Skip to content
PlagiatScanner.de
Higher education practice · template · HSP-16

Assess AI disclosure fairly and consistently

In brief: Assess a disclosure by whether it meets the rule published in advance and makes relevant use traceable, not by length or suspected volume. Minimum fields identify the tool or function, purpose, stage, retained contribution and student's verification. Quality criteria address limitations, source checking and responsibility. Permitted use should not be penalised merely because it was disclosed honestly, and ambiguous requirements must not be tightened retrospectively.

Rubric with minimum and quality criteriaSources reviewed 4 September 2026

Keep disclosure distinct from the quality of the work

A disclosure explains how a system entered the workflow. It does not directly grade the essay, code or research. A concise, precise entry can outperform a lengthy chat export. Apply the announced information criteria; use the assignment's separate rubric for subject quality.

Before work, students need to know which services and functions are permitted, what must be disclosed and how this affects marking. “Declare AI” is insufficient where translation, proofreading, search, code completion and generative drafting are treated differently. Give examples and one authoritative clarification route.

Programme specifications, assessment regulations and institutional policy remain controlling. A rubric cannot create an unapproved penalty or new assessment component. This is an adaptable template, not a universal institutional rule.

Authority asset: minimum and quality rubric

Discipline-open criteria with observable distinctions
Criterion Minimum Good quality Unclear
Identity Service or function and period named Version/model where displayed; unknowns marked Only “used AI”
Purpose Stage and intended assistance described Use and non-use boundaries stated No assignment connection
Retention Retained output or function identifiable Changes and own decisions explained “Inspiration” without example
Verification Subject check named Sources, tests or independent work linked “Everything checked”
Limitations Known uncertainty not hidden Effect on result considered System treated as authority
Compliance Permitted scope and privacy respected Boundary cases clarified in advance Unapproved or uncertain upload

Specify concise, privacy-aware evidence

A useful entry may be a short paragraph or table. It names the service, or function if exact identity is unavailable; purpose; workflow stage; consequential output; what was retained or rejected; and scholarly verification. Full prompts should be required only where the published rule provides a proportionate, lawful reason.

Credentials, protected research records and third-party personal information do not belong in disclosure. If material was uploaded improperly, privacy review is separate from academic quality. Transparency must not trigger further disclosure of sensitive data.

Accept “not displayed” where the interface does not expose a version. Students should not invent a model identifier. A dated export may support context but does not prove a complete use history.

Describe verification with discipline-open verbs

Turn “checked facts” into observable activity: opened sources, repeated the calculation independently, ran tests, compared translation with the original or sought contrary evidence. Appropriate checks vary by discipline and learning outcome. The rubric therefore describes functions rather than mandating one technique.

Strong reflection identifies limits without demanding a ritual negative opinion. A student may explain that an output helped yet invented sources, lost nuance or introduced a code defect. Assess the evidenced response, not a required confession.

Responsibility means that the submitting student owns selection, checking and final work. A tool is not an author. Disclosure does not excuse inaccurate citations or prohibited borrowing, but it makes the route assessable.

Do not punish honest permitted use

If use is allowed for a stage, disclosure alone should not cause an automatic deduction. That would reward concealment. Criteria address completeness, relevance and verification rather than moral assumptions about the amount of use.

Ambiguous or conflicting instructions require a fair clarification process, not retrospective certainty. Record the policy version in force. Treat comparable cases through the same criteria. Writing style and detector scores are not substitute disclosure evidence.

Handle disability-related tools and approved adjustments confidentially through the relevant process. Do not grade disability, language background or access to a paid service. Provide a workable equivalent route.

Calibrate with boundary examples

  1. Prepare anonymous examples of complete, concise and ambiguous disclosure.
  2. Ask markers to score independently and resolve wording-driven differences.
  3. Add discipline examples without introducing new criteria.
  4. Give students a fillable example before work.
  5. Record each decision against a criterion and visible evidence.
  6. Revise ambiguity for the next approved delivery.

Review anonymised marking patterns for inconsistent application, not for a preferred rate of AI use. A difference between disciplines can reflect different tasks; it should prompt interpretation rather than a universal quota. Document changes to training and rubric version.

The assessment design guide frames the tool rule. The group contribution record separates individual work. More rubrics sit in the higher education practice hub. The one foundation route is the guide to verifiable academic evidence.

A final evidence check prevents the rubric from creating false precision. Markers should first highlight statements that are actually present, then record which published criterion each statement supports. They should not infer undisclosed use from prose style, topic choice or familiarity with a tool. Where information is absent, the initial finding is “not evidenced”, not automatically “prohibited”. Only the rule supplied for that assessment determines whether absence has a consequence. If clarification is allowed, the student should be able to identify the relevant workflow stage without being required to reconstruct an unrealistically complete history. This sequence keeps disclosure quality, procedural compliance and the academic quality of the submitted work analytically separate.

Official and scholarly foundations

  1. UNESCO: Guidance for generative AI in education and research – human responsibility, privacy and competence.
  2. QAA: Maintaining quality and standards in the ChatGPT era – transparent assessment expectations.
  3. TEQSA: Assessment reform for the age of AI – institutional evidence and assessment perspective.

Sources reviewed 4 September 2026. The institution's approved rules govern application.