How a Kentucky school district is scaling writing feedback with Gemini

Source: Google Date: 2026-06-25 URL: https://blog.google/products-and-platforms/products/education/henry-county-public-schools/

Summary

Henry County Public Schools in Kentucky deployed Gemini to provide personalized writing feedback at scale, grounded in state rubrics and validated against prior official feedback before classroom rollout. The structural problem was load: teachers averaged 180 students each, making individualized written feedback impractical at frequency. After implementation, the share of high school students scoring in the novice tier (below grade level) dropped from 33% to 15% — roughly half the below-grade cohort advanced. Human oversight was preserved throughout: educators review all Gemini-generated feedback before it reaches students.

Implications

• Education-AI adoption. This is a concrete, before/after district-level outcome rather than a pilot study or vendor claim. A 55% reduction in below-grade-level writing scores is a meaningful signal, though a single district case study doesn’t establish causation — writing scores can move for multiple reasons. The value is in the implementation model: state-rubric grounding + curriculum specialist validation + teacher review before delivery is a replicable pattern that addresses the most common objections to AI in student assessment.
• Feedback as the most tractable classroom AI use case. Formative feedback on writing has a well-understood quality bar (alignment to rubric), a clear workload problem (teacher-to-student ratio), and a natural place for human review before any output reaches a student. That combination makes it more straightforwardly deployable than use cases with more ambiguous quality criteria.
• Replication template. The district’s approach — validate against known-good historical feedback, involve curriculum specialists, maintain teacher review — is a low-political-friction deployment template. Other districts watching for a responsible AI-in-schools model have a concrete playbook here rather than abstract principles.