert hourly rate | expert rates of up to $75 to $125+/hr |
| Doctors and Lawyers Are Joining AI's $400-an-Hour Gig EconomyBloomberg (via Bloomberg Law) | 2026 | mean hourly pay rate offered (company figure) | $110 per hour on average at Mercor; the share of tasks paying that rate is not known |
| Doctors and Lawyers Are Joining AI's $400-an-Hour Gig EconomyBloomberg (via Bloomberg Law) | 2026 | advertised hourly pay (expert roles) | up to $120 per hour (software engineers), $170 (condensed matter experts), $400 (civil rights attorneys) |
| Doctors and Lawyers Are Joining AI's $400-an-Hour Gig EconomyBloomberg (via Bloomberg Law) | 2026 | share of participants who are experts | 30% of Prolific's participants are its “experts”; how many tasks they complete was not disclosed |
| Handshake AI Program: advertised ratesHandshake | 2026 | advertised hourly pay (AI evaluation specialist) | up to $17/hr (AI Evaluation Specialist) |
| Handshake AI Program: advertised ratesHandshake | 2026 | advertised hourly pay (expert role) | up to $125/hr (Game Developer) |
| How China is using human labor to win the humanoid robot data raceRest of World | 2026 | hourly pay | 20 yuan ($3) per hour, six hours a day |
| I refused to train the AI that could replace meRest of World | 2026 | offered hourly pay vs national minimum wage | 600 rand ($37) an hour, against a national minimum wage of 30.23 rand ($2) an hour |
| I refused to train the AI that could replace meRest of World | 2026 | hourly pay | 250 rupees ($2.60) an hour |
| Lawyers are training AI how to think like themBusiness Insider | 2026 | advertised hourly pay | between $100 and $200 an hour for legal experts (Mercor) |
| Mercor competitor Deccan AI raises $25M, sources experts from IndiaTechCrunch | 2026 | hourly earnings range (company statement) | about $10 to $700 per hour; top contributors up to $7,000 a month |
| Now it's China's experts who are gig workers training AIRest of World | 2026 | pay per task | 100 to 500 yuan ($15 to $74) per task; a task usually takes a few hours |
| Surveillance at Scale: How AI Training Platforms Gather Personal Data from European WorkersDiPLab and Reversing.Works | 2026 | advertised rate against rate received | an advertised $20 an hour could drop to half after 10 to 20 minutes of work |
| The AI-powered World Cup runs on thousands of data workersRest of World | 2026 | pay per match | about 60 euros ($70) per match, plus transportation costs |
| The gig workers who are training humanoid robots at homeMIT Technology Review | 2026 | hourly pay | $15 an hour (a data recorder in Nigeria) |
| These workers help train robots who could one day steal their jobsNPR | 2026 | hourly pay | about $2.60 an hour, around $120 a month |
| These workers help train robots who could one day steal their jobsNPR | 2026 | hourly pay | about $2.60 an hour to record herself |
| Work and Wages of Data Workers in KenyaWageIndicator Foundation | 2026 | mean monthly earnings (women, main platform or BPO) | Ksh 21,849.72 a month (women) |
| Work and Wages of Data Workers in KenyaWageIndicator Foundation | 2026 | mean monthly earnings (men, main platform or BPO) | Ksh 26,106.83 a month (men) |
| Work and Wages of Data Workers in KenyaWageIndicator Foundation | 2026 | share whose earnings cover living expenses | 20% of women and 25% of men; against a Nairobi minimum wage for comparable occupations of Ksh 24,818.31 a month |
| Work and Wages of Data Workers in KenyaWageIndicator Foundation | 2026 | gender pay gap | 19% (Ksh 21,850 for women against Ksh 26,107 for men) |
| Work and Wages of Data Workers in KenyaWageIndicator Foundation | 2026 | share who did unpaid work | 27% performed unpaid work; some reported up to 200 unpaid hours |
| Work and Wages of Data Workers in KenyaWageIndicator Foundation | 2026 | weekly hours | 33.9 hours a week on average (individual answers from 4 to 70) |
| Workers are getting paid to teach AI how to do their jobsCBS News | 2026 | mean hourly pay rate offered (company figure) | an average of $105 an hour (Mercor), some people earn considerably more |
| AI Is Learning to Do the Jobs of Doctors, Lawyers, and ConsultantsTIME | 2025 | advertised mean hourly rate | an average hourly rate of $81 per hour, reaching over $200 per hour for Senior Domain Experts |
| AI Is Learning to Do the Jobs of Doctors, Lawyers, and ConsultantsTIME | 2025 | layoffs | 500 generalist data workers laid off (September 2025) |
| Data Work in Egypt: Who Are the Workers Behind Artificial Intelligence?arXiv (DiPLab report) | 2025 | mean monthly earnings | EGP 2,793.37 (USD 58.76) a month from data work platforms |
