If you need text labeled for an NLP model or an LLM, these are the 10 text annotation services to compare in 2026: Zilo AI, TELUS Digital, Appen, LXT, Scale AI, iMerit, Defined.ai, Shaip, Sama and CloudFactory. All of them turn raw text into labeled training data, and most now also help train LLMs with human feedback.
Check the names before you sign anything, because several have changed. TELUS International became TELUS Digital in September 2024. Meta bought 49% of Scale AI. Shaip became part of Ubiquity Global Services in February 2026, and EXL completed its purchase of iMerit in August 2026. Ownership matters, because a new owner can change how your data is handled.
Below: what each one labels, who it suits and how to choose. First, our pick for teams that need labeled data and the people to use it.
Text annotation companies at a glance
| # | Provider | Text work it highlights | Best for |
|---|---|---|---|
| 1 | Zilo AI | Sentiment, entities, classification, relation extraction, topic modeling | Labeled data plus IT and data hires |
| 2 | TELUS Digital | Classification, entities, intent, question answering, summarization | Large multilingual programs |
| 3 | Appen | Fine-tuning demonstrations, expert RLHF, red teaming | LLM alignment data |
| 4 | LXT | Entity tagging, sentiment, LLM red-teaming and RLHF | Rare languages and locales |
| 5 | Scale AI | Expert-built training sets and model evaluation | Large LLM programs |
| 6 | iMerit | Chatbot intents, search relevance, moderation, documents | Regulated, domain-heavy text |
| 7 | Defined.ai | Model-in-the-loop labeling, ready-made text datasets | Compliance-led buying |
| 8 | Shaip | Entity linking, subject-action-object tagging, coreference | Clinical and legal NLP |
| 9 | Sama | Instruction following, preference ranking, factuality | LLM tasks from a B Corp |
| 10 | CloudFactory | AI-assisted pre-labeling for LLM and NLP models | Labeling plus model oversight |
Details come from each provider’s own website, checked in September 2026, unless another source is linked.
What text annotation services include
A text annotation service has trained people mark up what text means, often starting from a model’s first guesses, so that your model can learn from the result. The work falls into five groups:
- Named entity recognition (NER): tagging people, companies, products, dates and amounts. Our guide to named entity recognition shows how it works.
- Classification: sorting whole texts by topic, intent or policy, such as routing support tickets.
- Sentiment: marking reviews, posts and chats as positive, negative or neutral, or by finer emotions.
- Relations and linking: connecting entities to each other, or to a knowledge base.
- LLM data: writing prompts and model answers, ranking responses for reinforcement learning from human feedback (RLHF), checking facts and red-teaming for unsafe output.
What you are really paying for is the part you never see: written guidelines, a review layer and a fair way to settle disagreements between annotators.
1. Zilo AI
Zilo does two jobs that usually sit with different vendors. It labels the text your model learns from, and it hires the people who build the model. Its text annotation service covers sentiment analysis, named entity recognition, text classification, relation extraction and topic modeling. Zilo says its team of more than 1,600 trained annotation and speech recognition experts has annotated over 10 million data points, for industries from retail to banking and healthcare. It also offers image and voice annotation, audio, video and multilingual transcription, translation in more than 25 listed languages, and staffing for IT roles from data scientists to cybersecurity analysts. The team is based in Bangalore.
Best for: teams that want labeled training data and specialist hires from one partner. See Zilo’s services.
2. TELUS Digital
TELUS Digital was called TELUS International until its September 2024 rebrand. Its text labeling covers classification, named entity recognition, keyphrase extraction, intent classification, question answering, sentiment analysis and summarization, using its own Ground Truth Studio platform. Its data annotation page cites more than 2 billion labels a year, a community of over 1 million annotators and linguists, and 500+ languages and dialects.
Best for: large programs that want many languages from one supplier’s platform and workforce.
3. Appen
Appen, headquartered in Sydney, has been building human-labeled language data since 1996, when it started with speech recognition and language processing. Its text work now leans toward large language models: fine-tuning demonstrations, RLHF from subject-matter experts, step-by-step reasoning traces and red teaming. It covers 500+ locales and is SOC 2 and ISO 27001 certified.
Best for: LLM teams that want a long-established vendor for fine-tuning and alignment data.
4. LXT
LXT, founded in 2010 and headquartered in Canada, specializes in language and speech data and states the widest language reach on this list: over 1,000 language locales in more than 150 countries. It bought clickworker, a German crowdsourcing platform, in a deal announced in December 2024, and now cites a crowd of more than 10 million contributors. Its text work runs from entity tagging and sentiment analysis to red-teaming and RLHF for LLMs.
Best for: projects that need rare locales or a very large crowd.
5. Scale AI
Scale AI, founded in 2016 and based in San Francisco, started with data for autonomous vehicles and now works on frontier AI models. Its Generative AI Data Engine builds training sets with vetted experts, linguists and coders, and Scale also runs model evaluations.
Meta took a 49% stake for $14.3 billion, and founder Alexandr Wang moved to Meta. Scale says it remains independent and committed to safeguarding customer data. If you compete with Meta, ask exactly how your data is kept apart.
Best for: large LLM programs that need expert-written training and evaluation data.
6. iMerit
iMerit is now part of EXL, the New York data and AI company, which completed the acquisition in August 2026. Its text work runs on its own Ango Hub platform and spans chatbot intents and entities, search relevance, content moderation, document extraction for finance and insurance, and compliance screening. For harder generative AI tasks it draws on Scholars, a network of more than 25,000 domain experts that includes physicians, scientists, engineers and linguists. It lists SOC 2, ISO 27001, GDPR and HIPAA compliance.
