01 · The short answer
The industry is real, and here is the proof
The strongest evidence is not marketing, it is what the AI companies say and spend. OpenAI's InstructGPT paper, the research behind the instruction-following behaviour in modern ChatGPT models, states that model behaviour is determined in part by the human feedback obtained from its contractors. The labs need human judgment, and they say so in their own research.
The money confirms it. In June 2025 Meta paid roughly $14.3 billion for a stake of about 49% in Scale AI, a data-labelling company. Mercor raised $350 million at a $10 billion valuation in October 2025. Toloka raised $72 million led by Bezos Expeditions. Grand View Research put the data annotation tools market at $1.0 billion in 2023 and projects $5.33 billion by 2030. Nobody invests at that scale in a category that does not exist.
So when someone asks whether data annotation is legit, the honest answer is that the work is real and the demand is real. The fair criticisms are about how some of that work has been organised and paid, which is the next section.
02 · The platforms
What each platform actually publishes
Below is what each company states about itself. Where a platform does not publish a pay rate, that is noted rather than filled in with a figure from a job aggregator, because those numbers routinely get attributed to companies that never said them.
DataAnnotation.tech
- Who they are:
- Forbes reports it is an affiliate of Surge AI
- Pay:
- Publishes rate ranges: generalists $25 to $30+ per hour, domain experts $50 to $100+, language and localisation $20 to $50+
- Worth knowing:
- Its FAQ states it has paid over $20 million to contractors since 2020, across over 100,000 contractors
Outlier
- Who they are:
- Its own FAQ describes it as a platform operated by Scale AI
- Pay:
- Declines to publish a rate. Its FAQ says rates vary by expertise, project complexity and location. Pays weekly on Tuesdays
- Worth knowing:
- Its homepage states $500M+ paid to experts and 100K+ experts
Surge AI
- Who they are:
- Surge Labs Inc, San Francisco, founded 2020 by Edwin Chen
- Pay:
- Forbes reports the work pays at least $20 an hour, and more than $40 for more specialised tasks
- Worth knowing:
- Forbes reports $1.2 billion revenue for 2024. Publicly documents Anthropic using its platform for RLHF
Mercor
- Who they are:
- Raised a $350 million Series C at a $10 billion valuation in October 2025
- Pay:
- Publishes live per-role rates on its jobs site, for example a generalist role listed at $50 to $60 per hour
- Worth knowing:
- Publishes a supported-countries list, which is unusual in this industry and useful
Toloka
- Who they are:
- A Nebius Group subsidiary. Raised $72 million led by Bezos Expeditions in May 2025
- Pay:
- Via Mindrift: entry level $15 to $30 per hour, domain experts $60 to $100+
- Worth knowing:
- Note the change: Toloka states tasks and projects have moved to Mindrift, so that is where workers now go
TELUS Digital
- Who they are:
- A publicly listed company, TSX and NYSE ticker TIXT
- Pay:
- Publishes no blanket rate. Pay is stated per job posting on its careers site
- Worth knowing:
- Recruits through an AI Community programme with roles listed per country
Appen
- Who they are:
- Publicly listed on the ASX, ticker APX. Reported statutory revenue of USD 230.8 million for FY25
- Pay:
- Publishes no blanket rate
- Worth knowing:
- Two changes worth knowing: Google terminated its contract, which Appen disclosed was over 30% of turnover, effective March 2024. Its worker platform Appen Connect is now CrowdGen by Appen
Clickworker
- Who they are:
- Clickworker GmbH, a German company
- Pay:
- Publishes no rate, and is unusually direct about it: its own page says monthly income can vary greatly, cannot be guaranteed, and is no substitute for full-time employment
- Worth knowing:
- Microtask model, so earnings depend heavily on task availability
Prolific
- Who they are:
- Prolific Academic Ltd, registered in England, company number 8991598, founded at Oxford in 2014
- Pay:
- Enforces a minimum of £6 or $8 per hour and recommends £9 or $12. Cash-out threshold £6 or $6
- Worth knowing:
- Research participation rather than annotation, but often appears in the same searches
03 · The honest part
What has genuinely gone wrong
This is the section most articles skip, and it is why so many people search whether this work is a scam. These are documented, attributable events. None of them means a platform is fraudulent, and each is reported here with its source.
