AI training jobs in 2026: pay, skills and who does the work

Q: How much do AI training jobs pay?

Of listings that state an hourly rate, 64% offer $35 an hour or more, 21% offer $20 to $35 and 15% offer under $20. Pay follows the kind of work: 88% of Domain expertise listings offer $35 or more, against 55% for Data annotation. Only 49% of listings state a rate at all.

Q: Are AI training jobs remote?

Of listings that state where the work is done, 27% are remote and open to people in any country, 61% are remote but limited to one country or region, and 13% are hybrid or on-site.

Q: How many hours a week is AI training work?

Of listings that state weekly hours, 62% ask for ten hours a week or less, 34% for ten to thirty hours and 5% for thirty or more.

Q: How do you get into AI red teaming jobs?

Red teaming listings ask most for Explaining findings (75%), Adversarial testing (74%) and Professional judgment (68%). Of its listings that state a level, 2% are entry level and 84% are senior or lead, and 8% of those that state a location are open to any country.

Explore the work,
read against the market.

Compare the abilities named in different kinds of work. These are market-level patterns; your personal experience map appears after you upload your CV.

Compare two kinds
of AI training work.

Choose two categories to compare what their listings ask for. Shared abilities describe the categories, not skills assessed in your CV.

CompareWith

The report / January–September 2026

What AI training jobs
really look like.

We read job listings and expert posts to see what AI trainer, data annotation and red teaming work pays, what it asks for and who does it.

Start reading

Key findings

Model evaluation is the biggest kind of AI training work, at 40% of listings.
64% of listings that state an hourly rate offer $35 or more, but only 49% of listings state a rate.
17% of listings that state a level are entry level. Data annotation is the way in, at 47%.
27% of listings that state a location are open to people in any country; 61% are remote but limited to one country or region.
62% of listings that state weekly hours ask for ten hours a week or less.
Getting in is what experts talk about most: 40% of their posts are about applying and qualifying.

The medium The work The skills The pay The people The conversation

What data do
AI trainers work on?

Most of the work is still words. Every listing, sorted by the kind of data it asks people to create or review.

Two in three jobs are about text.

They ask people to write, rate or check written language: answers, prompts, rubrics, chatbot replies. Audio, code and agent tasks make up most of the rest.

One square is one percent of listings.

Each kind of work has its own medium.

Annotation is where audio, images and video live. Building agent environments is almost entirely agent work, while evaluation and red teaming stay with text.

What kinds of
AI training jobs are there?

Six kinds of work, with different doors in. What people are hired to do, at what level, and from where.

Rating AI is the biggest job.

40% of listings are for evaluating model answers. That is more than annotation, the work most people picture when they hear “AI data”.

Annotation is the way in.

It is the only kind of work where entry-level roles are common. Red teaming and expert work hire almost only senior people.

Remote rarely means anywhere.

Most remote listings are limited to one country or region. Red teaming and agent work are the least open to people abroad, and agent work is often on-site.

Most of this work is a few hours a week.

When a listing says how many hours it wants, red teaming (88%) and evaluation (68%) mostly ask for ten a week or less. Writing and training is the one kind that usually wants more: 79% of its listings ask for ten to thirty hours.

Listings rarely state how long a project lasts; weekly hours are the commitment they do state. Kinds of work with fewer than 20 such listings are not shown.

Which skills get you
AI training work?

Rubrics and languages open the most doors. The skills listings name, and the skills that tend to come as a pair.

What your background
opens up.

Pick your field to see which kinds of work want it and what those listings come with.

Your field

What do
AI training jobs pay?

Expertise pays, and many listings won’t say how much. Hourly rates in US dollars, from listings and from experts’ own reports.

From a few dollars to over $60 an hour.

A third of the rates people mention are above $60 an hour. About a quarter are under $20.

Advertised and reported rates, not verified earnings.

Expert work sits at the top.

Domain experts and red teamers are mostly paid in the upper bands. Annotation is the only work with a real share of low rates.

Many listings keep the rate quiet.

Fewer than half of red-teaming and expert listings name a number. Writing and training listings are the most open about pay.

Almost all of it is paid by the hour.

Two exceptions: 40% of domain-expert listings pay per task, the consulting model, and 50% of RL and agent listings come with an annual salary, the only kind of work here that mostly sits outside the gig economy.

Kinds of work with fewer than 20 listings stating a pay type are not shown.

