What the work involves

Most tasks fall into a handful of shapes. You write German prompts and reference answers on assigned topics, then rewrite model responses that are grammatically valid but tonally or idiomatically wrong. You compare two model outputs and pick the better one, with a written rationale in English explaining what tipped the decision — register, case agreement, false friends, calques from English, or a factual error about German legal, administrative, or cultural context. Some projects add localization review: catching text that reads as machine-translated, mishandled formal/informal address, or Austrian and Swiss variants treated as errors when they are not.

  • Rewriting and grading German model outputs against a rubric
  • Authoring prompts and gold-standard responses in specific domains
  • Side-by-side preference ranking with English-language justifications
  • Flagging translationese, hallucinated regional detail, and register mismatches

What the platform screens for

micro1's screening is AI-led: a structured async interview that asks about your background, then probes it. Expect follow-ups that test whether your German is lived rather than studied — regional variation, register shifts, technical vocabulary in your stated field. Credentials that help include translation or linguistics degrees, CAT-tool or MT post-editing experience, teaching qualifications (DaF/DaZ), or professional work conducted in German. Equally important is evaluation judgment: whether you can rank two mediocre outputs consistently and explain the ranking in terms a rubric would recognize, rather than defaulting to "it sounds better."

Logistics

Fully remote and asynchronous. Work arrives in batches; contributors commonly report 10–25 hours per week when a project is live, with quiet gaps between engagements. Rates sit in the $30–65/hr range as observed on the platform, with the upper end going to specialists — legal, medical, engineering German — and to reviewers whose annotations survive quality audits. Reliable availability windows and a stated weekly capacity matter more than volume promises you cannot keep.