Delayed and cancelled payments
A Washington Post investigation in August 2023 interviewed 36 freelancers based in the Philippines working through Scale AI and Remotasks. All but two said payments had been delayed, reduced or cancelled after they completed tasks.
Unpaid training and assessment time, litigated and settled
Class actions in the Northern District of California against Scale AI and co-defendants alleged that workers recruited on promises of $25 to $40 per hour were paid for roughly five of every ten hours worked, partly because time spent reading instructions and completing training was unpaid. The matter was resolved by a $12.5 million settlement. A settlement is not a finding of wrongdoing, and it should not be read as one.
Entire countries cut off
Rest of World reported in March 2024 that Remotasks shut down entirely in Kenya, Nigeria and Pakistan, and halted new signups in Thailand, Vietnam and Poland, with Kenyan workers losing access that month. Scale AI said the discontinuation was due to enhanced security protocols, that notices were sent three weeks ahead, and that contractors in Kenya did not receive the notice because of operational errors.
The gap between advertised and realised pay
The Fairwork project at the Oxford Internet Institute reported in its 2025 cloudwork research that microworkers spent on average 27% of their time on unpaid tasks and earned an average of $2.15 per hour. That is the single most useful number on this page for setting expectations, because it quantifies the gap between an advertised hourly rate and what an hour of your life actually returns.
You will also find widespread reports on worker forums such as Reddit about accounts being deactivated with a balance owed, and per-task rates dropping without notice. We have not been able to corroborate those through any named outlet, so treat them as worker reports rather than established fact. They are worth knowing about before you rely on this as income.
04 · Can you actually sign up?
Country eligibility, from published lists only
This is the question that decides whether any of this is relevant to you, and almost nobody answers it. The table below is built only from eligibility documentation the platforms publish themselves.
| Country | Outlier | Mercor | TELUS Digital | Prolific |
|---|---|---|---|---|
| India | ||||
| United Kingdom | ||||
| Canada | ||||
| Philippines | ||||
| South Africa | ||||
| Kenya | ||||
| Nigeria |
The standout: Mercor publishes a supported-countries list that includes Kenya and Nigeria, while Outlier's published working guidelines do not. If you are in West or East Africa, that difference is the most useful fact on this page. Prolific limits participation to most OECD countries, which excludes India, Kenya, Nigeria and the Philippines. Outlier also states that using a VPN to mask your location is not permitted, so working around a country restriction risks your account.
DataAnnotation.tech is deliberately absent from this table. Its own About page refers to 50+ countries while third-party sites list a far shorter set, and we could not find official eligibility documentation, so publishing either figure would be guesswork.
05 · Actual fraud
How to spot a real scam
Genuine job fraud exists in this space, and it looks nothing like the platforms above. The Better Business Bureau reports roughly 50,000 employment-scam reports to its Scam Tracker over three years, with reports doubling in 2025. These are the patterns consumer-protection authorities actually document.
You are asked to pay anything at all
The FTC states plainly that honest employers will never ask you to pay to get a job, and that anyone who does is a scammer. That covers starter kits, training, certifications and placement fees. Every legitimate platform on this page pays you; none charges a signup fee. This single test separates the real industry from the fraud.
A cheque arrives before any work does
The FTC warns that no honest employer sends you a cheque to deposit and then asks you to send part of the money onward or buy gift cards with it. The cheque bounces later and the bank reclaims the full amount from you. The BBB reports that 36% of job-scam complaints it received involved a fake cheque.
The recruiter contacted you out of the blue
The FTC has issued specific alerts about scammers impersonating well-known companies to recruit for fake jobs on LinkedIn and other job platforms, and about unsolicited job-offer texts. Real annotation platforms make you apply and complete an assessment. They do not message you first offering work.
The conversation moves to WhatsApp or Telegram
The BBB documents this as a common pattern in employment scams: initial contact on a job platform, then a rapid move to an encrypted messaging app where there is no record and no platform oversight. Legitimate platforms keep hiring inside their own systems.
