Before learners start: make AI-use rules specific to the assessment

By the EduMark team

Give learners a short AI-use statement alongside each assessment brief, before they begin. Specify five things: permitted help, prohibited assistance, what they must demonstrate independently, how to acknowledge permitted use and where to ask questions. Replace vague instructions such as “use AI responsibly” with actions learners can recognise.

Acknowledging AI use does not make prohibited assistance acceptable. Permission and disclosure are separate questions: first establish whether the help is allowed, then explain how learners should record it.

Start with the rules that govern this assessment

Check the qualification specification, assessment instructions and institutional policy before drafting permissions. NCFE’s assessment guidance distinguishes between assessment types and directs providers to qualification-specific instructions. Internet access does not, by itself, establish permission to use AI. Do not extend a coursework permission to an examination or controlled task. ([ncfe.org.uk](https://www.ncfe.org.uk/customer-and-learner-support/assessment-support/use-of-artificial-intelligence-ai/))

In its higher education context, King’s College London’s student guidance says learners should receive clear guidance about appropriate AI use for each assessment. Its suggested levels can be adjusted by programme and module leaders; the page also explains that policy takes precedence over guidance. These are institutional arrangements, not a universal permission scale for schools, colleges or awarding organisations. ([kcl.ac.uk](https://www.kcl.ac.uk/about/strategy/learning-and-teaching/ai-guidance/student-guidance))

Describe the boundary using the assessed skill

Write down what the learner must demonstrate before deciding which assistance is acceptable. For a supplier-comparison task, that might be selecting relevant evidence, weighing competing claims and justifying a recommendation. Then describe permitted and prohibited actions against those skills.

NCFE’s definition of AI misuse includes unauthorised assistance with tasks directly related to assessment objectives. Its example is instructive: if essay structure is being assessed, AI must not supply that structure. The boundary depends on what is being assessed, not simply which application is used. ([ncfe.org.uk](https://www.ncfe.org.uk/customer-and-learner-support/assessment-support/use-of-artificial-intelligence-ai/ai-misuse-definition/))

For your brief, name the action rather than relying on a tool label:

• “Identify possible punctuation errors” is more precise than “grammar tools allowed”.
• “Do not generate your comparison or recommendation” is clearer than “do not overuse AI”.
• Specify whether the restriction covers preparation, drafting, editing and the final submission.

Editable AI-use statement

Copy and adapt the following wording. Replace every bracketed field, remove irrelevant options and obtain the appropriate approval. This is a drafting aid, not a substitute for governing rules.

AI use for [assessment title and version]

What you must demonstrate: You must independently [identify the knowledge, reasoning or practical skills being assessed]. This statement applies during [preparation, drafting, editing and submission, as applicable].

Permitted help: You may use [approved tools or functions] only to [specific actions]. For example, [one permitted action]. You remain responsible for checking the output.

Not permitted: You must not use AI to [specific actions]. This includes [one prohibited example], even if you acknowledge it.

Acknowledgement: Record permitted use in [location and required format]. Include [required tool details, purpose and affected sections]. Keep [specified prompts, outputs or other records] and submit them [where and when required]. If you used no AI, [state the required declaration].

Questions and support: Ask [named role and contact route] before using help that is not clearly covered. Contact [support route] about approved adjustments or access needs; do not assume this statement changes an existing support arrangement.

Fictional example: grammar feedback or rewritten reasoning?

Imagine an assignment asking learners to compare two supplier proposals and justify a choice. In this fictional brief, subject to the applicable rules, staff permit grammar feedback on the learner’s own draft but prohibit AI-generated analysis, evaluation and recommendations.

Add an instruction for unexpected output: if a tool supplies rewritten reasoning despite a restricted request, do not incorporate it; ask the tutor what to do next. If language accuracy itself is assessed, review whether grammar assistance is appropriate rather than copying this example.

Show learners what an acknowledgement should contain

Follow the applicable format rather than inventing one standard for every setting. NCFE directs learners to acknowledgement requirements associated with JCQ guidance; King’s uses coursework declarations and does not treat generative AI as an authoritative reference-list source. ([ncfe.org.uk](https://www.ncfe.org.uk/customer-and-learner-support/assessment-support/use-of-artificial-intelligence-ai/))

Where your rules permit this format, a useful acknowledgement for the fictional task might read: “I used [approved tool] to identify possible grammar errors in my draft. I reviewed the suggestions and made corrections myself. It did not generate my comparison or recommendation.” Require any additional records specified by your assessment rules.

Staff review before release

• Does the statement match current qualification and institutional requirements?
• Are permitted actions and independent skills explicit?
• Does it cover embedded writing assistance as well as standalone chatbots?
• Are acknowledgement, record-keeping and question routes clear?
• Have approved adjustments and access needs been checked?
• Can learners explain the boundary using the worked example?

Check platform settings separately from the wording. Aurora, EduMark’s AI assistant, is part of EduMark’s beta AI provision, with human review throughout. Its public page describes human approval of marking and assessment boundaries, including being switched off in protected assessments. A written statement does not itself configure those controls or enforce restrictions across other tools. Review the learner instructions and the actual assessment setup together. ([edumark.ai](https://www.edumark.ai/aurora))