Contracted AI Raters Train Chatbots Under Grueling Conditions

Thousands of contract workers power Google’s Gemini and AI Overviews, moderating and rating outputs under grueling conditions. Recruited under vague job titles, many, like Texas writer Rachael Sawyer, found themselves unexpectedly reviewing violent or explicit AI content without warning or consent. Paid $16–$21 an hour via contractors such as Hitachi’s GlobalLogic, raters face shrinking deadlines, opaque guidelines, and mental health strain. Workers describe editing outputs on topics such as medicine, finance, or astrophysics, despite lacking expertise. At the same time, new rules loosen the guardrails around hate speech and explicit material if initiated by the user. Despite being crucial to AI safety and accuracy, raters feel expendable, hidden, and vulnerable to layoffs. Critics argue “AI isn’t magic; it’s a pyramid scheme of human labor,” with ethics sacrificed for speed and profit.

The Guardian, Varsha Bansal – How thousands of ‘overworked, underpaid’ humans train Google’s AI to seem smart – 11th September 2025