As generative AI reshapes higher education, institutions increasingly require students to declare their use of it. But generic, binary declarations a single checkbox reading "I used GenAI" cannot capture how these tools are actually applied across different academic tasks. This work contributes a design artefact and a position: a framework of two task-specific declaration structures, one for writing-focused activities and one for coding assessments, built on an existing taxonomy of GenAI usage. By categorising AI use across distinct cognitive and developmental stages structural planning versus content generation, code improvement versus code generation the framework prompts students to reflect on their own learning and clarifies the boundary between acceptable assistance and academic misconduct.
Many universities now require a declaration of GenAI use, but the typical form is a single yes/no question. When a declaration cannot distinguish brainstorming from ghost-writing, it fails three times over: as an integrity safeguard, as a reflective prompt for the student, and as a usable signal for the educator adapting their teaching.
The evidence is blunt. In one business-school study, 74% of students left a mandatory declaration blank interviews attributed this to fear of academic consequences and to perceiving the form as an admission of plagiarism rather than a neutral act of transparency. A coarse instrument can undermine the very honesty it was meant to develop.
The framework builds on an established taxonomy of GenAI use in computing education, which characterises AI-mediated work in terms of nine aspects. That taxonomy was designed as an analytic instrument a way to compare interventions across subject areas. This work translates those abstract aspects into concrete, assessment-facing categories that students and instructors can use directly when declaring GenAI use in specific coursework.
The contribution sits deliberately between two existing levels. Policy-level instruments such as the AI Assessment Scale operate at the educator's design decision what level of assistance is permitted. The taxonomy operates at the analytic level what kinds of activity occur. This framework operates at the student's declaration level how GenAI was actually used within a task whose permitted scope has already been set.
Five principles shift the declaration from a compliance artefact into a reflective instrument.
Six activity categories for writing-focused work a short literature review, a professional report. For each, the student states whether GenAI was used, the extent of use, and a short explanation with example prompts. The extent column below shows the scale a student fills in; the lit segments are illustrative entries.
A parallel structure of five categories mirrors the phases of software development. The same extent scale and example-prompt pattern apply. The structure lets both sides distinguish GenAI as a learning aid understanding an error message from GenAI as an authoring tool generating core solution code.
The goal is not a precise measurement instrument but a structured way to describe use and open more nuanced conversations about acceptable assistance. One three-point scale is shared across both forms.
The same two structures support very different institutional stances, because the declaration describes practice rather than prescribing it. It sits below policy instruments, supplying the granular disclosure layer those instruments do not specify.
A
CHECKBOX
CANNOT
DESCRIBE
A
WORKFLOW.