| Data Work in Egypt: Who Are the Workers Behind Artificial Intelligence?arXiv (DiPLab report) | 2025 | mean hourly earnings | USD 1.22 per hour on average |
| Data Work in Egypt: Who Are the Workers Behind Artificial Intelligence?arXiv (DiPLab report) | 2025 | share who depend on platform work to pay bills | 3 out of 4 (514 of 683) depend on platforms to earn enough to pay their bills |
| Desperate for work, translators train the AI that's putting them out of workRest of World | 2025 | hourly pay | $20 for every hour worked |
| Fairwork Cloudwork Ratings 2025: Advancing Standards in Digital Labour and AI Supply Chain GovernanceFairwork | 2025 | share experiencing non-payment | 31.2% non-payment, 38% late payment |
| Fairwork Cloudwork Ratings 2025: Advancing Standards in Digital Labour and AI Supply Chain GovernanceFairwork | 2025 | share feeling secure about earnings | only 35% feel secure about their earnings all or most of the time |
| Fairwork Cloudwork Ratings 2025: Advancing Standards in Digital Labour and AI Supply Chain GovernanceFairwork | 2025 | share for whom platform work is main income | 58% say cloudwork is their primary source of income |
| How thousands of 'overworked, underpaid' humans train Google's AI to seem smartThe Guardian | 2025 | starting hourly pay | $16 an hour (generalist raters), $21 an hour (super raters) |
| How thousands of 'overworked, underpaid' humans train Google's AI to seem smartThe Guardian | 2025 | workforce size | almost 2,000 super raters at peak, shrinking to roughly 1,500 after rolling layoffs in 2025 |
| How thousands of 'overworked, underpaid' humans train Google's AI to seem smartThe Guardian | 2025 | time allowed per task | timer of 30 minutes per task shrank to 15, for about 500 words per response |
| The hidden Kenyan workers training China's AI modelsRest of World | 2025 | daily workload | up to 26,000 videos per person a day, up to 12 hours for beginners |
| Venezuelan migrants relied on clickwork to survive. Now AI is replacing themRest of World | 2025 | piece rate | 5 to 25 cents per task |
| Google cuts off an $83 million ghost labor contractRest of World | 2024 | hourly pay | $14.50 an hour after a contract-wide raise (2023) |
| Labelers training AI say they're overworked, underpaid and exploited by big American tech companiesCBS News | 2024 | hourly pay | $2 per hour, gross before tax |
| Microtrabalho no Brasil: Quem são os trabalhadores por trás da inteligência artificialBBC News Brasil | 2024 | mean monthly earnings | R$ 583,71 por mes, about 15.5 hours a week |
| Microtrabalho no Brasil: Quem são os trabalhadores por trás da inteligência artificialBBC News Brasil | 2024 | mean hourly earnings | US$ 1,60 (about R$ 9) per hour |
| Microtrabalho no Brasil: Quem são os trabalhadores por trás da inteligência artificialBBC News Brasil | 2024 | share earning only enough for basic bills | 66% earn only enough to pay the most basic bills |
| Scale AI's Remotasks platform is dropping whole countries without explanationRest of World | 2024 | countries where the platform shut down | shut down entirely in Kenya, Nigeria, and Pakistan; new sign-ups stopped in Thailand, Vietnam, and Poland |
| Side Hustle or Scam? What to Know About Data Annotation WorkTIME | 2024 | advertised hourly pay | around $20 per hour; $40 for coding tasks (DataAnnotation.tech); $60 per hour for chemistry tasks (Outlier.ai) |
| AI Is a Lot of WorkThe Verge | 2023 | hourly earnings range | dropped to between $1 and $3 per hour (early 2023), from about $5 to $10 when Remotasks arrived |
| AI Is a Lot of WorkThe Verge | 2023 | hourly earnings range | generally between $10 and $25 per hour |
| AI Is a Lot of WorkThe Verge | 2023 | hourly earnings range | between $15 and $30 per hour; Surge says it has 100,000 annotators |
| AI Is a Lot of WorkThe Verge | 2023 | advertised hourly pay | $45 an hour (law), $25 an hour (poetry) |
| China's AI boom depends on an army of exploited student internsRest of World | 2023 | monthly pay | 3000 to 4000 yuan ($409 to $545) a month, of which schools took 600 to 1000 yuan |
| China's AI boom depends on an army of exploited student internsRest of World | 2023 | share of workforce who are student interns | 20% to 30% of China's data annotation work force |
| Fairwork Cloudwork Ratings 2023: Work in the Planetary Labour MarketFairwork | 2023 | mean hourly earnings | an average of $2.15 USD an hour ($50.30 per week) |
| Fairwork Cloudwork Ratings 2023: Work in the Planetary Labour MarketFairwork | 2023 | unpaid time share | 6 hours and 18 minutes of 23 hours 30 minutes a week unpaid (26.85%) |
| OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less ToxicTIME | 2023 | hourly take-home pay range | between around $1.32 and $2 per hour |
| OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less ToxicTIME | 2023 | hourly rate paid by client to vendor | $12.50 per hour paid by OpenAI to Sama, six to nine times workers' take-home pay |
| OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less ToxicTIME | 2023 | monthly base salary | 21,000 Kenyan shillings ($170) per month |
| OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less ToxicTIME | 2023 | workload per shift | between 150 and 250 passages of text per nine-hour shift |
| OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less ToxicTIME | 2023 | hourly take-home pay range (company statement) | between $1.46 and $3.74 per hour after taxes; 70 passages per nine-hour shift |
| Scale AI is on a hiring spree for speakers of under-represented languagesRest of World | 2023 | advertised hourly pay by language | $21.55 per hour (German) vs $1.43 per hour (Telugu) |
| Scale AI is on a hiring spree for speakers of under-represented languagesRest of World | 2023 | advertised hourly pay by country | up to $8.20 an hour (Portugal) vs $3.97 per hour (Brazil) for identical jobs |
| Why Silicon Valley's biggest AI developers are hiring poetsRest of World | 2023 | advertised hourly pay (standard vs expert) | $13.98 per hour (standard data worker) vs up to $50 per hour (expert poet, editor or creative writer) |
| How the AI industry profits from catastropheMIT Technology Review | 2022 | mean hourly earnings (company statement) | an average of a little more than 90 cents an hour |
| How the AI industry profits from catastropheMIT Technology Review | 2022 | workforce size | an estimated 200,000 Venezuelans, 75% of the platforms' workforces, by mid-2018 |
| How the AI industry profits from catastropheMIT Technology Review | 2022 | weekly paid-hours cap | anyone who worked over 60 hours a week would not be paid for extra time |
| Work in the Planetary Labour Market: Fairwork Cloudwork Ratings 2022Fairwork | 2022 | unpaid hours per week | over 8.5 hours per week on unpaid tasks |
| Work in the Planetary Labour Market: Fairwork Cloudwork Ratings 2022Fairwork | 2022 | share who completed a task and were not paid | almost a third (186 of 613) |
| Ghost Work: How to Stop Silicon Valley from Building a New Global Underclassghostwork.info (authors' book site) | 2019 | share of population that has done the work | an estimated 8 percent of Americans |
| Online Labour Index: Measuring the Online Gig Economy for Policy and ResearchMunich Personal RePEc Archive | 2018 | share of vacancies posted by US employers | roughly 52 percent of all the vacancies |
| Income security in the on-demand economy: Findings and policy lessons from a survey of crowdworkersInternational Labour Organization | 2016 | mean hourly earnings (US AMT workers) | average $5.55 per hour, median $4.65 (American AMT workers) |
| Income security in the on-demand economy: Findings and policy lessons from a survey of crowdworkersInternational Labour Organization | 2016 | median hourly earnings (Indian AMT workers) | Indian AMT: average $3.17, median $1.65; Crowdflower: average $1.77, median just under $1 |
| Income security in the on-demand economy: Findings and policy lessons from a survey of crowdworkersInternational Labour Organization | 2016 | unpaid time share | 23.2 per cent of working time unpaid; 18 minutes per hour of paid work |
| Income security in the on-demand economy: Findings and policy lessons from a survey of crowdworkersInternational Labour Organization | 2016 | share with work rejected or payment refused | 94 per cent have had work rejected or were refused payment |
| Research in the Crowdsourcing Age, a Case StudyPew Research Center | 2016 | share earning under $5 per hour | 52% report earning less than $5 an hour; 8% earn $8 or more |
| Research in the Crowdsourcing Age, a Case StudyPew Research Center | 2016 | share earning $8 per hour or more | Only 8% say they earn $8 per hour or more |
| Research in the Crowdsourcing Age, a Case StudyPew Research Center | 2016 | share of tasks paying 10 cents or less | 61% were microtasks paying 10 cents or less |
| Research in the Crowdsourcing Age, a Case StudyPew Research Center | 2016 | share for whom platform work is main income | 25% earn all or most of their incomes on MTurk; 53% say very little |
Sample. 130 figures from 52 studies, reports and investigations, each from the publication linked beside it.
