Data Annotation Jobs: What They Are, Who Hires and How to Start

Data annotation jobs have gone from a niche corner of the internet to one of the most searched forms of remote work, especially in the United States and India. The reason is simple. Every AI model needs humans to show it what a good answer looks like, and to check whether it got things right.

This guide uses only facts published by the companies themselves, by regulators and by journalists, all checked in November. It explains the work, who recruits, what they say about pay and payment, which countries are affected and how to tell a real opportunity from a scam.

What these figures show: the work is real and companies publish meaningful hourly ranges. Scammers have noticed the trend too. Keep both ideas in mind as you read on.

What are data annotation jobs, exactly?

Data annotation means adding labels or judgments that an AI system can learn from or be tested against. You might draw a box around a pedestrian in a photo, tag a customer message as a “complaint,” transcribe a voice clip or decide which of two chatbot answers is more accurate. Thousands of these small decisions together become training or evaluation data.

There is also a newer, “generative” side to the job: writing prompts, ranking and correcting AI responses, or rewriting a weak answer into a strong one. This work can feed RLHF (reinforcement learning from human feedback), the technique used to fine-tune chat assistants. Not every annotation job is RLHF, though, and much of the work is still classic categorization. For a closer look at RLHF and red-teaming, read our guide to getting paid to train AI with RLHF, annotation and red-teaming.

Search engine evaluation is a close cousin. Raters judge whether a result meets the intent behind a query (“Needs Met”) and how good a page is (“Page Quality”), including E-E-A-T: experience, expertise, authoritativeness and trustworthiness. Google publishes its Search Quality Rater Guidelines as a public PDF (182 pages in the version dated September 11). These ratings help measure search systems. They do not directly change how a given page ranks. Our dedicated guide covers search engine evaluator jobs and how to apply.

If you searched in French for « travail annotation de données », the same platforms and rules apply. Most projects run in English, but several recruiters also hire for language-specific work, such as Appen/CrowdGen's linguist projects.

What kinds of data labeling jobs exist?

“Data labeling jobs” covers very different activities. Some need little more than patience, while others need a degree or real coding skills. The table below gives an overview. The skills and tools are general descriptions, because every project uses its own interface and guidelines.

What this table shows: as you move down from image labeling to ranking and prompt writing, the work becomes more about judgment and writing than speed. This is also where the highest published rates appear, such as DataAnnotation's coding and STEM ranges.

Illustrative example: Priya, a chemistry graduate in Pune, will probably find STEM evaluation projects a better fit than image boxes. Her advantage is spotting a wrong formula in an AI answer, not clicking quickly. Kunle, a careful reader in Ibadan with no degree, may start with text classification and build up a track record of accurate work. Neither is promised any work, but each knows where to look first.

Who hires for remote data annotation jobs?

Most remote data annotation jobs go through a few specialized companies and their crowd communities. Here is what each one publishes. Where a cell says “not verified,” the official sources we checked did not give that information, so we don't guess.

What this table shows: very few companies publish a clear hourly rate. Payment methods depend on where you live, and many of them publish assessments, certifications or ID checks as entry steps. Strictly speaking, Prolific is a research-study platform rather than an annotation platform.

The market also moves quickly. TELUS International became TELUS Digital on September 18, 2024, according to its official announcement. Google gave Appen notice on January 20, 2024 that their contract would end, and it ended by March 19, 2024 at the latest. Before you rely on any blog post, check the platform's own site, and that includes this one.

How much do AI training jobs really pay?

The chart below shows selected hourly figures the companies themselves publish. For each DataAnnotation category, the lower and upper ends of the range appear as separate bars.

What this chart shows: specialist work (coding, STEM) is where the high published ceilings are. Generalist work sits much lower. Three caveats matter more than the bars themselves:

• An hourly rate is not an hourly income. Projects start and stop, and qualifying for one does not mean steady hours.
• Country changes everything. TELUS Digital's rater offers are estimated at USD 13.50 an hour in Canada but USD 5 an hour in Mexico for comparable work.
• Some projects pay per task, not per hour. On some Alignerr projects, only accepted tasks are paid. Appen/CrowdGen's high figures (USD 160/h for accounting, 450/h for clinicians) are caps for experts, not a general rate.

Is DataAnnotation.tech legit? A data annotation tech review from published facts

This is one of the most searched questions about the field, so here is a data annotation tech review based on what the company states in its official FAQ. We have not tested the platform, and we don't report anonymous earnings claims.

• Published rates: USD 25–50 an hour for general projects, USD 40–150+ for coding and USD 40–125+ for STEM and specialized professions.
• Payment: PayPal, so you need a PayPal account that can receive money in your country.
• Entry: a single “Starter Assessment” of about 1 hour (1–2 hours for specialties). There is no second chance if you fail.
• Countries: the list of eligible countries is not published, so you can't confirm eligibility in advance.

