DataAnnotation's job board lists 43 language roles among 133, all at up to USD 40 per hour (checked 2026-11-09), while Luel states USD 7.50 per hour for partner audio in five low-resource languages against USD 12 for American English (checked 2026-11-17).
Language work is one route into AI training data licensing for people rather than companies. No AI developer publishes which languages it buys human work in, and no platform publishes task volume per language, so a listed role is not a promise of hours.
What language work in AI training covers: writing, rating, recording and annotation
Language work in AI training takes three forms on the platforms' own pages: written and evaluation tasks in a language, spoken recordings, and annotation of language data.
DataAnnotation lists language roles on its job board, Luel pays for recorded conversations and partner audio, and TELUS Digital describes its AI Community tasks as "annotation, rating, data collection tasks" with annotation in more than 500 languages and dialects (checked 2026-11-09). SpeechData.ai states 7,250 native speakers.
Recording work produces speech data that the platform or vendor then holds, and text work produces demonstrations and ratings. The sibling pages named in the last section compare those sides, so this page stays with which languages, what pay and how much work.
Platforms that state language roles or language tasks, and the languages they name
Four companies state language roles, language tasks or language coverage for contributors or speakers on their own pages, and only one of them names the individual languages. The table is alphabetical, as of 2026-11-17, with page checks of 2026-11-09 unless marked.
- Inclusion means the company's own page names language roles, named languages, native speakers or a language and dialect count.
- "Not stated" means that no page read gives it.
- Pay is in the pay table further down, not here.
| Company | Contracting or legal entity on record | Language scope as stated (its wording) | Contributor or speaker rule as stated | Source |
|---|---|---|---|---|
| DataAnnotation | Surge Labs, Inc. (dba DataAnnotation) | 43 language roles among 133 roles on its job board | "fluent in English" (FAQ, re-read 2026-11-16) | dataannotation.tech; dataannotation.tech/faq |
| Luel | entity not in the record | task cards for American English conversations and for partner audio in Uyghur, Tibetan, Hmong, Jamaican Creole and Guarani; "low-resource language audio" among data wanted (re-read 2026-11-17) | not stated; "900,000+ contributors" (Luel's claim) | luel.ai/contribute |
| SpeechData.ai | Spire Light ApS, CVR 44262509 (Denmark) | "60 languages and dialects"; 7,250 native speakers; 73k hours | every speaker signed explicit commercial-use consent; no public contributor program | speechdata.ai |
| TELUS Digital | TELUS International (Cda) Inc.; wholly owned by TELUS Corporation since 2025-10-31 | "500+" languages and dialects in annotation; AI Community of one million contributors across more than 118 countries (FY2024 annual report) | not stated on its pages | telusdigital.com/solutions/data-for-ai-training; sec.gov |
Luel is the only company that names languages, five low-resource languages plus American English. DataAnnotation counts roles, while TELUS Digital and SpeechData.ai count languages and dialects, and each count is the company's claim, so the counts are not comparable with each other.
Platforms that publish location lists rather than languages (Outlier, Alignerr and the Handshake AI Fellowship) are treated under availability below.
Demand signals on record: role counts, language counts and a regional revenue split
No AI developer publishes which languages it buys human work for, so the demand signals on record come from the supply side: how many language roles a platform lists, how many languages a vendor states, and one regional revenue split.
DataAnnotation's job board of 2026-11-09 lists 133 roles, of which 43 are language roles (about 32 %), the largest category ahead of Medical (19) and Finance (14).
TELUS Digital's 500+ languages and dialects and Luel's "low-resource language audio" are each a company's claim about its own offer, not a measure of what buyers ask for.
The one regional figure in a filing-grade source is Appen's FY25 investor presentation (period ending 2025-12-31): Appen China revenue of USD 102.9 million out of group revenue of USD 230.8 million, about 45 %, up 75 % on FY24. That is a geographic business segment, never a language or a share of contributor work.
