Four kinds of employer
There is no single list of good annotation companies, and this course will not give you one. Reputations in this industry change fast: a platform that paid promptly and treated people decently two years ago may have lost the client, cut rates or gone quiet since. Naming names would age badly and could send you somewhere that is no longer safe.
What is stable is the shape of the market. Almost all legitimate work comes from one of four places.
Annotation vendors and outsourcing firms. Companies whose whole business is supplying labelled data to AI developers. They hire in volume, often as contractors, sometimes as employees. They usually have a real careers page, a registered company address and named staff.
Crowdsourcing marketplaces. Open platforms where you create an account, qualify for projects and pick up tasks from a queue. Fast to join, high variability in pay and availability, and mostly piece rate.
Contractor programmes run for AI labs. Larger AI developers hire raters and reviewers, nearly always through a staffing or vendor partner rather than directly. These roles are often advertised with titles that do not contain the word annotation at all.
Transcription, localisation and language firms. Long-established businesses that now also supply speech and language data. If you have a second language or a specific accent, this is often the most accessible door.
In plain English
- Vendor:
- A company that supplies annotated data to AI developers and hires the annotators, usually as contractors.
- Qualification test:
- An unpaid assessment you must pass to get access to a project. Normal in this industry, unlike a fee.
- Independent worker reviews:
- What current workers say on forums, subreddits and review sites the company does not control.
- Registered entity:
- A company that legally exists somewhere, with a findable registration number and address.
- Job title drift:
- The same work advertised under many names, such as AI trainer, search quality rater or content analyst.
Finding the openings
Search using the industry's own vocabulary rather than the phrase you have in your head. Roles get posted as data annotator, AI trainer, search quality rater, transcriptionist, linguistic data collector, content analyst, model evaluator and AI tutor, among others.
Then take one step that removes most of the danger: once you have a company name, close the message or advert and navigate to that company's own website yourself. Apply from their careers page. Never apply through a link someone sent you.
Fake careers pages that copy a real company's branding are common and cheap to build. Typing the company name into a search engine yourself, and checking the web address carefully, costs ten seconds and defeats most of them.
"[company name]" payment problems
"[company name]" not paying
"[company name]" review worker
"[company name]" scam
site:reddit.com "[company name]" annotation
Read the negative results properly. Every large employer has
unhappy people. What you are looking for is a PATTERN: many
different workers, over months, describing the same failure to pay.
The five tests that separate real from fake
One. Money flows towards you only. No fee for training, no fee for equipment, no fee for a starter kit, no deposit, no software you must buy. There is no legitimate exception to this.
Two. Independent evidence exists. Workers discussing the platform somewhere the company cannot moderate. Total silence about a company that claims thousands of workers is itself a warning.
Three. Something is in writing. A contract or terms document naming the parties, the rate, the payment schedule and the grounds for rejecting work. If nobody will put the rate in writing, there is no rate.
Four. The payment mechanism is boring. Bank transfer, a known payment provider or a normal payroll system, on a stated schedule. Cryptocurrency, gift cards, or a request to receive money and forward it onward are not payment methods for a job.
Five. The company is findable. A registered entity, an address, a real website with history, staff who exist elsewhere on the internet. A brand-new page and a messaging app account are not a company.
Checkpoint
Money only ever flows towards you, evidence of real workers exists, terms are in writing, payment is through ordinary channels, and the company is findable. Five tests, applied to every platform.
BEFORE SIGNING UP
[ ] Found their careers page myself, not via a link I was sent
[ ] Company has a registered entity and a physical address
[ ] Independent worker discussion exists and is not uniformly awful
[ ] No payment of any kind is requested from me, ever
[ ] Written terms state rate, schedule and rejection grounds
[ ] Payment threshold and method are clearly stated
[ ] I know who to contact when payment goes wrong
AFTER ONE WEEK
[ ] Tasks were actually available
[ ] Completed work appeared in my balance as expected
[ ] My questions to support got a real answer
[ ] My tracked effective hourly rate is worth the hours
Qualification tests deserve a word. Being asked to complete an unpaid assessment of an hour or two, before you are given access to paid work, is standard practice here. Being asked for money is not. That is the line, and it never moves.
Subject: Question about the data annotator role
Hello,
I am interested in the data annotator position advertised on your
careers page. Before applying I would like to confirm three things:
- Is the role paid per task or per hour, and is training time paid?
- What is the payment schedule and method?
- Does any project I would work on involve graphic or distressing
content?
I am available around 15 hours per week and I work in English and
[your other language].
Thank you,
[Your name]
Sending that message is also a test. A real employer answers it. A scam either dodges the questions or floods you with urgency, which is exactly what the next lesson is about.
๐ Quiz
Question 1 of 4You find a company name in a job advert. What is the safest way to apply?