Preregistration: Do AI chatbots ground their grading of student essays in the text? A feasibility pilot
Preregistration of a feasibility pilot that tests how three AI chatbots (ChatGPT, Gemini, Grok) grade three German student essays from the GENIUS Keylog Corpus: whether their quotes occur in the essay, how stable the grades are over three repetitions, and whether they detect one inserted error per essay. 54 grading runs, done by hand. Registered before any run. The study state is frozen at commit bfcd872 of github.com/LAKITALKS/Inspektorium. The Markdown file is identical to studies/school-pilot-2026/prereg/preregistration.md (SHA-256 760dbbb3593f76bbfcf13d2d45891830a1212b6371823f0fa27ee40752c864d0).
Authors
- Lazaros Varvatis
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-11
- DOI
- https://doi.org/10.5281/zenodo.24131847
- Primary Topic
- Artificial Intelligence in Education
- Type
- article
- Field-Weighted Citation Impact
- 0.00