Each figure is as the study states it, for its own population, place and year: they are not adjusted for inflation or currency.
Compiled by. AITraining.jobs Research, What Studies Find About Pay in AI Data Work, October 2026 edition.
Figure 2
Hourly pay in the studies, against what open roles advertise today
How do the hourly earnings researchers measured compare with today's advertised rates?
| Figure | Year | Measure | US dollars an hour |
|---|---|---|---|
| Handshake AI Program: advertised rates: advertised hourly pay (expert role)Handshake | 2026 | Advertised | $125.00 |
| Doctors and Lawyers Are Joining AI's $400-an-Hour Gig Economy: mean hourly pay rate offered (company figure)Bloomberg (via Bloomberg Law) | 2026 | Advertised | $110.00 |
| Workers are getting paid to teach AI how to do their jobs: mean hourly pay rate offered (company figure)CBS News | 2026 | Advertised | $105.00 |
| AI Is Learning to Do the Jobs of Doctors, Lawyers, and Consultants: advertised mean hourly rateTIME | 2025 | Advertised | $81.00 |
| Research in the Crowdsourcing Age, a Case Study: share earning under $5 per hourPew Research Center | 2016 | Measured | $52.00 |
| Desperate for work, translators train the AI that's putting them out of work: hourly payRest of World | 2024-2025 | Measured | $20.00 |
| Side Hustle or Scam? What to Know About Data Annotation Work: advertised hourly payTIME | 2024 | Advertised | $20.00 |
| Handshake AI Program: advertised rates: advertised hourly pay (AI evaluation specialist)Handshake | 2026 | Advertised | $17.00 |
| The gig workers who are training humanoid robots at home: hourly payMIT Technology Review | 2026 | Measured | $15.00 |
| Google cuts off an $83 million ghost labor contract: hourly payRest of World | 2023 | Measured | $14.50 |
| OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic: hourly rate paid by client to vendorTIME | 2021-2022 | Measured | $12.50 |
| A Data-Driven Analysis of Workers' Earnings on Amazon Mechanical Turk: mean hourly rate paid by requestersarXiv | 2014-2017 | Measured | $11.58 |
| Prolific's payment model: minimum hourly pay ruleProlific | 2026 | Measured | $8.00 |
| Research in the Crowdsourcing Age, a Case Study: share earning $8 per hour or morePew Research Center | 2016 | Measured | $8.00 |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of work: mean hourly earnings (freelance platforms)International Labour Organization | 2019-2020 | Measured | $7.60 |
| Income security in the on-demand economy: Findings and policy lessons from a survey of crowdworkers: mean hourly earnings (US AMT workers)International Labour Organization | 2015 | Measured | $5.55 |
| Digital labour platforms and the future of work: mean hourly earnings (paid work only)International Labour Organization | 2017 | Measured | $4.43 |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of work: mean hourly earnings (paid and unpaid time)International Labour Organization | 2017-2020 | Measured | $3.40 |
| Digital labour platforms and the future of work: mean hourly earnings (paid and unpaid time)International Labour Organization | 2017 | Measured | $3.31 |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of work: mean hourly earnings (microtask platforms)International Labour Organization | 2017 | Measured | $3.30 |
| A Data-Driven Analysis of Workers' Earnings on Amazon Mechanical Turk: median hourly earnings excluding unpaid workarXiv | 2014-2017 | Measured | $3.18 |
| A Data-Driven Analysis of Workers' Earnings on Amazon Mechanical Turk: mean hourly earningsarXiv | 2014-2017 | Measured | $3.13 |
| Quantifying the Invisible Labor in Crowd Work: median hourly earnings (including invisible labour)arXiv | 2021 | Measured | $2.83 |
| These workers help train robots who could one day steal their jobs: hourly payNPR | 2026 | Measured | $2.60 |
| These workers help train robots who could one day steal their jobs: hourly payNPR | 2026 | Measured | $2.60 |
| Digital labour platforms and the future of work: median hourly earnings (paid and unpaid time)International Labour Organization | 2017 | Measured | $2.16 |
