What the work actually involves

A generalist writing contract on an AI data platform is not content marketing. On a typical day you pick up a batch with a task spec — often several pages of instructions with edge cases — and produce text against it: writing prompts designed to stress a model in a particular way, drafting the reference answer a model should have given, ranking two or three model outputs and explaining in writing why the winner won, or editing a model's draft into something a careful reader would accept. Batches change subject with little warning: a week of consumer-finance explainers, then travel itineraries, then email rewrites in someone else's voice, then summarising a dense document without importing claims it never made. The constant is not the topic, it's your ability to internalise a spec quickly and stay inside it for hours.

What the screen looks for

micro1's process is largely AI-led — an asynchronous interview plus written work samples, with your reasoning probed under follow-up. Screeners are looking for range you can evidence, not range you claim: published or professional writing across genuinely different registers, and the ability to name why a piece of prose fails rather than just calling it weak. Expect to be asked to justify a judgment and then be pushed on it; the follow-up is the test. Instruction fidelity is weighted heavily, because the most common failure mode among strong writers is quietly improving on a spec they disagreed with. Fluency assessments and grammar precision checks are standard, and originality matters — undisclosed model-generated submissions are the fastest route out of the pipeline.

Logistics

  • Fully remote, contractor status, invoiced hourly against approved time
  • Largely asynchronous; occasional live calibration or onboarding sessions, often US-timezone-leaning
  • Volume fluctuates with client batches — many contributors treat this as 10–25 hrs/week alongside other work rather than a full load
  • Pay in the $40–50/hr range is what contributors have reported for this tier; rates vary by project, geography, and specialisation, and nothing about the band is a guarantee
  • Work is usually under NDA, which limits how much of it you can ever show anyone

The people who last are the ones who find the spec itself interesting — who enjoy the puzzle of "what does this rubric actually mean by unhelpful" — rather than those looking for a place to do their own writing on someone else's budget.