Rater

See also: Machine learning terms

A rater is a person (or, increasingly, a model) who assigns labels, scores, or judgments to data items so those items can be used to train, evaluate, or align a machine learning system. The rater is the human (or proxy) who decides what counts as the correct answer, and almost every supervised model in production today rests on a pile of rater decisions made at some earlier stage. Without raters there is no labeled data, no benchmark, no RLHF reward model, and no way to tell whether a system actually behaves the way its developers claim. The work is now a multibillion-dollar industry: the global market for AI data labeling was worth roughly $1.9 billion in 2025, and platforms such as Scale AI and Surge AI field hundreds of thousands of contract raters between them. ^[21][11]

The word is used loosely. The same role goes by different names depending on the field, the task, and the company doing the hiring. In computer vision the person drawing bounding boxes is usually called a labeler. In NLP they are usually called an annotator. In information retrieval they are an assessor. In LLM evaluation they are a judge. Search teams call them quality raters. The underlying job is the same: look at an item, apply a guideline, and produce a structured response.

What is a rater called in different fields?

Different communities settled on different vocabulary, often before the fields started talking to each other.

Scroll sideways for more →

The boundaries blur in practice. A Scale AI worker producing preference rankings for a frontier model might be called a rater, an annotator, or a labeler in different documents on the same day.

What are the main types of raters?

Not every rater is the same kind of person, and the choice of pool drives both quality and cost.

Domain experts. Radiologists labeling chest X-rays, lawyers tagging contract clauses, biologists identifying species in camera trap photos. Expert raters are slow and expensive but produce labels in domains where lay annotators cannot reliably tell what they are looking at. Medical imaging datasets like CheXpert leaned on board-certified radiologists for the test sets specifically because crowd labels were not trustworthy enough.

Trained crowdworkers. Vetted contractors who go through onboarding and qualification tasks before they are allowed to work on production data. Surge AI, iMerit, and Scale AI all rely heavily on trained pools because frontier-lab work involves nuanced rubrics that cannot be picked up in five minutes.

Untrained crowdworkers. Anyone who picks up a task on an open marketplace like Amazon Mechanical Turk. Cheap and fast, but quality varies enormously and many tasks need redundancy and gold-standard checks to be useful.

End users. Implicit raters who never volunteered for the role. Clicks, dwell time, thumbs-up, star ratings, and the choice of which AI response to copy are all signals that get folded back into ranking models and reward models. End-user data is abundant and cheap but noisy and biased toward whatever the product surfaces.

LLMs as raters. Since 2023 it has become routine to use a strong language model as a stand-in for a human judge. The LLM-as-judge approach trades off some quality for orders-of-magnitude lower cost and is now standard in benchmarks like MT-Bench and AlpacaEval. ^[1]

Which platforms do raters work through?

Most rater work today flows through a handful of platforms. Each has a different posture on quality, pay, and the kind of task it is built for.

Scroll sideways for more →

The platform market consolidated sharply in 2025. Meta paid $14.3 billion for a 49% non-voting stake in Scale AI in June 2025, valuing the company at $29 billion and moving founder Alexandr Wang to lead Meta's superintelligence effort. ^[18] Weeks later Scale laid off about 14% of its staff, roughly 200 full-time employees, and ended contracts with about 500 data-labeling contractors, with interim CEO Jason Droege telling staff, "we ramped up our GenAI capacity too quickly over the past year." ^[19] Rival Surge AI, founded by Edwin Chen in 2020 and bootstrapped without venture capital, reported roughly $1.4 billion in revenue in 2025 and opened its first outside fundraise in July 2025, seeking up to $1 billion at a valuation above $15 billion. ^[20]

Which rater programs shaped modern AI?

A handful of rater programs have shaped the systems most people interact with every day.