A different market sits beside this one: data holders who license existing material. Company-data programs that ask for English or welcome other languages, such as Mercor's publisher and video listings stating "multilingual content welcome" (checked 2026-11-09), are covered under multilingual and low-resource language data.
What language roles pay, as each platform states it
As of 2026-11-17 two platforms publish pay by language: DataAnnotation states up to USD 40 per hour for all 43 language roles, and Luel pays USD 12 per hour for American English conversations and USD 7.50 for partner audio in five low-resource languages. The table is alphabetical by company, and every value is the company's own claim.
| Company and role | Language or scope | Pay as stated | Unit | Who pays | Stated on | Source |
|---|---|---|---|---|---|---|
| DataAnnotation, language roles (job cards) | all 43 language roles | up to USD 40 | per hour | the platform | 2026-11-09 | dataannotation.tech |
| DataAnnotation, FAQ starting rates | multilingual; general for comparison | multilingual USD 25 to 40; general USD 25 to 50 | per hour | the platform | 2026-11-16 | dataannotation.tech/faq |
| Luel, three-person conversations | American English | USD 12 (USD 0.20 per minute) | per hour | the platform | 2026-11-17 | luel.ai/contribute |
| Luel, "Record with Partner" audio | Uyghur, Tibetan, Hmong, Jamaican Creole, Guarani | USD 7.50 (USD 0.13 per minute) | per hour | the platform | 2026-11-17 | luel.ai/contribute |
| Outlier | any; "Rates vary by expertise, project complexity, and location" | no rate published | hourly or per task | the platform | 2026-11-09 | outlier.ai |
| Prolific, participant studies | set per study by the researcher | minimum GBP 6 / USD 8; recommended GBP 9 / USD 12 | per hour | the researcher or AI developer | help page dated 2026-07-17 | researcher-help.prolific.com |
| SpeechData.ai, dataset sales (buyer side) | English (US); Spanish (MX); range across 60 languages and dialects | USD 90; USD 65; USD 60 to 95 | per hour of audio | the buyer pays the vendor; speaker pay not disclosed | 2026-11-09 | speechdata.ai |
| TELUS Digital AI Community | 500+ languages and dialects (annotation) | not published on its pages | n/a | n/a | 2026-11-09 | telusdigital.com |
The SpeechData.ai row is a sale price to buyers, never a speaker's pay.
Luel's low-resource rate is 62.5 % of its English rate (USD 7.50 divided by USD 12).
DataAnnotation's language ceiling of USD 40 sits below its generalist ceiling of USD 50 on its job cards, and its FAQ starting range for multilingual work (USD 25 to 40) sits below the general range (USD 25 to 50), as re-read on 2026-11-16. No platform's own page states a premium for a rare language.
The buyer side of a speech hour is a different number. SpeechData.ai sells audio at USD 60 to 95 per hour (English US USD 90, Spanish MX USD 65, checked 2026-11-09) and does not disclose speaker pay, so a vendor price is never the speaker's pay.
Rates across all fields, dated, sit under pay for AI training work by platform and field.
How much work a language role brings: availability, locations and English requirements
A listed language role is not a promise of hours: the platforms state no minimum or fixed hours, pay per task or per minute of recording, and limit who may work by location and by English fluency.
DataAnnotation states "no minimum hours" and a Starter Assessment of about an hour that can be taken once, and its FAQ says workers must be "fluent in English" (checked 2026-11-09 and re-read 2026-11-16).
Mindrift states "no employment contract and no fixed hours" with pay per completed task, and Luel's task rates are per minute of recording. No platform publishes task volume per language.
Outlier lists 117 working locations in its Flexible Working Guidelines effective 2026-07-15, with Kenya, Nigeria and Pakistan not listed, requires an ID and a mobile phone from the country of residence, prohibits VPN use and states that rates vary by location (checked 2026-11-09).
The Handshake AI Fellowship is for US-based fellows, and Alignerr accepts applicants worldwide but only US identity documents. How platforms test and onboard is compared under AI trainer assessments.
