Anchor Is the Key: Toward Accessible Automated Essay Scoring with Large Language Models Through Prompting

Synthesis: Choi, Tate, Ritchie, Nixon & Warschauer (2025) investigate the most practical approach to LLM-based automated essay scoring — prompting — and find that providing anchor papers (example essays with scores) significantly improves LLM-human agreement, bringing it close to human-human scoring reliability. GPT-4o mini achieves comparable results to GPT-4o at substantially lower cost, making accessible, teacher-friendly AES feasible.

Key Findings

Implications

This study advances Automated Essay Scoring by shifting focus from resource-intensive model optimization to accessible prompting strategies. For teachers, the finding that GPT-4o mini with anchor papers approaches human reliability means practical AES is within reach — no expensive compute, no large pre-scored essay banks. The anchor paper approach connects to Prompt Engineering best practices and suggests a pathway for Writing Education where teachers can calibrate AI scoring to their own assessment standards rather than relying on black-box systems.

The work complements Psyscore Essay Scoring ZPD Feedback research on psychometrically-aware scoring and Icle Plus Plus Essay Scoring work on fine-grained trait scoring, showing that prompt design alone — particularly anchor inclusion — can achieve strong holistic scoring. For AI Literacy, this empowers educators to understand and control AES rather than treating it as an opaque tool.

Connected Concepts

Connected Articles

Citation

Choi, J., Tate, T., Ritchie, D., Nixon, N., & Warschauer, M. (2025). Anchor Is the Key: Toward Accessible Automated Essay Scoring with Large Language Models Through Prompting. EdArXiv. doi:10.35542/osf.io/cbhgz_v1.