Google Search Quality Raters. Google has used internal quality raters since around 2005 and published the first complete public version of the Search Quality Rater Guidelines in November 2015. ^[4] The guidelines run to 182 pages as of the September 11, 2025 revision and are the definition of the E-E-A-T framework (Experience, Expertise, Authoritativeness, Trust), which was extended from E-A-T in December 2022. ^[4][5] Google has described the program as involving more than 10,000 raters worldwide, and external estimates place the current number near 16,000. ^[15] The September 2025 revision expanded the Your Money or Your Life category to cover "Government, Civics and Society" and added guidance for judging AI-generated answers such as AI Overviews. ^[16] Quality raters do not directly change rankings; they evaluate sample result pages, and Google uses the aggregated scores to test ranking changes. As Google puts it in its How Search Works documentation, the ratings raters provide "don't directly impact how a page or site appears in Search." ^[17]

ImageNet labelers. Fei-Fei Li and her team built ImageNet between 2007 and 2010 by routing more than 160 million candidate images through Mechanical Turk. Roughly 49,000 workers in 167 countries did the labeling, and by 2012 ImageNet was the largest academic user of Mechanical Turk in the world. The final dataset had about 14 million labeled images across 22,000 categories and is the dataset that the AlexNet result, and most of the modern computer vision boom, was built on. ^[7]

LMArena human raters. Chatbot Arena (now LMArena) collects pairwise preferences from anonymous public visitors who blind-compare two model outputs and pick a winner. The aggregated Elo ratings have become a de facto leaderboard for frontier LLMs and are widely cited by labs in their model release notes.

Wikipedia editors. Not raters in the formal sense, but Wikipedia's volunteer editors produce one of the largest human-curated text corpora in existence and one that nearly every modern LLM trains on. The community's content policies effectively act as a labeling guideline for the rest of the internet.

RLHF labelers at OpenAI, Anthropic, and Google. The reward models that shape ChatGPT, Claude, and Gemini were initially trained on roughly 50,000 human preference rankings each, collected from contracted rater pools. ^[8][9] The exact pool composition is rarely disclosed publicly.

How is rater quality controlled?

Rater output is noisy by default. A few standard techniques are used to push the noise down to a usable level.

Gold-standard questions. Items where the correct answer is already known are sprinkled into the workload. Raters whose accuracy on the golds drops below a threshold get retrained or removed from the pool.

Qualification tasks. A short scored test that workers must pass before they can take paid work on a project. Used aggressively on Mechanical Turk and Toloka because the worker pool is otherwise unfiltered.

Multiple raters per item. The same item is shown to three or five raters and the labels are aggregated by majority vote, by weighted vote, or by an expectation-maximization model that estimates each rater's reliability and the latent true label simultaneously.

Adjudication. When raters disagree past a threshold, a senior rater or domain expert reviews the item and decides the ground truth. Common in medical and legal labeling.

Inter-rater agreement metrics. Quantitative measures of how consistently the pool labels the same items.

Scroll sideways for more →

Under the widely used Landis and Koch scale, a Cohen's kappa between 0.61 and 0.80 is considered substantial agreement, and 0.81 or higher almost perfect. ^[13] Krippendorff's alpha at or above 0.80 is the conventional acceptable threshold for drawing conclusions from coded data. ^[12]

How does rater bias shape models?

The people doing the labeling shape the model that learns from their labels, and that effect is bigger than it looks. Mor Geva, Yoav Goldberg, and Jonathan Berant showed in their 2019 EMNLP paper Are We Modeling the Task or the Annotator? that NLU models pick up on the writing patterns of specific annotators in the training set and fail to generalize to examples produced by annotators they have not seen. Their headline recommendation was that "test set annotators should be disjoint from training set annotators," which most benchmarks were not doing at the time. ^[2]

The broader lesson is that crowdsourced datasets reflect their crowd. If your annotator pool is concentrated in one country, one age range, or one political demographic, your model will inherit those defaults. RLHF preference data is especially exposed to this because the rater is making subjective calls about helpfulness and tone, not labeling something with a verifiable ground truth.

Can an LLM be a rater?

Using a strong language model as a rater for other model outputs has become standard practice since the publication of Lianmin Zheng et al.'s 2023 NeurIPS paper Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena. The paper found that "strong LLM judges like GPT-4 can match both controlled and crowdsourced human preferences well, achieving over 80% agreement, the same level of agreement between humans." ^[1] That result made it cheap to run automated evaluations at scale, and most modern leaderboards (MT-Bench, AlpacaEval, Arena-Hard) lean heavily on judge models.

