The work
You put on headphones and listen to AI-narrated audiobooks in Brazilian Portuguese the way an ordinary listener would — except you stop, rewind, and write down what went wrong. Typical findings: a word skipped or hallucinated, `R$ 1.200,00` read as digits instead of currency, an abbreviation expanded incorrectly, stress landing on the wrong syllable in a proper noun, a question read with declarative intonation, a breath or clipping artifact between sentences. Each issue gets a category and a precise timestamp span in the review tool, plus a plain-language note in English so the team can act on it.
Beyond error logging, you report on the listening experience as a whole: where the narration held up over a long chapter, where it became tiring, and whether you would actually keep listening. That judgment — the difference between a technically correct read and one a Brazilian listener would enjoy — is the part the models can't self-assess.
What the screen looks for
Mercor's screening is AI-led and follow-up heavy. Expect it to test three things: that your Brazilian Portuguese is genuinely native-level (regional pronunciation norms, number and date reading conventions, treatment of foreign loanwords), that you are a real audiobook listener with opinions rather than someone who has heard of audiobooks, and that you can describe a defect precisely enough for an engineer to reproduce it. Vague answers get probed; specific ones get followed up on. Prior annotation, transcription, proofreading, or linguistics work is asked about directly.
Logistics
- Remote, location-flexible, though candidates living in a Brazilian-Portuguese-speaking region are preferred.
- Part-time, around 20 hours per week — roughly 4 hours a day, largely async on your own schedule.
- Working English is required for reports and team communication; you do not need to be bilingual.
- Quiet listening conditions and decent headphones matter more here than they do in text annotation work.