How consistent is the algorithm? Examining the intra- and inter-rater reliability of LLM-based writing assessment

A study published in Frontiers in Education examined the reliability of Large Language Model (LLM)-based writing assessment, specifically using OpenAI's ChatGPT. The research focused on assessing 192 essays from EFL learners, applying a criterion-based analytic rubric. The study found high intra-rater reliability for the LLM, indicating consistency in its scoring over time.

Why it matters

The consistent performance of AI in assessment offers potential for scalable and objective evaluation processes across various domains. Understanding the reliability of LLM-based tools is crucial for their integration into existing frameworks and for developing future strategies that leverage artificial intelligence.

Key insights

• LLM-based scoring demonstrates high intra-rater reliability, meaning consistent scores for the same essays over time.
• OpenAI's ChatGPT was utilized to assess EFL learner essays using a criterion-based analytic rubric.
• The assessment rubric covered task achievement, grammatical range and accuracy, lexical resources, organization, and mechanics.
• The study replicated scoring after three weeks under similar conditions to assess intra-rater reliability.

Source

Frontiers in Education — https://www.frontiersin.org/articles/10.3389/feduc.2026.1861960