How we vet AI trainers: five gates and a probation
Why the cheap checks come first, why a person runs the last two, and what happens when quality slips.
The cheap checks come first
A sign-up starts with a residency self-certification, an IP check and a device and phone check. Then come a native-language test and a skill test, both automated and free for the candidate.
Passing those three is Tier 1, which we never sell. The paid identity check comes next, so we only pay for it for people who have already shown they can do the work.
A person runs the last two
Gate four is a document and liveness check followed by a proctored retest on new items, with webcam and screen recorded. Gate five is a recorded work sample: a 20-minute task with camera and screen on, then a 2-minute spoken explanation in the candidate’s language, scored by a native reviewer.
Recording a test alone does not stop someone using an AI tool on a phone. Hearing them explain their choices aloud, in their own language, makes that much harder.
Probation on real work
Tier 2 is what clients hire. Every worker’s first 10 billed hours on a new project are double-reviewed. Tier 3 takes 200 or more billed managed hours with a QA score of at least 92%.
One score, one set of rules
The QA score is a worker’s gold-set accuracy over the last 30 days. Under 85% on two working days in a row means a pause and a paid recalibration; still under 85% two weeks later means a tier down and off the project.
Nobody is banned on an automated flag alone, and any decision can be appealed within 14 days to a different reviewer.