What the work involves
You'll spend most of your time producing and judging text. On the authoring side, that means writing prompts in Netherlands Dutch that stress-test how a model behaves around sensitive subject matter — self-harm, extremism, medical and legal advice, fraud, dual-use technical information — including phrasings that a Dutch speaker would recognise as evasive, coded, or regionally specific but a translation-trained system might not. On the review side, you apply a rubric to prompts and conversations: does this fall inside or outside policy, is the model's refusal calibrated or over-broad, is this an escalation pattern or a single benign question. Every judgment comes with a written rationale in English, and those rationales are what actually get graded.
What the platform screens for
Mercor's screen is AI-led and follow-up heavy. Expect it to verify three things: that your Dutch is genuinely Netherlands-native rather than fluent-second-language or Flemish-inflected, that you can reason in writing about a borderline case without collapsing into "it depends," and that you handle dual-use material soberly — neither refusing to engage with the category nor treating harmful specificity as an interesting puzzle. No AI/ML background is required and claiming one you don't have tends to backfire under probing. Trust and safety, moderation, editorial grading, or red-teaming experience is a genuine plus but not a gate; a bachelor's degree, finished or in progress, is stated as a requirement.
Logistics
- Fully remote, but you must currently be based in the Netherlands — this listing covers Netherlands Dutch, with Belgian Dutch handled as a separate role.
- Asynchronous task queues; hours are typically flexible and self-scheduled, with throughput expectations rather than fixed shifts.
- Observed pay band of $48–52/hr, as reported for this listing — rates on Mercor vary by project and are not guaranteed.
- Start date is immediate, so availability in the next week or two is a real factor in selection.