| Fairwork Cloudwork Ratings 2023: Work in the Planetary Labour Market: mean hourly earningsFairwork | 2023 | Measured | $2.15 |
| World Employment and Social Outlook 2021: The role of digital labour platforms in transforming the world of work: median hourly earningsInternational Labour Organization | 2017-2020 | Measured | $2.10 |
| Labelers training AI say they're overworked, underpaid and exploited by big American tech companies: hourly payCBS News | 2024 | Measured | $2.00 |
| A Data-Driven Analysis of Workers' Earnings on Amazon Mechanical Turk: median hourly earningsarXiv | 2014-2017 | Measured | $1.77 |
| Income security in the on-demand economy: Findings and policy lessons from a survey of crowdworkers: median hourly earnings (Indian AMT workers)International Labour Organization | 2015 | Measured | $1.65 |
| Microtrabalho no Brasil: Quem são os trabalhadores por trás da inteligência artificial: mean hourly earningsBBC News Brasil | 2024 | Measured | $1.60 |
| Digital labour platforms and the future of work: mean hourly earnings by region (paid and unpaid time)International Labour Organization | 2017 | Measured | $1.33 |
| Data Work in Egypt: Who Are the Workers Behind Artificial Intelligence?: mean hourly earningsarXiv (DiPLab report) | 2024 | Measured | $1.22 |
| How the AI industry profits from catastrophe: mean hourly earnings (company statement)MIT Technology Review | 2022 | Measured | $0.90 |
Sample. Hourly figures in US dollars from the studies (35), and the advertised hourly pay of the AI training roles open on the last day of this site's live window.
Advertised pay is what a posting offers for an hour of accepted work; the studies mostly measure what workers actually earned per hour, including unpaid time searching for tasks. The two are not the same measure, and the gap between them is part of the finding.
Studies in other currencies are listed in the first figure, not converted here.
Compiled by. AITraining.jobs Research, What Studies Find About Pay in AI Data Work, October 2026 edition.
Figure 3
How many people do this work
How large is the online workforce that labels data and trains AI, by the studies' estimates?
| Study | Year | Measure | Estimate |
|---|---|---|---|
| Working Without Borders: The Promise and Peril of Online Gig WorkWorld Bank | 2023 | workforce size (upper bound) | as high as 435 million online gig workers |
| How many online workers are there in the world? A data-driven assessmentOpen Research Europe | 2021 | workforce size (registered profiles) | 163 million freelancer profiles |
| Working Without Borders: The Promise and Peril of Online Gig WorkWorld Bank | 2023 | workforce size (lower bound) | 154 million unique registered online gig workers |
| Working Without Borders: The Promise and Peril of Online Gig WorkWorld Bank | 2023 | workforce size (main gig workers) | 132.5 million main gig workers |
| How many online workers are there in the world? A data-driven assessmentOpen Research Europe | 2021 | workforce size (ever obtained work) | approximately 19 million obtained work at least once; 5 million completed at least 10 projects or earned at least $1000 |
| Demographics and Dynamics of Mechanical Turk WorkersPanos Ipeirotis (author's site) | 2018 | workforce size | more than 100K workers available; more than 2K active at any given time |
| Data Workers' InquiryData Workers' Inquiry | 2024 | number of worker-researchers | 19 data workers in Venezuela, Kenya, Syria, Brazil, France and Germany |
| How the AI industry profits from catastropheMIT Technology Review | 2022 | workforce size | an estimated 200,000 Venezuelans, 75% of the platforms' workforces, by mid-2018 |
| How thousands of 'overworked, underpaid' humans train Google's AI to seem smartThe Guardian | 2025 | workforce size | almost 2,000 super raters at peak, shrinking to roughly 1,500 after rolling layoffs in 2025 |
| China's AI boom depends on an army of exploited student internsRest of World | 2023 | share of workforce who are student interns | 20% to 30% of China's data annotation work force |
Sample. 10 estimates of the size of the online gig and data-work workforce.