Best for: regulated, domain-heavy text such as clinical notes, insurance claims and policies.
7. Defined.ai
Defined.ai, founded in Seattle in 2015 by its CEO Daniela Braga, runs two businesses side by side: a marketplace of ready-made training datasets, text included, and custom collection and annotation with a model in the loop. It lists ISO 27001, 27701 and 42001 certification, the last being the standard for AI management systems, and says it is GDPR and HIPAA compliant. It puts its crowd at 1.6 million-plus people in 150+ countries, covering 500+ languages, dialects and locales.
Best for: buyers whose legal or procurement teams will scrutinize how the data was sourced.
8. Shaip
Shaip, operating since 2019, has a strong healthcare slant: it sells medical datasets and medical annotation alongside general AI data. In February 2026 it became part of Ubiquity Global Services, a New York-based operations and customer experience company. Its text work includes named entity recognition linked to knowledge bases such as UMLS and ICD-10, subject-action-object tagging and coreference resolution. It routes work by domain, with clinical specialists for healthcare and JD-credentialed reviewers for legal text, and states 150+ languages, including low-resource Indic and African ones.
Best for: clinical and legal NLP, and Indian-language text.
9. Sama
Sama was founded in 2008 by the late Leila Janah, on the belief that dignified, living-wage work is the best way out of poverty, and it was one of the first AI companies certified as a B Corp. Its text annotation targets generative AI: instruction following, preference ranking and factuality. Most case studies on its homepage are computer vision projects, so ask for text references before you commit.
Best for: LLM text tasks from a supplier with an impact-sourcing model.
10. CloudFactory
CloudFactory was founded in Nepal in 2010 and built its business hiring and training a global AI workforce. It says it has helped more than 700 clients, and it has since moved beyond labeling: annotation is now the data layer of an AI platform that also covers model oversight and human review of AI decisions in production. That data layer supports LLM and NLP models, with AI-assisted pre-labeling.
Best for: companies that want labeling and production oversight from one supplier.
Multilingual text annotation services
If your data spans many languages, compare the coverage each provider states. As of September 2026:
| Provider | Language coverage it states |
|---|---|
| Zilo AI | 25+ languages listed, from German and Arabic to Vietnamese and Bahasa |
| LXT | 1,000+ language locales |
| TELUS Digital | 500+ languages and dialects |
| Appen | 500+ locales |
| Defined.ai | 500+ languages, dialects and locales |
| Shaip | 150+ languages |
A locale is a language as used in one place, such as French in Canada, so locale counts run higher than language counts. Read all of these as the size of a network, not proof of quality in your language. Put your hardest, lowest-volume languages into the pilot, and ask whether a native speaker reviews each one.
Outsource text annotation or use a platform?
You have three routes. Software alone, like the tools in our guide to text annotation software, suits teams with their own annotators and data that cannot leave their environment. A managed service suits teams whose volume swings, who need many languages or domain experts, or whose ML engineers have ended up doing the labeling. The third route is both at once: iMerit delivers its work on Ango Hub, and TELUS Digital on Ground Truth Studio. Whichever you pick, keep ownership of the guidelines and the gold-standard examples, because they are what let you switch vendors later.
How to choose a text annotation company
- Labeled text plus hires for your own team: Zilo AI.
- Many languages at production scale: LXT, TELUS Digital, Appen or Defined.ai.
- LLM fine-tuning, ranking and evaluation data: Scale AI, Appen, iMerit or Sama.
- Clinical, legal or insurance text: Shaip or iMerit.
- Strict security reviews: Defined.ai, iMerit or Shaip, which all list ISO, GDPR and HIPAA credentials.
- Labeling plus oversight of models in production: CloudFactory.
Then run the same small pilot with two or three of them. Use your real schema and your hardest examples: sarcasm, mixed languages, overlapping entities. Agree the pass mark before work starts, such as a minimum agreement rate between annotators, and ask who settles disagreements and how guideline changes reach every annotator. Our guide to data quality for machine learning covers how to measure the result. Last, read the change-of-control clause. iMerit and Shaip were both sold this year, so decide now what happens to your data if your vendor changes hands.
Frequently asked questions
What are text annotation services?
Companies that label raw text so NLP models and LLMs can learn from it. The work includes tagging entities, sentiment and intent, linking related terms, and ranking or fact-checking model answers, all backed by written guidelines and a review layer.
What is the difference between text annotation and text labeling?
In practice there is none, and vendors use the two terms interchangeably. Where teams do draw a line, labeling means giving a whole text one category, and annotation means marking spans inside it, such as entities.
Who provides multilingual text annotation at production scale?
Zilo AI pairs text annotation with multilingual transcription. Among providers that publish figures, as of September 2026 LXT states 1,000+ language locales, TELUS Digital 500+ languages and dialects, Appen 500+ locales and Defined.ai 500+ languages, dialects and locales.
Should I outsource text annotation?
Outsource when your volume swings, when you need many languages or domain experts, or when your ML engineers are spending their time labeling. Keep it in-house when the data cannot leave your environment or the labeling rules still change every week.
What should I look for in a text annotation platform?
Check that it handles your label types, such as spans, relations and document classes, imports pre-labels from your own model, includes a review step and exports straight into your training format. If you also need the annotators, look at providers that bundle both, such as iMerit with Ango Hub or TELUS Digital with Ground Truth Studio.