Who are the people
behind AI training?

People spread across 5 continents. 17% of the people who say where they are write from India and 17% from the United States; no other country passes 12%. Together, Americas and Asia hold 61%.

Where people say they are

Only people who say where they live or come from are counted. The forums are in English, so English-speaking countries are over-represented.

Who they are

41%
are students.

Of the people who say what else they do, most are studying, between jobs or holding another job. Only 2% do this work full time. Those who name a background come mostly from software, science and finance; half of those who mention a degree hold a bachelor’s.

What else they do

Their field

Their education

What do AI trainers
talk about?

Getting in is the hardest part. What experts post about in their own communities.

Two in five posts are about getting in.

Applications and qualification tests come up more than anything else, and even these posts lean negative.

Problems are as common as getting in.

Two in five posts are about something going wrong: work drying up, payment trouble, accounts removed. About one in eight is good news.

And the work itself

41% of projects
end within a month.

From experts’ own accounts of how long a project ran. At the other end, 22% ran for more than six months.

AI training jobs,
question by question.

The questions people ask most about this work, answered from the same data as the charts.

AI training jobs by kind of work
Kind of work	Share of listings	Entry level	Open to any country	$35 an hour or more
Model evaluationRating, ranking or reviewing an AI’s answers, and writing the rubrics that say what a good answer is.	40%	15%	27%	58%
Data annotationLabelling raw material (text, images, audio, video) so a model can learn from it.	18%	47%	37%	55%
AI training & writingWriting the prompts, model answers and examples an AI is trained on, or tutoring it as a trainer.	15%	15%	30%	66%
Red teamingTrying to make an AI misbehave or break its rules, so its makers can fix it before release.	12%	2%	8%	78%
Domain expertiseBringing a profession (medicine, law, finance, science, a language) to write or check answers in that field.	10%	5%	39%	88%
RL & agent tasksBuilding the tasks and environments where an AI agent practises acting: browsers, tools, software.	4%	5%	9%	Too few rates

Entry level and location are shares of the listings that state them; pay is a share of the listings that state an hourly rate. The highest in each column is highlighted.

What is an AI trainer?

AI trainer is the common name for people paid to produce and check the data AI models learn from. Listings split the work into six kinds: Model evaluation (40%), Data annotation (18%), AI training & writing (15%), Red teaming (12%), Domain expertise (10%) and RL & agent tasks (4%).

What is data annotation?

Labelling raw material (text, images, audio, video) so a model can learn from it. It is 18% of listings, and 47% of its listings that state a level are entry level, against 17% across all listings.

How much do AI training jobs pay?

Of listings that state an hourly rate, 64% offer $35 an hour or more, 21% offer $20 to $35 and 15% offer under $20. Pay follows the kind of work: 88% of Domain expertise listings offer $35 or more, against 55% for Data annotation. Only 49% of listings state a rate at all.

Can you get an AI training job with no experience?

17% of listings that state a level are entry level. Data annotation is the way in, with 47% of its listings at entry level. Red teaming is the hardest to start in, at 2%.

Are AI training jobs remote?

Of listings that state where the work is done, 27% are remote and open to people in any country, 61% are remote but limited to one country or region, and 13% are hybrid or on-site.

How many hours a week is AI training work?

Of listings that state weekly hours, 62% ask for ten hours a week or less, 34% for ten to thirty hours and 5% for thirty or more.

What skills do AI training jobs ask for?

What listings most often ask people to be able to do: Explaining findings (61%), Evaluating AI answers (56%), Professional judgment (48%) and Finding errors (44%). Most listings ask for several abilities at once.

What is AI red teaming?

Trying to make an AI misbehave or break its rules, so its makers can fix it before release. It is 12% of listings and mostly experienced work: 84% of its listings that state a level are senior or lead.

What does an AI evaluator do?

Rating, ranking or reviewing an AI’s answers, and writing the rubrics that say what a good answer is. It is the biggest kind of AI training work, at 40% of listings. What evaluator listings ask for most: Evaluating AI answers (91%), Explaining findings (81%) and Finding errors (68%).

How do you get into AI red teaming jobs?

Red teaming listings ask most for Explaining findings (75%), Adversarial testing (74%) and Professional judgment (68%). Of its listings that state a level, 2% are entry level and 84% are senior or lead, and 8% of those that state a location are open to any country.

What is an AI prompt writer?

