An RLHF job is paid work rating, ranking and rewriting AI model answers so the model can be trained on human judgment. RLHF stands for reinforcement learning from human feedback. Listings rarely use the term. They say AI trainer, AI tutor, evaluator or annotator. On 15 November 2026 public listings paid generalists up to $30 or $50 an hour and subject experts far more, for example up to $180 an hour for physicians. The work is remote contract work, limited by country on some programs, and it starts and stops with each project.
- RLHF is a training method. The job is giving the feedback: comparing answers, scoring them and writing better ones.
- Published pay on 15 November 2026: ceilings of $30 to $50 an hour for generalists, $45 to $180 for the expert listings in the table.
- Search for the job titles, not the acronym: AI trainer, AI tutor, evaluator, rater, annotator.
- No program promises hours. Projects pause and close with the lab's need for feedback.
What is an RLHF job?
It is paid work judging and improving what an AI model writes, so that the model can learn from those judgments. You are the human in reinforcement learning from human feedback.
The method comes from research. A 2017 paper trained systems on "human preferences between pairs of trajectory segments": show a person two attempts, ask which is better, and learn from the choice. OpenAI's 2022 paper on instruction-following models applied it to language: "We then collect a dataset of rankings of model outputs, which we use to further fine-tune this supervised model using reinforcement learning from human feedback."
Those rankings are the job. The industry uses the word loosely for all human feedback work. A trade publication described Welo Data as supporting "AI development via data generation, RLHF, and model evaluation", and Outlier describes itself as a platform that "connects experts with leading AI companies to provide human feedback that improves large language models (LLMs)."
What does an RLHF task look like?
Most tasks show you a prompt and one or two model answers and ask for a judgment with a reason. A task takes from a few minutes to an hour or more, depending on the field.
The table lists the five task types you will meet, with the words public sources use for them. On expert projects the same person often does all five on one item.
| Task | What you do | How a public source words it |
|---|---|---|
| Compare two answers | Read both, pick the better one and say why in a sentence or two | micro1's guide: "Tagging content, ranking responses, A/B comparisons" |
| Rate one answer | Score it for accuracy, completeness and tone against written guidelines | Outlier's home page: "Rate and Rank Answers" |
| Write the ideal answer | Answer the prompt the way the model should have | The 2022 paper calls these "labeler demonstrations of the desired model behavior" |
| Write a hard prompt | Ask something in your field the model is likely to get wrong | Outlier's home page: "Write a Challenging Prompt" |
| Write the rubric | Set out what a correct answer must contain | Outlier's home page: "Create Grading Rubrics" |
The reason matters more than the choice. Guidelines usually ask you to name the specific error: a wrong fact, a skipped step, an unsafe suggestion or an answer to a different question. Writing test items that models fail is a specialty of its own, covered in AI benchmark writer jobs.
How much do RLHF jobs pay?
Generalist listings we read on 15 November 2026 published ceilings of $30 and $50 an hour, and one program's own guide puts general annotation at $15 to $40. Expert listings in the same week ran from $45 to $180 an hour depending on the field.
These are contract rates, not salaries. There is no annual figure, no benefits and no promised hours, and you are paid for time logged or tasks approved. Outlier's FAQ says it "pays for completed work, either hourly or per task." "Up to" is a ceiling: the rate on a given project can be well below it.
| Level | Program and listing | Published pay |
|---|---|---|
| General annotation | micro1's guide for experts | Typical range $15-$40 an hour |
| Generalist | Handshake AI, generalist AI trainer | Up to $30 an hour, depending on the project |
| Generalist | DataAnnotation, Generalist | Up to $50 an hour, by task difficulty |
| Per-task rater | TELUS Digital, Evaluator (United States) | 0.2333 USD per task, estimated at 14 USD an hour |
| Math, master's or PhD | xAI, AI Tutor, Math | $45 to $75 an hour for US-based candidates |
| Licensed lawyers | Handshake AI, Lawyers | Up to $110 an hour |
| Physicians | Handshake AI, Physicians | Up to $180 an hour |
| Verified specialists | micro1's guide for experts | Typical range $100-$200+ an hour |
More fields and programs are in AI trainer pay rates compared. What a license or degree changes is in do credentials raise AI training pay?.
Can beginners get RLHF jobs with no experience?