The approach has known failure modes. LLM judges show position bias (preferring the first answer presented), verbosity bias (preferring longer answers regardless of quality), and self-enhancement bias (judging outputs from their own family more favorably). A related authority bias leads judges to favor answers that cite sources even when the citations are fabricated. ^[22] Mitigations include randomizing answer order, controlling for length, and using multiple judges from different model families. Purpose-built evaluators can narrow the gap with human raters: the open Prometheus 13B judge model reported a Pearson correlation of 0.897 with human scores when supplied with detailed rubrics. ^[23]

How do raters fit into the RLHF workflow?

RLHF is the most rater-intensive part of modern LLM training. The standard pipeline runs roughly like this:

1. A model generates two or more candidate responses to the same prompt.
2. A human rater reads both and picks the preferred one, sometimes with a written justification.
3. Roughly 50,000 of these preference pairs are collected. ^[9]
4. A reward model is trained to predict which response a human would prefer.
5. The base policy is fine-tuned (typically with PPO or DPO) to maximize the reward model's score.

Quality of rater output bottlenecks every step downstream. A reward model trained on inconsistent or biased preferences will steer the policy toward whatever the raters happened to prefer, including superficial qualities like response length or formatting flourishes that look authoritative without being correct.

How much do AI raters get paid?

The rater workforce that powers modern AI is largely contract labor in low-wage countries, and the conditions have repeatedly been the subject of investigative reporting. A January 2023 TIME investigation by Billy Perrigo revealed that OpenAI had contracted with the San Francisco firm Sama to have Kenyan workers label graphic content (including descriptions of child sexual abuse, bestiality, torture, and suicide) so that ChatGPT could be trained to refuse such content. Take-home pay was between roughly $1.32 and $2 per hour. ^[3] Multiple workers reported lasting psychological harm from the material; Sama cancelled the OpenAI contract in February 2022, eight months ahead of schedule. Kenyan content moderators later petitioned their parliament for stronger protections. ^[14]

Scale AI has also faced criticism. A 2024 lawsuit alleged that contractors regularly experienced delayed, reduced, or cancelled payments after completing assignments, and Scale's 240,000-strong contractor pool has been described in court filings as poorly insulated from the psychological costs of frontier AI work. ^[11] The general pattern is that the labeling labor sits at the bottom of the AI value chain and captures very little of the value the labels create.

Where are raters used in AI today?

Frontier AI development as of 2026 leans on raters in a few specific places. Post-training pipelines need preference data for RLHF and direct preference optimization. Safety evaluations need raters to score model behavior against red-team prompts and to grade refusals. Domain-specific deployments (medical, legal, code) need expert raters to build evaluation suites the labs themselves cannot internally produce. Active learning loops use rater attention selectively, queueing only the items the model is least sure about for human review rather than labeling everything.

Two shifts are reshaping rater demand. First, reinforcement learning with verifiable rewards (RLVR) has moved part of the training signal away from subjective human preference labels. Reasoning models such as DeepSeek-R1 are trained with rule-based rewards, accuracy checks for math and compiler feedback for code, that need no rater in the loop for domains with an objective ground truth. ^[24] Second, LLM judges now absorb a large share of routine evaluation. Even so, human annotation still dominated the labeling market in 2025, when manual workflows held roughly 78% of activity and the overall AI data labeling market was worth about $1.9 billion, and human raters remain the only source of signal for open-ended, subjective, or safety-critical judgments that no verifier or model can settle. ^[21]

The industry has not yet figured out a sustainable model for this work. LLM judges absorb some of the demand, but the human signal at the top of the funnel is still what defines what a good response looks like. Whoever the rater is, in whatever country, on whatever pay scale, is implicitly setting the values that the eventual model will optimize for.