Estimates count different populations (all online gig work, microtask work, one platform's registered workers) and are not comparable with one another.
Compiled by. AITraining.jobs Research, What Studies Find About Pay in AI Data Work, October 2026 edition.
For labs, data vendors and researchers
Buy the data behind this report
The figures above are a summary. Behind them are records we collect ourselves, every day:
0 roles tracked, each with its history
Every role on 0 job boards, read since n/a (every hour since 18 Oct 2026): its pay and place, and the first and last day it was open.
0 apply clicks in 28 days
Which roles people actually apply to, by field, pay and kind of work. It happens on this site, so nobody else can count it.
39 lab and vendor hiring boards
The job boards of 25 AI labs and 14 human-data vendors, read every day: who is hiring for human data, and who is hiring to buy it.
What you can buy
The records
The rows behind any figure, with their history, not only the totals.
A regular feed
The same figures every week or month, as files.
A custom cut or benchmark
Your countries, languages or fields, or your own platform against the market.
Building AI and need the experts behind this data? (you contract them yourself) or (we deliver the expert work).
Investing in AI-data companies? Hiring radar is these records role by role, every day, as files.
Method and definitions
The studies, reports and investigations themselves, each linked beside its figure; today's advertised rates from the job boards this site reads every day. Every row links the publication it is taken from, with the figure as that publication states it; nothing is estimated or converted. Each figure states what it includes beneath it.
- Measured earnings
- What a study found workers actually earned per hour (often counting unpaid time spent finding and qualifying for tasks), for its population, place and year.
- Advertised pay
- The hourly figure an open AI training role states (the middle of a stated range), on the last day of this site's live window. A median needs at least 3 roles stating a rate.
- As published
- Every figure keeps its study's own definition, currency and year: nothing is converted, deflated or pooled.
What this data cannot tell you
- Compiled from public sources: an event that was never announced or reported is missing, and private companies' figures are what they or the press stated, not audited.
- Studies measure different populations, years and definitions; set side by side they show a range, not a trend.
- Advertised pay is an offer for an accepted hour and leans towards the expert end of the work; measured earnings include unpaid time. The comparison shows the gap between the two, not one market's change.
How the job feeds are read and roles are classified: site methodology. How fresh, complete and consistent the data is, checked every day: data quality and coverage.
Editions
The live report is the dataset as last reviewed; new events are added as they are published. Each complete month is also published as an edition, frozen two days after the month ends, so a figure you cite from an edition never moves.
- Live dataAs last reviewed
- October 2026 edition (this page)Frozen 11 Nov 2026
Cite and reuse
Free to quote, chart and republish under CC BY 4.0 (creativecommons.org/licenses/by/4.0) with a link to this page. This edition is frozen, so its figures will read the same later.
Citation
AITraining.jobs (2026). What Studies Find About Pay in AI Data Work, October 2026 edition. https://cyberjobs.cc/research/research-on-ai-data-work-pay/2026-10
BibTeX
@misc{aitrainingjobs_research_on_ai_data_work_pay_2026_10,
author = {{AITraining.jobs}},
title = {What Studies Find About Pay in AI Data Work, October 2026 edition},
year = {2026},
howpublished = {\url{https://cyberjobs.cc/research/research-on-ai-data-work-pay/2026-10}},
}More research
- Funding and Valuations of AI Training Data CompaniesHow much money has gone into the companies that supply human data to AI labs, and at what valuations?Read the report
- Revenue and Scale of AI Training Data CompaniesHow much revenue do the companies that supply human data to AI labs make, and how many people do they say work for them?Read the report
- Lawsuits, Regulation and Labour Actions in AI Training WorkWhich lawsuits, rulings, investigations, laws and labour actions concern the people who train AI, and where do they stand?Read the report
- Which AI Labs Are Hiring for Human DataWhich AI labs are hiring for the teams that buy, run and evaluate human data, and for what roles?Read the report
- Who Is Buying AI Training DataWhich AI labs and data vendors have people whose job is to source, license or acquire training data right now?Read the report
- Where Human-Data Vendors Are BuildingHow fast are the human-data vendors hiring their own staff, and for which functions?Read the report
- The State of AI Training PayWhat does AI training work pay, by kind of work and by platform?Read the report