You are asked to pay to withdraw your own earnings
The BBB lists fees to unlock withdrawals as a red flag, and reports that task-scam complaints more than doubled in 2025. No real platform charges you to access money you have already earned.
The one test that settles it
Every platform named on this page pays you and charges nothing to join. If money is flowing from you to them at any point, for training, a starter kit, a certification, a placement fee or to unlock a withdrawal, it is not a job. The FTC puts it bluntly: honest employers will never ask you to pay to get a job, and anyone who does is a scammer. You can report suspected job scams to the FTC at reportfraud.ftc.gov.
06 · Questions
Frequently asked questions
Is AI data annotation work legit?
The industry is unambiguously real. OpenAI's own InstructGPT paper states that model behaviour is determined in part by human feedback obtained from its contractors, and Meta paid roughly $14.3 billion in 2025 for a stake of about 49% in Scale AI, a data-labelling company. Grand View Research values the data annotation tools market at $1.0 billion in 2023, projected to reach $5.33 billion by 2030. So the work exists and the money is real. That is a separate question from whether any given platform treats workers well, which is where the documented complaints sit.
Is DataAnnotation.tech legit?
It is a real operating business. Forbes reports it is an affiliate of Surge AI, and DataAnnotation publishes specific rate ranges (generalists $25 to $30+ per hour) and states it has paid over $20 million to more than 100,000 contractors since 2020. What we cannot tell you is which countries it accepts: its own About page says 50+ countries while third-party sites claim a much shorter list, and we found no official eligibility page, so we are not going to guess.
Is Outlier AI legit?
Yes, it is a real platform and its own FAQ describes it as operated by Scale AI. Its homepage states it has paid over $500 million to experts. Note two things: it deliberately does not publish an hourly rate, saying rates vary by expertise, project complexity and location, so any specific figure you see quoted elsewhere did not come from Outlier. And its published working guidelines exclude several countries, including Kenya and Nigeria.
Why do so many people online say data annotation is a scam?
Because real, documented problems get compressed into the word scam. Workers interviewed by the Washington Post in 2023, 36 freelancers based in the Philippines, said payments had been delayed, reduced or cancelled after work was completed, with all but two reporting this. Class actions in California alleged workers recruited on promises of $25 to $40 per hour were paid for roughly half the hours they worked, partly because training and instruction time was unpaid; those claims were resolved by a $12.5 million settlement without an admission of wrongdoing. Those are serious grievances about working conditions, which is different from the platform being fake.
Can I do data annotation work from Kenya or Nigeria?
It depends entirely on the platform, and this is where most guides are unhelpfully vague. Mercor publishes a supported-countries list that includes both Kenya and Nigeria. Outlier's published working guidelines, effective July 2026, do not include either. There is history here: Rest of World reported in March 2024 that Remotasks shut down entirely in Kenya, Nigeria and Pakistan, with Kenyan workers cut off that month. Scale AI said the discontinuation was due to enhanced security protocols and that Kenyan contractors did not receive advance notice because of operational errors.
How much can you actually earn doing data annotation?
Advertised rates and realised earnings are different numbers, and you should plan on the gap. Platforms that publish rates advertise figures like $25 to $30 per hour for generalist work and higher for specialists. But the Fairwork project at the Oxford Internet Institute found in its 2025 cloudwork research that microworkers spent on average 27% of their time on unpaid tasks and earned an average of $2.15 per hour. The difference is largely unpaid time: reading instructions, taking assessments, and waiting for work to appear.
How do I tell a real annotation platform from a scam?
One test does most of the work: real platforms pay you and never ask you for money. The FTC states that honest employers will never ask you to pay to get a job, and that anyone who does is a scammer. Add three more checks: they never send you a cheque before you have worked, they do not contact you out of the blue offering work, and they do not move the conversation to WhatsApp or Telegram. Every platform named on this page charges nothing to join.
Keep going
Company figures on this page are those each company publishes about itself, checked in July 2026. Reported events are attributed to the outlet, court record or research body that reported them, and a legal settlement is not a finding of wrongdoing. Pay rates, eligibility and platform terms change frequently, so confirm current details on the platform's own site before applying. This is general information, not employment or financial advice.