Abstract:

This intervention study explored how AI-guided feedback can support Finnish students preparing for writing tasks on their English-language matriculation examination. The writing section of the exam accounts for 99 points out of the exam's 299 total points. It also requires students to produce texts of a maximum 1,300 characters across five text types: descriptive, narrative, instructive, opinion-based, and reflective. The official assessment rubric for marking the exam, prioritizes communicative effectiveness ("Viestinnällisyys"), supported by four additional criteria: content quality, text structure, linguistic range, and grammatical accuracy.

Twenty Finnish secondary students completed past-paper writing tasks under examination conditions across multiple text types. Each student's work was numerically assessed from four complementary perspectives: student self-evaluation (developing metacognitive awareness), classroom teacher assessment (applying examination expertise), university writing instructor evaluation (providing higher education perspective), and AI-generated scoring (offering consistent, detailed analysis).

The AI system analyzed each student's complete writing portfolio to identify recurring strengths and weaknesses based on the seven texts they had produced. AI focused particularly on issues affecting communicative effectiveness—the primary assessment criterion in the Finnish system. Students then received personalized feedback highlighting specific areas for improvement that could transfer across multiple text types and could yield cross-genre improvements. After receiving this targeted feedback, students completed an additional writing task. This allowed us to measure whether AI-guided instruction led to sustained improvement in writing quality.

The study examined how different assessment perspectives—student, teacher, university instructor, and AI—complemented one another in supporting student learning, and whether AI-identified patterns aligned with expert human judgment. The findings offer practical insights for educators seeking to integrate AI feedback tools that support rather than replace human assessment in high-stakes examination preparation.

Keywords:

AI feedback, writing intervention, human-human feedback, calibrating formative assessment, matriculation examination, collaborative assessment, automated writing evaluation.