Making learning visible in the age of generative AI: reported assessment practices for cognitive engagement and self-regulated learning, and the development and external appraisal of a higher education assessment framework
Year: 2026
Type: article
Abstract: Introduction Generative artificial intelligence (GenAI) makes it harder to infer a student’s own capability from a submitted product, so assessment increasingly needs evidence of how students reason and regulate their learning. This study examined four questions. Which assessment practices do higher education educators publicly report, and how do they describe their implementation? Which cognitive-engagement and self-regulated learning (SRL) processes are these practices intended to elicit? How does a framework derived from these reports fare against separately sampled external documents? Methods Public posts and comments were collected with TinyFish from Reddit (the principal source), Hacker News, Stack Exchange, Lemmy and Mastodon, covering December 2022 to October 2026. Of 29,355 unique records, 13,903 were eligible, and 5,091 reported or discussed at least one of 11 assessment practices (75.2% from Reddit; 3,741 in educator voice; 1,336 threads). Records were coded with transparent dictionaries, which were validated against an independent model-based coder on a hold-out sample (any practice: κ = 0.49; implemented vs. proposed: κ = 0.69). A seven-component framework built from the corpus was then appraised against 40 external documents not used in its construction. Results Detection and sanction (43.2%) and supervised in-class tasks (28.5%) dominated discussion. Detection’s share fell from 56.4% in 2023 to 37.6% in 2026, while oral, process-based and stakes-related practices roughly doubled. Reasoning was the intended process described most often, yet only 2.2% of records mentioned planning, monitoring and reflection or revision together. Several practice–mechanism associations were largely lexical: reflection vocabulary accompanying reflective tasks fell from 94.2% to 5.5% when shared words were masked. Learning-oriented records described monitoring, application and reasoning more often than detection-only records. The framework described all 40 external documents (13 implemented cases, 13 institutional guidance documents, 14 commentaries; mean fit 3.9/5). However, those documents emphasised authentic task design and reflection rather than secured checkpoints, and they prompted an eighth component, student co-design and transparency. Discussion Public discussion favours securing assessment over making learning visible, and no single practice is described as eliciting both cognitive engagement and a connected SRL cycle. This supports combining complementary forms of evidence. The study documents patterns in public reporting and offers a provisional design framework; it does not measure students’ cognition or learning outcomes.
Source: Frontiers in Psychology
Authors: Olusiji Lasekan(Conceptualization, Data curation, Funding acquisition, Methodology, Resources, Validation, Visualization, Writing – original draft, Writing – review & editing), Lucía Ubilla(Formal analysis, Funding acquisition, Investigation, Methodology, Validation, Visualization, Writing – review & editing), Blessy Sarah Mathew(Formal analysis, Investigation, Methodology, Resources, Software, Visualization, Writing – review & editing)
Language: English
Subfield: Education
Field: Social Sciences
Domain: Social Sciences
Keywords: generative AI, higher education assessment, assessment practices, student self-regulation, cognitive engagement, evaluation framework, AI detection, student reflection, student co-design, assessment transparency, Hacker News, Stack Exchange, Mastodon, learning-oriented assessment, metacognition, metacognitive reflection
SDG: Quality education
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7236358325