Language work for contributors: benefits, risks, downsides and alternatives
Language work lets native speakers of languages that platforms list earn per hour or per task from home, but the rates on record are not higher for rarer languages and the work belongs to the platform once submitted.
The lists set out the contributor's position, and every figure is a claim or a record as dated.
Benefits
- A large share of listed roles at one platform: 43 of 133 on DataAnnotation's job board (2026-11-09).
- A low entry requirement of one assessment that can be taken once, as DataAnnotation states (2026-11-09).
- Recording tasks that Luel states are paid daily (2026-11-17).
Risks
- Outlier's terms (effective 2026-05-11) have the contributor irrevocably assign all rights worldwide in work product (s.4(b)) and require individual arbitration with a 30-day opt-out for US residents (s.10).
- noyb and three groups filed five GDPR complaints against Scale AI over Outlier in France, Italy, the Netherlands and Austria on 2026-10-27; they are complaints, not findings.
- Classification disputes: McKinney v. Scale AI (Superior Court of California, County of San Francisco, CGC-24-620481) is a class settlement of gross USD 12,500,000, final as of 2026-11-17 (approval order filed 2026-10-30), with payments reported on Form 1099. The Cavalier v. Surge Labs complaint (San Francisco Superior Court), filed 2025-05-20 against the company behind DataAnnotation, contains allegations only.
Downsides
- Lower stated rates for low-resource audio and for language roles than for generalist roles (Luel, DataAnnotation, as above).
- No stated hours, so income per week is not predictable from a listing.
- English fluency is required even for language roles at DataAnnotation.
Alternatives
- Volunteer recordings for an open corpus: Common Voice Scripted Speech v27.0 (2026-09-21) covers 295 languages under CC0, unpaid.
- Licensing existing material instead of producing new work, such as translation memories and parallel text.
- Expert work in another field the contributor holds.
- Not doing AI training work at all.
What the public record does not show about language work
The record leaves four things open for language work, and none is stated here beyond that gap:
- How many tasks each language receives.
- Which languages AI developers buy human work in.
- Rates per language on platforms other than Luel and DataAnnotation.
- Whether job-board listings match the platforms' own pages, so only the platforms' own pages are used here.
Where language work meets multilingual data, platforms, experts and taxes
This page covers contributor work by language; the pages below cover the wider human data market, the platforms compared, expert fields, taxes for contributors abroad, privacy, legitimacy checks, speech data and the open speech corpus, and how this site is funded.
- Human data for AI: the companies, their platforms and their pay
- AI training jobs and platforms compared, including countries accepted
- Expert contributor work by field, where languages are one of six
- Taxes for AI training contributors, including work from abroad
- Contributor privacy on human data platforms: ID checks and location verification
- Is AI training work legit: checking a listing against the platform's own page
- Selling speech and audio data for AI training
- Mozilla Common Voice, the open volunteer speech corpus
- How DataPacts is funded
Frequently asked questions
Which languages pay the most in AI training work?
No platform's own page states a premium for a rarer language. Luel states USD 7.50 per hour for five low-resource languages against USD 12 for American English (2026-11-17), and DataAnnotation states one ceiling of USD 40 for all 43 language roles (2026-11-09), as in the pay table above.
Can translators do AI training work?
Translators can take language roles where platforms list them, for example the 43 on DataAnnotation's job board (2026-11-09). Translation work for AI is contributor work, while a translation agency's existing memories are licensed as data, as the translation memories page above explains.
Do you need fluent English for multilingual AI training work?
DataAnnotation's FAQ requires workers to be fluent in English, including for language roles (re-read 2026-11-16). Luel states no language rule beyond its task cards (2026-11-17).
Can you do language work for AI training from any country?
No, location rules differ by platform. Outlier lists 117 working locations, with Kenya, Nigeria and Pakistan not listed (2026-07-15), and the Handshake AI Fellowship is for US-based fellows (2026-11-09). Taxes abroad are on the taxes page linked above.