What the work actually involves

You install a specialized build of an AI assistant app on your personal Mac and use it the way a real person would — dictating messages and emails in Thai, asking it to format a document, apply a spreadsheet formula, generate an image, or summarize a file. The point is natural speech, not scripted prompts, so a lot of the value comes from you genuinely using productivity apps you already know rather than performing test cases. Each interaction then becomes data: you capture the audio, write a verbatim transcript of what you said, produce a cleaned-up version, rate the assistant's response, and enter the whole thing into a structured tracking sheet with recordings uploaded to a shared drive.

Expect the logging to take as long as the speaking. A single two-minute voice task can mean five to ten minutes of transcription, rating, and sheet entry. Reviewers care that the transcript matches the audio exactly, that Thai orthography and spacing are consistent, and that your ratings are defensible against what the assistant actually produced — not what you hoped it would.

What the platform screens for

Mercor runs a roughly 25-minute AI-led conversational interview covering your background, your day-to-day app habits, and your motivation for a short project. Two things get verified hard: that Thai is genuinely your first language with Thailand-native usage, and that your machine is an M1/M2/M3 Mac on macOS Sequoia 15.0 or later — this is a stated hard requirement, not a preference, and Intel Macs cannot run the build. English comprehension is checked functionally: all instructions, rubrics, and sheet headers are in English while your outputs are in Thai.

Logistics

  • Fully remote, asynchronous, on your own hardware
  • Four-week duration with immediate start
  • Observed rate of $15/hr; Mercor states hours and rates per project, and nothing here is guaranteed
  • Onboarding includes a Mercor Expert email setup and a technical setup step before task work begins
  • Self-directed pacing, but the tracking sheet makes throughput and quality visible, so consistent weekly blocks work better than sporadic bursts