What the work involves

You'll be recording spoken audio against supplied transcripts so an AI lab can train and evaluate how its model handles speech. The brief calls for a "customer service" register — neutral to upbeat, clearly enunciated, consistently paced, without the performance flourishes that would be welcome in an audiobook or character read. Sessions are self-directed: you receive scripts, record in your own space, do light post-production, and submit finished files against a turnaround. A kickoff call at the start of the project sets specification, file format, and QA expectations.

The detail that trips people up is punctuation. Transcripts are to be read exactly as written — commas, em dashes, ellipses and sentence breaks are signal, not suggestion — or, where the instruction allows, edited to reflect natural pacing and flagged as such. Silently "fixing" awkward phrasing, dropping a repeated word, or smoothing a stilted clause corrupts the alignment between text and audio that the whole dataset depends on. Expect a share of your submitted audio to be spot-checked and returned if it drifts from the script or the tonal spec.

What the platform screens for

Mercor's screen is an AI-led interview, and for this role it leans heavily on two things: provable provenance of your accent and the reality of your recording chain. Born and raised in the United States, native English speaker — these are stated gates, not preferences. On equipment, be ready to name your microphone, interface, treatment, and editing software specifically; vague answers about "a good USB mic setup" read as a signal that noise floor and room reflections will fail QA. Professional voice acting credits help but are explicitly not required — music, podcasting, streaming, radio, comedy and YouTube experience all count if you can talk about mic technique and clean delivery.

Logistics and pay

  • Fully remote and largely asynchronous, except for the kickoff call
  • Roughly 1–2 weeks from your start date, with a March 9, 2026 project start
  • 2–10 hours per week commitment
  • Pay is quoted by the platform as $250 USD per finished hour of submitted audio — a finished hour is audio delivered, not time at the mic, so your effective clock rate depends on how many takes and how much editing each page costs you. The $50/hr band shown on this card reflects a conservative observed clock-hour equivalent, not a guarantee. Clean readers with a treated room land well above it; heavy retakers land below.