The work

You receive batches of short audio samples — often a few seconds each — in which a medication name is spoken aloud. Your job is to decide whether the name was pronounced accurately and clearly enough to be unambiguous in practice, and to record that judgment in a rubric with defined score levels. Where a pronunciation is wrong or borderline, you flag the specific error: misplaced stress, a substituted or dropped syllable, a vowel shift that pushes the name toward a look-alike/sound-alike drug, or delivery so rushed or muffled that the name can't be recovered. Written feedback is expected to be short and diagnostic — "stress on second syllable rather than first; still identifiable" rather than a paragraph of commentary.

The volume of judgment calls is what makes this harder than it sounds. You will hit names with more than one defensible American pronunciation, regional variants you have heard daily in your own pharmacy, brand names whose manufacturer-preferred pronunciation differs from common usage, and clips where the speaker is a non-native English speaker but entirely intelligible. The rubric, not your personal preference, decides these. Calibration exists precisely to find out whether you can hold that line.

What the screen looks for

Mercor's AI-led interview probes verifiable specifics first: your licence or certification, setting, years in practice, and how much of your day involved saying drug names out loud to patients and prescribers. It then tests depth — expect to be asked to pronounce or describe the stress pattern on names like eszopiclone, dexmedetomidine, empagliflozin, or hydroxychloroquine, and to explain which sound-alike pairs you consider genuinely dangerous when mispronounced. Finally it tests evaluation judgment: whether you can separate "different from how I say it" from "wrong," whether you defer to a rubric you disagree with, and whether you stay consistent across a long batch rather than drifting.

Logistics

  • Fully remote, U.S.-based, asynchronous — batches are claimed and completed on your own schedule.
  • A paid or unpaid calibration exercise comes before production work; scoring agreement there gates access to batches.
  • Requires quiet listening conditions and decent headphones — laptop speakers are not adequate for distinguishing subtle vowel and stress differences.
  • Work arrives in waves rather than as a steady weekly load; treat it as supplemental income with the possibility of further review projects.
  • No patient care, no clinical decision-making, no licence exposure.