In short, the company is transparent about pay and payment but not about where it operates. The one-shot assessment is the main risk, so treat it like a job interview: prepare first, and take it when you have a quiet hour. Outlier works similarly, with 30–90 minutes of onboarding, an ID check and tests, according to its FAQ.

Does your country matter? What Remotasks taught annotators

Yes, and sometimes brutally so. In March 2024, Remotasks (another Scale AI platform) closed in Kenya and excluded workers from Nigeria and Pakistan. New sign-ups were also stopped in Thailand, Vietnam and Poland, as Rest of World reported. These are 2024 events, not a current list, but the lesson still holds: a platform can leave your country with little warning.

Restrictions exist elsewhere too. Prolific's list of eligible countries includes India, the US, the UK, Canada, France and Belgium, but not Nigeria, Cameroon, Côte d'Ivoire, Senegal, Morocco or Madagascar. Toloka publishes its own geographic restrictions. Mercor's payment country list includes Morocco and Senegal, while Outlier's generalist offer is limited to six English-speaking countries.

Illustrative example: Chinedu in Lagos has a solid science background. Before investing time in an assessment, he should check that the platform accepts Nigerian residents and that he can actually receive the payment method it uses. Our guide to making money online in Nigeria goes through these checks. The practical rule is to never depend on a single platform.

Platforms close, too. Amazon Mechanical Turk, launched on November 2, 2005, closed permanently on September 30, according to the notice on mturk.com. Its FAQ explains how approved HITs are paid. If you used to work there, see our list of alternatives to MTurk.

How to become an AI data annotator: the typical path

No single degree is required everywhere to become an AI data annotator. Recruiters mostly look at your country of residence, your language level and how careful you are, then, for the best-paid projects, a verifiable specialty: coding, maths, science, law or a less common language. The path almost always looks like this.

How do you pass a data annotation assessment?

Assessments are the real gate into data annotation jobs, and some give you only one try. This approach works across platforms.

Passing is only the start. Quality reviews continue on real tasks, and the habits that pass an assessment (reading instructions carefully, checking your work) are the same ones that keep your work accepted. Our guide on passing quality checks the first time covers these habits in more detail.

Is data annotation legit, or is it a scam?

Data annotation itself is legitimate work done for real companies. But its popularity has fueled a wave of fake “task jobs.” The US Federal Trade Commission documented the trend in a December 2024 data spotlight.

What this chart shows: there were about four times as many reports in half a year as in the whole previous year. Reported losses to job scams of all kinds reached USD 223 million in the first half of 2024, against USD 286 million for all of 2023. The pattern is always similar. You are shown fake earnings on a dashboard, then asked to deposit money (often in crypto) to “unlock” tasks or withdraw. The rule is simple: never pay to get paid.

If even one red flag appears, stop. Our checklist for legit work from home without paying upfront covers the other common traps.

Is the data labeling market really growing?

Market researcher Grand View Research estimates the data annotation tools market at about USD 1.7 billion and forecasts USD 10.3 billion by 2033, a growth rate of about 25.4% a year. Treat this as a commercial estimate sold to businesses, not an official statistic.

Two nuances matter for workers. First, the figure measures software tools, not wages paid to annotators. Second, growth at the level of the industry hides constant churn at the level of individual platforms: Remotasks' exits, the end of the Google–Appen contract, TELUS International's rebrand and Toloka joining Mindrift. Demand for human judgment looks durable, but any single job source can disappear.

Where does Microtaches fit in?

To be clear, Microtaches is a French-speaking platform, open worldwide, that offers various micro-tasks: you accept a task, follow the instructions displayed and submit your answer. If the concept is new to you, read our explanation of what micro-tasks are and how you get paid.

The rules are public. Sign-up is free and reviewed by the team, so approval is not instant, and a completed profile earns a 10 Ops bonus. Each task has a 60-minute limit, and you can have up to 3 tasks in progress. You can abandon a task without penalty during the first 10 minutes, but a task that expires without an answer costs 20 Ops. Tasks are reviewed before you are credited. Work availability varies by day, country and profile, and no income is guaranteed.

Earnings are in Ops. 1 Ops is worth €0.0042 when withdrawn and €0.0082 in the shop (gift cards, products), so the shop values your Ops about 1.95 times better. Our guide to converting Ops into euros and gift cards details both options. Before your first withdrawal or shop order, you complete an identity check, available from a balance of 1,000 Ops and processed within up to 7 business days. You will need an ID document and proof of address, and the name must match the receiving account. Our guide explains how the KYC process works.

