---
title: "When parents use AI to regrade: A new actor in the assessment debate"
author: Martin Ebner
date: 2026-11-12
url: https://postgenerative.at/indications/2026-11-12/when-parents-use-ai-to-regrade-a-new-actor-in-the
lang: en
type: Indication
summary: "A German-language practice article adds a dimension to the assessment debate: AI-assisted parents question school grades and seek transparent marking criteria."
tags:
  - Education
  - Assessment
sources:
  - title: "Lucca Spohn (2026): ChatGPT gibt meinem Kind die bessere Note. Und jetzt? – deutschsprachiger Praxisbeitrag"
    url: https://luccaspohn.substack.com/p/chatgpt-gibt-meinem-kind-die-bessere
  - title: "Zhang et al. (2024): Evaluating large language models for criterion-based grading from agreement to consistency"
    url: https://pmc.ncbi.nlm.nih.gov/articles/PMC11683144/
  - title: "Karjus et al. (2026): Machine-Assisted Grading of Nationwide School-Leaving Essay Exams with LLMs and Statistical NLP – Preprint"
    url: https://arxiv.org/abs/2601.16314
license: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
alternate:
  de: https://postgenerativ.at/indizien/2026-11-12/wenn-eltern-mit-ki-nachbenoten-ein-neuer-akteur-in-der
---

# When parents use AI to regrade: A new actor in the assessment debate

Writing on 4 November 2026, Lucca Spohn considers parents using ChatGPT to obtain a second assessment of their children’s schoolwork. His German-language practice article is not an empirical study of how widespread such cases are. He advocates checking specific objections against explicit criteria; a different AI-generated grade alone does not establish a marking error. [1]

### What the research adds

A study of IELTS writing tasks illustrates this distinction: Detailed criteria improved grading, yet even the best prompt tested achieved only moderate absolute agreement with official scores and assigned systematically lower marks. Similar rankings therefore do not mean identical grades. The findings concern specific models and one assessment format. [2]

An Estonian preprint from 2026, by contrast, reports results within the range of human marking for nationwide trial school-leaving essays. This depended on a carefully developed, rubric-driven process with human oversight. It does not demonstrate that an improvised chatbot conversation achieves the same quality. [3]

### Why this is a sign of change

In our interpretation, this adds a dimension to discussions of tests and examinations: Alongside learners and teachers, **AI-assisted parents emerge as additional actors**. They can confront school assessments with a detailed counterargument. For assessment design, this raises a practical question: How can tasks, criteria and feedback be documented so that reasoned objections can be examined together? The quality of that examination depends on transparent standards, rather than the persuasiveness of a chatbot’s prose.