Yes for generalist work, where programs test your reading and writing instead of asking for prior AI work. Expert listings ask for proof of the profession, not of AI experience.
DataAnnotation's generalist listing says "No degree required" and asks for "Strong reading comprehension and attention to detail." Handshake AI's generalist listing says "We'll provide a brief training, and you'll take a single assessment before starting." At the expert end, a micro1 listing for attorneys says "no prior AI experience required".
The test is where most people stop. DataAnnotation says "We review submissions for thoroughness and accuracy rather than speed, so taking time to craft quality answers improves your chances." No real program charges you to start. The FTC says "Honest employers, including the federal government, will never ask you to pay to get a job." See the DataAnnotation assessment and AI training jobs for students.
Are RLHF jobs remote, and can you do them outside the US?
They are remote, and working from home is the norm. Country rules differ by program and sometimes by project.
Outlier publishes a list headed "You may work on Outlier from the following countries" and adds that customers may "require contributors from specific states, regions, or countries". Mercor says it "does not engage contractors located in, countries outside of those listed in our Supported Regions Policy".
In the UK the work exists at both ends. A TELUS Digital listing for UK English read on the same day paid per task: "The rate of pay is 0.2200 GBP per task, with an estimated earning potential of 13.20 GBP per hour." An xAI math listing gave its location as "Remote (United States, United Kingdom, Canada, Australia, and India)". See AI training jobs in the UK and every country. Origin takes inquiries from every country.
Why does RLHF work stop and start?
Because each project is one lab's order for a fixed amount of feedback, and it ends when the order is filled. No program promises hours.
Programs put this in writing. Handshake AI: "Projects may be paused or closed with little notice based on the needs of our partner labs." Mindrift: "the platform cannot guarantee that any specific task will be continuously available to every user". Mercor: "We can not guarantee a minimum number of hours of work per week, or when a project may come to a close." micro1's status page has a state that reads "Hiring is temporarily paused for this role."
Plan for it. Treat RLHF income as variable, keep your main work, and expect quiet weeks after a project closes. No tasks available? explains what each program says about empty queues.
Is RLHF the same as AI training or data annotation?
RLHF is one method inside AI training. Data annotation labels raw data, RLHF ranks and corrects what a model wrote, and evaluation measures the result.
In practice the job titles overlap and one contract can cover all three. If you want the neighbouring kinds of work, see AI evaluation jobs, remote data annotation jobs and AI tutor jobs.
Tell us your field
Origin Data Partners takes inquiries from people in every country, with or without a specialist background. Tell us what you know well, where you live and how many hours a week you have. If you hold a degree or a license, say which. We reply by email with the feedback work that is open to you.
Tell us your field, your country and the hours you have, and we'll email you what is open. It's free.
Origin Data Partners is independent of the programs on this page. A program may pay Origin when Origin introduces someone who joins it, which never changes what you are paid (how Origin is paid).
Common questions
What does RLHF mean?
Reinforcement learning from human feedback. People compare and rank a model's answers, and the model is trained further on those rankings. The term comes from AI research papers published in 2017 and 2022.
How much do RLHF jobs pay per hour?
On 15 November 2026 generalist listings published ceilings of $30 and $50 an hour. Expert listings ran higher: $45 to $75 for math, up to $110 for lawyers and up to $180 for physicians.
Are RLHF jobs remote work from home?
Yes. The listings we read are remote contract roles done on your own computer. Some programs limit which countries, and some projects which states, you can work from.
Can you get RLHF jobs with no experience?
Yes, at the generalist level. Programs screen with a writing and reasoning assessment, not a work history. One generalist listing says no degree is required.
Are there RLHF jobs in the UK?
Yes. On 15 November 2026 a TELUS Digital listing for UK English paid 0.2200 GBP per task, estimated at 13.20 GBP an hour, and an xAI math listing named the United Kingdom among its remote locations.
Do RLHF jobs pay a salary?
Almost never. The listings are hourly or per-task contracts with no promised hours and no benefits. Full-time salaried RLHF roles exist inside AI labs, but those are research and engineering jobs.
Is RLHF the same as AI training?
It is one part of it. AI training also covers labeling raw data, writing test questions and red teaming. Job listings usually say AI trainer, tutor or evaluator and rarely use the acronym.
