What the work actually involves
You will spend most of your time reading model outputs about games and deciding whether they are actually correct. That means catching a model that confuses Street Fighter II's Champion Edition roster with Super Turbo's, misstates frame data or input windows, invents a patch note, gives strategy advice that was valid two metas ago, or describes an arcade cabinet's hardware that never existed. Tasks arrive in batches with a rubric: label a sample, rank two responses, flag a factual error, or write a prompt designed to expose a weakness in the model's game knowledge. Prompt-writing tasks ask you to lean on the specific corners you know well — a genre, a competitive scene, a console generation, an emulation quirk — rather than general trivia.
What the screen looks for
AfterQuery's intake is AI-led and follow-up heavy. It is trying to separate someone who has logged thousands of hours and can cite specifics from someone who has read a lot of gaming journalism. Expect to be asked which titles, which platforms, which years, which ranks or tournament results, and to be pushed one layer deeper on whatever you claim. Verifiable artefacts help: a ladder rank, a shipped credit, a QA employer, a content channel, a tournament bracket placing, a speedrun leaderboard entry. Vague breadth ("I play everything") screens worse than narrow depth.
Logistics
- Fully remote and asynchronous; you pick up available task batches rather than working set shifts.
- Paid per approved task, so effective hourly rate depends on your accuracy rate and pace — rejected work is unpaid. The $10–30/hr band reflects observed outcomes, not a guarantee.
- Referral and speed bonuses sit on top of base task pay.
- Volume fluctuates by project. Treat this as supplementary rather than a replacement for primary income.
- No fixed weekly minimum is stated, but batches can expire, so responsiveness matters more than raw hours.