What the work actually is

Each item arrives as a complete package: the task prompt as it was written, a transcript of what the AI agent did, whatever files it produced (a marked-up draft, a filed story, an edit decision list, a cut, a delivery spec sheet), and a scoresheet an AI grader has already filled in. Your job is to audit that package from the position of someone who does the work for a living. Four questions, in order: is this a task that would genuinely come up in this job, described the way a real commissioner or producer would describe it; is the domain content right, including style rules, sourcing standards, legal constraints, broadcast specs, figures and dates; do the grader's eleven yes/no answers survive contact with the files; and is the reported score defensible against the band you would give the work yourself.

The fourth question is the one that separates this from proofreading. You band the work — could a practitioner use it, accept it, or would it come back — and then check the reported score against your band. Where you disagree with the grader, you write a short reason naming the specific line, frame, figure or clause that drives the disagreement. Budget about an hour per item. You will not be asked to rewrite copy, re-report a story, or recut a sequence.

What the platform screens for

Three tracks, and you apply to the one that is actually yours:

  • Editors — 4+ years editing other people's copy to a house style, with responsibility for what published.
  • News Analysts, Reporters and Journalists — 4+ years reporting and filing your own stories, with sourcing you had to stand behind.
  • Film and Video Editors — 4+ years cutting finished pieces to a professional timeline, delivered to a client or broadcast spec.

Screens probe for the editorial chain, not the byline. Expect follow-ups on a story you killed, a sourcing problem you sent back, a legal or factual catch you made before publication, or a delivery you rejected against spec. Content marketers writing their own copy with no editor above them, and people whose only editing experience is on their own work, tend not to clear. Near-native business English is required because every judgment carries a written reason, and you need enough comfort with a spreadsheet or PDF to check whether the numbers in one reconcile.

Logistics

Remote and asynchronous, work drawn from a queue on your own schedule. Observed pay for this listing is $45–50/hr — as reported, not guaranteed, and subject to the platform's own banding at offer. Volume fluctuates with project demand; most reviewers treat this as part-time alongside their working practice rather than a full load.