How to become an AI trainer: the job titles to search instead

The title is rarely on the listing

If you want to know how to become an AI trainer, start with this: almost nobody advertises the job under that name. Of the 761 listings live on this site on 12 October 2026, four have "trainer" in the title, all at micro1, and only two of those say "AI Trainer". The work is posted as expert and specialist roles, as nursing or legal jobs, and as evaluator or transcriber roles. A search for "AI trainer jobs" finds a handful of them. A search for your own profession finds far more.

What an AI trainer does

An AI trainer supplies the human judgement a model learns from: labelling raw material, rating and correcting model answers, or doing professional work that becomes training and test data. Which of the three a listing means decides the pay and the bar to entry; what AI training work actually is explains each.

The titles the work is posted under

We put every title into one family, checking for trainer, evaluator, annotator and reviewer words before expert or specialist, so "Evaluation Expert" counts as evaluation. A title that matches none is a plain job title. The median floor is the median of the lowest advertised rate among listings that pay by the hour (723 of the 761).

Count "expert" or "specialist" anywhere in a title, including the evaluation and QA titles, and the total is 367: Mercor 185, micro1 160, Invisible 22. Nearly half the board is titled as expert work, while the literal trainer roles sit second from the bottom on pay. micro1's Senior AI Trainer advertises $14–36 an hour for generalist chat and search evaluation.

The name does turn up elsewhere. Outlier's homepage describes what you do "as an AI trainer", but Outlier has no open roles listed here, so you mostly meet the term on sign-up pages rather than in job titles.

Which search fits your background

If you have no degree or specialism, search "annotator", "transcription", "evaluator" and "data contributor", and filter by level. The beginners guide goes platform by platform. General & Data Collection holds 72 listings and AI Safety & Evaluation 64.

If you have a profession, search your job title or open your field's page. A nurse should search "registered nurse" or "RN", not "AI trainer": four listings here name nurses, at $55–120 an hour, and none uses the word trainer. Mercor's RN Annotators and micro1's Registered Nurse are both titled for the profession. Lawyers have the most title matches, with 42 titles saying attorney, lawyer, counsel or legal; physicians have 16 and teachers 3. By field, counting a listing once in each field it covers:

• Languages & Linguistics: 120 listings, median floor $20
• Business & Finance: 114, median floor $70
• Engineering: 111, median floor $50
• Health & Medicine: 94, median floor $100
• Software & IT: 94, median floor $57.50
• Law, Policy & Security: 75, median floor $90
• Science & Math: 67, median floor $70

If you speak a second language, 185 listings require one other than English. The largest groups are German (17), Spanish (16), Japanese, French and Portuguese (15 each) and Korean (12). Median floors run from $20 for Portuguese to $50 for Spanish; the bilingual guide explains the spread.

An AI trainer job description, from real listings

The skills on our listings show what platforms screen for. The most common is "attention to detail" (84 listings), then Python (59), rubric writing (57), red teaming (56) and technical writing (43). Grouped loosely, 223 listings name some kind of writing, 220 some kind of evaluation and 138 annotation or labelling.

Three listings show the range:

• Senior AI Trainer, micro1: test chat and search tools under repeatable conditions, grade and rank responses against guidelines, write prompts that push models into edge cases, and recheck ratings where score and rubric disagree. No degree or minimum years are stated.
• AI Agent Evaluator, micro1, $25–45: run two AI agents through the same real errand on your own phone, screen record it, score each run and write feedback. It asks for US residence, a bachelor's degree and two to three years of work experience.
• Registered Nurse, micro1, $60–70: annotate inpatient records and judge AI output for clinical accuracy and safety. It asks for US RN practice and recent bedside experience, and no AI experience.

Seniority leans high. We tag 326 listings senior, 225 mid-level, 134 junior and 76 entry, reading each listing's requirements.

AI trainer salary: what the listings pay

Very little of this is salaried: 723 listings pay by the hour, 33 per task and 5 by the year. Across the hourly ones the median floor is $50 and the median ceiling $90. By level, the median floor is $80 for senior, $50 for mid-level, $20 for junior and $30 for entry. Entry beats junior because 19 of Mercor's entry roles are voice-cloning and ChatGPT-user projects at $50 or more, while junior holds 21 transcription roles and 14 of Invisible's $17–21 evaluation roles.

These are posted rates, not earnings. Several micro1 listings pay per accepted task, so the band is what a fast contributor reaches; see how you actually get paid. No listing guarantees hours either, and being accepted then waiting for work is common enough to have its own guide.

First steps

1. Find your family in the table and search by your profession or the task name, not "AI trainer". Save the field and language pages that match you.
2. Check eligibility and the assessment before anything else. How to get started puts the steps in the order that wastes least time.
3. Rewrite your CV for a matching system, naming your subfield, volumes and written outputs; the AI training resume has examples.
4. With no specialism, start with evaluation or annotation. AI response evaluation jobs explained and what is data annotation show what a task involves.
5. Apply to more than one platform, because matching is slow; see applying to several platforms at once.

How we counted

• Families come from words in each title. Every listing on this board is AI training work, so the plain job title family is professional work done for training data.
• Pay medians use the lower end of each hourly range and leave out per-task and yearly listings. Field counts overlap.
• We could not establish how often each title leads to paid work; no platform publishes acceptance rates.

Sources

• The 761 listings live on this site on 12 October 2026, for every count, rate and quotation
• Outlier's homepage (outlier.ai), read 12 October 2026, for its use of "AI trainer"

We are not affiliated with any of these platforms and we do not process applications.