Withdrawals have a minimum of 5,000 Ops (€21) per request, and 10,000 Ops stay in reserve. Your balance must therefore reach at least 15,000 Ops for a first bank transfer. A bank transfer outside Europe costs a fixed 800 Ops, and payment arrives within up to 3 weeks after your request is approved. Microtaches also applies an internal cap of €2,500 per year, which is a company rule, not a law, as explained in our article on the €2,500 annual cap. The community has more than 44,000 approved workers, over two-thirds of them in Africa, with more than 500 in India and about 350 in Nigeria. Microtaches does not offer M-Pesa, Airtel Money or Payoneer; the available methods are listed below.

Withdrawal methods on Microtaches

Your Ops can be withdrawn to a bank account, a mobile money account or an online wallet. Bank transfer is available from the usual minimum; MVola, Orange Money, Wave, MTN MoMo, PayPal, Skrill and Neteller are available from 25,000 Ops; Revolut and Crypto are coming soon.

Are data annotation jobs worth it? Our verdict

Data annotation jobs are worth trying if you approach them as a selective, unstable source of income rather than a career guarantee. Here is how the trade-offs look.

Sources

• DataAnnotation FAQ: published hourly ranges, PayPal payment and the one-shot Starter Assessment.
• Outlier FAQ: onboarding, identity verification and weekly payment methods.
• CrowdGen (Appen) payments: payouts via Dots, with methods by country.
• Rest of World on Remotasks: the 2024 country closures and exclusions.
• Prolific, eligible countries: the official list of participant countries.
• FTC Data Spotlight: task-scam report numbers and reported job-scam losses.
• Grand View Research: commercial estimate of the annotation tools market.
• Google Search Quality Rater Guidelines: what search raters evaluate and how.

How do I become an AI data annotator with no experience?

First check that your country and language are accepted, then take the entry assessment of one or more platforms. No single degree is required everywhere: accuracy and following guidelines matter most. Meanwhile, simple micro-tasks let you build experience without paying anything.

Are data annotation jobs legit?

Yes, the work itself is legitimate and done for established companies such as Appen/CrowdGen, TELUS Digital, Outlier or OneForma. Fake “task jobs” that ask you to deposit money are a growing scam, with about 20,000 reports to the FTC in the first half of 2024. A real platform never asks you to pay to get paid.

How much do data annotation jobs pay?

It varies widely by skill and country. DataAnnotation publishes USD 25–50 an hour for general projects and up to USD 150+ for coding, while one Outlier generalist offer pays up to USD 18 an hour. These figures are published by the companies and are not guaranteed, and work comes in projects rather than steady hours.

Is DataAnnotation.tech legit?

It publishes its rates, payment method (PayPal) and entry process in an official FAQ. Its Starter Assessment takes about 1 hour and gives no second chance. The list of eligible countries is not published, so you cannot confirm eligibility in advance.

Can I do data annotation jobs remote from Nigeria or India?

Sometimes, depending on the platform. Prolific lists India but not Nigeria, and Remotasks excluded Nigerian workers in March 2024. Always check the official eligibility page and make sure you can receive the payment method in your name.

Do I need a degree for data labeling jobs?

Not always. General labeling and classification mainly require attention and consistency, while the highest published rates (coding, STEM, clinical or accounting work) go to people with real expertise. Either way, you will usually have to pass an assessment.

How does Outlier pay annotators?

Weekly. According to its FAQ and offers, payments go through ACH, PayPal or AirTM. Onboarding takes 30 to 90 minutes and includes identity verification and tests.

Should I pay a fee to start AI training jobs?

No. Legitimate platforms do not charge to join, unlock tasks or withdraw your earnings. A request for a deposit, often in crypto, is the core mechanism of the task scams described by the FTC.

Is Amazon Mechanical Turk still open?

No. MTurk closed permanently on September 30, according to the official notice on mturk.com, and its FAQ explains how approved HITs are paid. Alternatives include Prolific, CloudResearch Connect, Clickworker and Microworkers.

Is data annotation the same as RLHF?

No. RLHF (reinforcement learning from human feedback) uses human rankings and corrections of AI answers to fine-tune models, and some annotation work feeds it. Most annotation, such as image labeling, classification and transcription, is not RLHF.

Does Microtaches offer data annotation work?

Microtaches offers various micro-tasks that you complete by following the instructions displayed, with no income guaranteed. Bank transfer withdrawals start from the 5,000 Ops minimum (balance at least 15,000 Ops for the first one), and MVola, Orange Money, Wave, MTN MoMo, PayPal, Skrill and Neteller are available from 25,000 Ops, with payout within up to 3 weeks.