Write and refine the prompts or instructions that steer an AI model. 18% of listings ask for it, most often in AI training & writing (51% of its listings).

Are there AI training jobs for doctors, nurses and lawyers?

Yes. Medicine is named in 10% of listings, most often in Domain expertise (21% of its listings); 61% of those that state a rate offer $35 an hour or more. Law is named in 3% of listings, most often in Domain expertise (9% of its listings); too few of them state a rate to compare pay.

Are AI training jobs legit?

Experts report both good and bad. In their posts about platforms, 12% are about new work or good pay, 11% report a payment problem and 3% ask or warn whether a platform is legitimate. The largest topic is applying and qualifying, at 40%.

Who does AI training work?

Of the people who say what else they do, 41% are students, 26% have another job and 27% are looking for work. The countries people name most are India, United States and United Kingdom.

What the words mean

Kinds of work

Model evaluation: Rating, ranking or reviewing an AI’s answers, and writing the rubrics that say what a good answer is.
Data annotation: Labelling raw material (text, images, audio, video) so a model can learn from it.
AI training & writing: Writing the prompts, model answers and examples an AI is trained on, or tutoring it as a trainer.
Domain expertise: Bringing a profession (medicine, law, finance, science, a language) to write or check answers in that field.
Red teaming: Trying to make an AI misbehave or break its rules, so its makers can fix it before release.
RL & agent tasks: Building the tasks and environments where an AI agent practises acting: browsers, tools, software.

Data types

Text: Written language: prompts, answers, conversations, rubrics.
Audio & speech: Voice, speech or other audio, recorded or transcribed.
Code: Source code and programming tasks.
Agent & computer use: Recordings of an AI operating a computer: clicks, tool calls, multi-step tasks, RL environments.
Images: Images and other visual material.
Video: Moving images, recorded or reviewed.
Physical world: Robotics, sensor, driving and other physical-world data.
Mixed: Several data types with none dominant.
Not specified: The listing does not say.

Abilities

Labelling data: Label or tag raw data, or extract information from documents, as annotation work.
Following a rubric: Follow a given rubric, guideline, policy or taxonomy when rating or labelling.
Professional judgment: Apply a professional or academic background in a specific field, such as medicine, law, finance, engineering or a science.
Judging language: Judge or produce language quality: fluency, translation, localisation or cultural fit.
Auditing data quality: Quality-check data that other people have labelled or produced.
Explaining findings: Explain a judgment, feedback or findings in clear, structured writing or speech.
Coordinating work: Organise tasks, deadlines and collaboration within a team.
Writing guidelines: Design the labelling schemas, guidelines, rubrics or reference examples that others follow.
Designing test tasks: Create tasks, test cases, scenarios or benchmarks that show how well an AI system performs.
Finding errors: Identify and document specific errors, bugs or failure cases, beyond giving an overall rating.
Evaluating AI answers: Compare, rank or assess an AI model’s outputs for quality, accuracy or following instructions.
Audio & speech work: Transcribe, record or assess audio or speech.
Reviewing visuals: Evaluate images, video or design output for quality or correctness.
Adversarial testing: Probe an AI system with adversarial inputs to make it fail, and document what happened.
Solving technical problems: Work out correct solutions to mathematical, scientific or engineering problems.
Checking facts: Check claims, calculations or citations against reliable sources.
Writing & reviewing code: Write, test, debug or review software code.
Writing prompts: Write and refine the prompts or instructions that steer an AI model.

Skills

Programming: The listing names programming or a programming language.
Writing & editing: The listing names writing, editing or proofreading.
Languages: The listing names a language skill, translation or localisation.
Math & statistics: The listing names mathematics or statistics.
Medicine: The listing names medical or clinical expertise.
Finance & economics: The listing names finance, accounting or economics.
Law: The listing names legal expertise.
Science: The listing names physics, chemistry or biology.
Evaluation & rubrics: The listing names rubrics, fact-checking or evaluating model output.
Security: The listing names red teaming, adversarial testing or security.
Audio, image & video: The listing names audio, image or video work.
Agents & robotics: The listing names agents, RL environments, computer use or robotics.

Levels

Entry: No experience required, or an internship.
Mid-level: Some experience expected; works alone rather than leading.
Senior: Senior or expert, several years of experience.
Lead: Leads a team or runs a programme.