References

1. Lianmin Zheng et al. *Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena*. NeurIPS 2023. https://arxiv.org/abs/2306.05685 ↩
2. Mor Geva, Yoav Goldberg, Jonathan Berant. *Are We Modeling the Task or the Annotator? An Investigation of Annotator Bias in Natural Language Understanding Datasets*. EMNLP-IJCNLP 2019. https://aclanthology.org/D19-1107/ ↩
3. Billy Perrigo. *Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic*. TIME, January 18, 2023. https://time.com/6247678/openai-chatgpt-kenya-workers/ ↩
4. Google. *General Guidelines* (Search Quality Rater Guidelines), September 11, 2025. https://guidelines.raterhub.com/searchqualityevaluatorguidelines.pdf ↩
5. Google Search Central. *Our latest update to the quality rater guidelines: E-A-T gets an extra E for Experience*, December 15, 2022. https://developers.google.com/search/blog/2022/12/google-raters-guidelines-e-e-a-t ↩
6. Wikipedia. *Amazon Mechanical Turk*. https://en.wikipedia.org/wiki/Amazon_Mechanical_Turk ↩
7. Wikipedia. *ImageNet*. https://en.wikipedia.org/wiki/ImageNet ↩
8. Wikipedia. *Reinforcement learning from human feedback*. https://en.wikipedia.org/wiki/reinforcement_learning_from_human_feedback ↩
9. Hugging Face. *Illustrating Reinforcement Learning from Human Feedback (RLHF)*. https://huggingface.co/blog/rlhf ↩
10. Sacra. *Surge AI revenue, funding & news*. https://sacra.com/c/surge-ai/ ↩
11. Sam Blum. *Lawsuit Calls Scale AI 'the Sordid Underbelly of the Generative AI Industry'*. Inc., 2024. https://www.inc.com/sam-blum/scale-ai-lawsuit-alexandr-wang/91064163 ↩
12. Encord. *Understanding Krippendorff's Alpha: Inter-Rater Data Reliability Metric in Machine Learning*. https://encord.com/blog/interrater-reliability-krippendorffs-alpha/ ↩
13. McHugh, Mary L. *Interrater reliability: the kappa statistic*. Biochemia Medica, 2012. https://pmc.ncbi.nlm.nih.gov/articles/PMC3900052/ ↩
14. TechCrunch. *Workers that made ChatGPT less harmful ask lawmakers to stem alleged exploitation by Big Tech*, July 14, 2023. https://techcrunch.com/2023/07/14/workers-that-made-chatgpt-less-harmful-ask-lawmakers-to-stem-alleged-exploitation-by-big-tech/ ↩
15. Google Search Central. *Search Quality Rater Guidelines: An Overview*, November 2023. https://developers.google.com/search/blog/2023/11/search-quality-rater-guidelines-update ↩
16. Barry Schwartz. *Google Search Quality Raters Guidelines Updated 9/11*. Search Engine Roundtable, September 15, 2025. https://www.seroundtable.com/google-search-quality-raters-guidelines-update-40092.html ↩
17. Google. *How Search Works: Rigorous testing*. https://www.google.com/search/howsearchworks/how-search-works/rigorous-testing/ ↩
18. Janakiram MSV. *Meta Invests $14 Billion In Scale AI To Strengthen Model Training*. Forbes, June 23, 2025. https://www.forbes.com/sites/janakirammsv/2025/06/23/meta-invests-14-billion-in-scale-ai-to-strengthen-model-training/ ↩
19. Julie Bort. *Scale AI lays off 14% of staff, largely in data-labeling business*. TechCrunch, July 16, 2025. https://techcrunch.com/2025/07/16/scale-ai-lays-off-14-of-staff-largely-in-data-labeling-business/ ↩
20. Reuters. *Exclusive: Scale AI's bigger rival Surge AI seeks up to $1 billion capital raise*. July 2025. https://finance.yahoo.com/news/exclusive-scale-ais-bigger-rival-152327620.html ↩
21. Mordor Intelligence. *AI Data Labeling Market Size, Share & Trends*. 2025. https://www.mordorintelligence.com/industry-reports/ai-data-labeling-market ↩
22. Wikipedia. *LLM-as-a-Judge*. https://en.wikipedia.org/wiki/LLM-as-a-Judge ↩
23. Seungone Kim et al. *Prometheus: Inducing Fine-grained Evaluation Capability in Language Models*. ICLR 2024. https://arxiv.org/abs/2310.08491 ↩
24. DeepSeek-AI. *DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning*. 2025. https://arxiv.org/abs/2501.12948 ↩