Where the work is done

Global remote: Fully remote and open to people in any country.
Regional remote: Remote, but only for one country or region.
Hybrid: Partly on-site.
On-site: At a fixed location.

Weekly hours

10 hours a week or less: The listing asks for ten hours a week or less.
10–30 hours a week: The listing asks for between ten and thirty hours a week.
30 hours a week or more: The listing asks for thirty hours a week or more.

How it is paid

By the hour: Paid per hour worked.
Per task: Paid per task or unit completed.
Monthly: A monthly wage or stipend.
Annual salary: An annual salary, as in employment.

Project length

Under a week: A project of a few days up to a week.
1–4 weeks: A project of one to four weeks.
1–3 months: One to three months.
3–6 months: Three to six months.
6+ months: More than six months.
Ongoing: Still running, no end stated.

What experts post about

Applying & qualifying: Applying, assessments, qualifications, interviews or waiting for approval.
New work available: New tasks, projects or invitations; the poster is getting work.
Work drying up: An empty queue, no tasks or hours cut while the account stays active.
Payment problems: Unpaid, late or reduced pay, or a payment dispute.
How-to questions: A how-to question with no signal about the platform.
Other complaints: A negative experience not covered by the other topics, such as unresponsive support.
Removed from a platform: The account was deactivated, banned or removed from the platform.
Is it legit?: Asking or warning whether a platform is a scam, legitimate or safe.
Good pay: Good pay, a raise, or fast and reliable payment.
Good experiences: A positive experience not covered by the other topics.

What else they do

Students: Says they are a student.
Looking for work: Says they are unemployed or between jobs.
Have another job: Has another job; this work is on the side.
Caregivers: A stay-at-home parent or caregiver.
Do this full time: Does this work as their main occupation.
Retired: Says they are retired.

Their field

Software & IT: A background in software, IT or data.
Science & engineering: A background in science, engineering or mathematics.
Finance & business: A background in finance, accounting or business.
Medicine & health: A background in medicine, nursing or health care.
Arts & design: A background in arts, design or entertainment.
Other fields: Another field the person names.
Writing & media: A background in writing, journalism, editing or media.
Law: A background in law.
Languages: A background in translation, interpreting or languages.
Education: A background in teaching or education.

Their education

Bachelor’s: A bachelor’s degree, or studying for one.
Master’s: A master’s degree, or studying for one.
Doctorate: A PhD or other doctorate, or studying for one.
Professional degree: A professional degree such as MD, JD or a nursing licence.
Secondary school: High school or secondary school.

A clearer picture of what you bring.

See the full map.Make it yours.

Explore the work, read against the market.

Compare two kinds of AI training work.

What AI training jobs really look like.

Key findings

What data do AI trainers work on?

Two in three jobs are about text.

Each kind of work has its own medium.

What kinds of AI training jobs are there?

Rating AI is the biggest job.

Annotation is the way in.

Remote rarely means anywhere.

Most of this work is a few hours a week.

Which skills get you AI training work?

What your background opens up.

What do AI training jobs pay?

From a few dollars to over $60 an hour.

Expert work sits at the top.

Many listings keep the rate quiet.

Almost all of it is paid by the hour.

Who are the people behind AI training?

41% are students.

What else they do

Their field

Their education

What do AI trainers talk about?

Two in five posts are about getting in.

Problems are as common as getting in.

41% of projects end within a month.

AI training jobs, question by question.

What is an AI trainer?

What is data annotation?

How much do AI training jobs pay?

Can you get an AI training job with no experience?

Are AI training jobs remote?

How many hours a week is AI training work?

What skills do AI training jobs ask for?

What is AI red teaming?

What does an AI evaluator do?

How do you get into AI red teaming jobs?

What is an AI prompt writer?

Are there AI training jobs for doctors, nurses and lawyers?

Are AI training jobs legit?

Who does AI training work?

Kinds of work

Data types

Abilities

Skills

Levels

Where the work is done

Weekly hours

How it is paid

Project length

What experts post about

What else they do

Their field

Their education

See the full map.
Make it yours.

Explore the work,
read against the market.

Compare two kinds
of AI training work.

What AI training jobs
really look like.

What data do
AI trainers work on?

What kinds of
AI training jobs are there?

Which skills get you
AI training work?

What your background
opens up.

What do
AI training jobs pay?

Who are the people
behind AI training?

41%
are students.

What do AI trainers
talk about?

41% of projects
end within a month.

AI training jobs,
question by question.