AI in Assessment & Feedback

For University of Reading staff who design, approve, deliver or support assessment and feedback.

Generative AI is changing what students can produce, how they learn, and what assessment can tell us about their achievement. This page brings together the University’s current requirements, practical frameworks and developing work to help staff design assessment and feedback that remain valid, inclusive, transparent and educationally purposeful.

The aim is not to make every assessment resistant to AI. It is to be clear about permitted use, preserve meaningful human judgement, make learning visible and assure achievement across modules and programmes.

 

Start here

When designing or reviewing an assessment:

  1. Decide which GenAI category applies, based on the learning outcomes and assessment conditions.
  2. State clearly what students may and may not do. A category label on its own is not enough.
  3. Use I3 where helpful to describe the forms of GenAI use that are appropriate or inappropriate.
  4. Decide what assessment approach will support a valid judgement about learning, including process evidence, dialogue or supervised activity where appropriate.
  5. Consider assessment type balance and assurance of learning across a programme.

 

1. Set clear GenAI expectations for an assessment

The University uses three categories to communicate whether students may use GenAI for an assessment. The category must be communicated clearly through the assignment brief. If different rules apply to different parts of a task, those differences should be explicitly communicated to students.

Category 1: Independent work

What it means

 

Students must demonstrate their own knowledge and skills without GenAI assistance.

When it is likely to apply

Usually where assessment conditions allow independent performance to be assured, such as an invigilated examination, viva, supervised test or observed practical activity.

What staff need to do

  • Check that GenAI use can be restricted to a reasonable degree in the assessment conditions.
  • Explain permitted assistive or accessibility technologies separately.
  • Assess meaningful application, analysis or judgement rather than recall alone.

Category 2: AI-supported

What it means

Students may choose to use GenAI to support their learning and the development of their work. For example, they might use it to brainstorm, ask questions, test an outline, plan a structure or refine work they have already produced. GenAI use is not itself assessed and the final submission must remain the student’s own work.

What staff need to do

  • State the boundaries of permitted use in language that makes sense for the discipline and task.
  • Identify aspects for which GenAI must not be used because they are central to the learning outcomes.
  • Require students to acknowledge how they used GenAI.
  • Do not make optional GenAI use a hidden expectation or disadvantage students who choose not to use it.

Category 3: AI-integrated

What it means

Purposeful engagement with GenAI, or critical evaluation of its outputs, forms part of the learning being assessed. AI literacy is therefore visible in the assessment criteria.

Students should have an appropriate route to demonstrate this through:

  • active use – creating prompts, iterating and explaining why outputs were accepted, adapted or rejected; or
  • critical review – analysing provided or publicly available outputs where students prefer not to generate content themselves for ethical or other legitimate reasons.

What staff need to do

  • Align the GenAI activity explicitly with the learning outcomes.
  • Assess judgement, evaluation and disciplinary application, not simply tool use or prompt volume.
  • Explain what evidence students should retain or submit, keeping the burden proportionate.
  • Provide an equitable route for students who should not be required to use a particular commercial tool.

Important: Do not rely on the category name alone. Students need assessment-specific examples of permitted, prohibited and required activity. Where no guidance is provided, students are currently instructed to assume Category 1.

Category guidance

 

2. Use I3 to explain how GenAI may be used

 

How the category system and I3 work together

 

Category system 

 I3 framework 

Is GenAI permitted for this assessment?

How might GenAI be used, and what judgement is required?

Supports assessment rules

Supports learning conversations

Permission-focused

Judgement-focused

I3 was developed at Reading as a descriptive framework for forms of GenAI use already appearing in student and staff practice:

Inquire

Use GenAI to explain concepts, summarise material, explore an unfamiliar topic or support early sense-making.

Improve

Use GenAI to refine work that the person has already produced, for example its grammar, tone, clarity, structure, expression or length.

Boundary: Improve means improving something that already exists. If writing, expression, disciplinary voice or argument construction are themselves learning outcomes, this assistance may be inappropriate.

Integrate

Work iteratively with GenAI to compare ideas, test interpretations, synthesise material or "co-create" an output.

 

Interrogate throughout

Interrogation runs across all three domains. It includes checking accuracy, sources, bias, fabrications, altered meaning and disciplinary fit, and asking whether the person remains in the loop and in the driver’s seat.

 

I3 is not a ladder from “low” to “high” AI use. Inquire, Improve and Integrate are different forms of use with different educational benefits and risks.

 

The I3 framework is available to download here

 

How to use I3 in assessment design

  • Start from the learning outcomes - what do you want your students to be able to do on their own? what aspects are ok for students to use GenAI to help develop their understanding?
  • Describe the permitted forms of GenAI use and their boundaries in the assignment brief using I3 as a framework, e.g. "you may use GenAI to inquire about X, Y and Z, but do not use it to improve your written expression, which must be your own".
  • Give discipline-specific examples rather than generic permission.
  • Describe why it is important that students interrogate GenAI outputs by referring to some of the risks illustrated in I3, and if appropriate, how they can demonstrate interrogation of outputs within the assessment itself.

Explore the I3 poster and worked examples.

 

3. Use GenAI responsibly in marking and feedback

GenAI may have value in bounded feedback activities, for example by helping staff make human-generated comments clearer, more accessible, more consistent or easier to act on. It must not replace academic judgement, determine marks or generate evaluative feedback independently.

Six principles

1. Human academic judgement cannot be delegated to GenAI

Marks, evaluative decisions and responsibility for feedback remain with appropriately qualified staff.

2. Fairness, validity and reliability must be maintained

Staff remain responsible for accuracy, bias, disciplinary fit, consistency and alignment with assessment criteria.

3. Student data and intellectual property must be protected

Student work must not be uploaded into GenAI tools or AI detection tools under the current interim approach.

4. Transparency should go beyond simple disclosure

Students should understand why GenAI was used, what role it played, what it did not do and how human judgement remained central.

5. Feedback should support engagement, dialogue and action

GenAI should help students make sense of and act on feedback, not simply increase its volume or displace its relational and educational value.

6. Innovation should be supported through responsible experimentation

Novel, less established or higher-risk uses should be discussed before a pilot so appropriate expertise and safeguards can be brought in early and learning can inform future guidance.

Current practical boundaries

  • GenAI must not determine marks.
  • GenAI must not independently produce evaluative feedback.
  • Bounded assistance may be appropriate only after human academic judgement has been made and where all outputs are fully reviewed.
  • Student work, personal data, sensitive data and restricted information must not be entered into GenAI tools.
  • Student work should not be uploaded into AI detection tools.
  • Staff should use institutionally supported tools and follow DTS and IMPS guidance.

For novel or higher-risk proposals, email cqsd-tandl@reading.ac.uk before beginning a pilot.

 

  • Read the fuller use of GenAI in marking & feedback guidance (available soon)

 

4. Explore and pilot the Assessment Toolkit

About the QAA Collaborative Enhancement Project

The University of Reading is a partner in Enhancing Assessment and Programme Design in the AI Era, a QAA-funded Collaborative Enhancement Project led by SOAS, University of London.

The project brings together partner institutions to pilot, evaluate and refine a practical Assessment Toolkit. It is investigating how the toolkit can support programme and module teams to design assessment that is authentic, inclusive, AI-aware and capable of providing credible assurance of learning across online, blended and face-to-face contexts.

Evaluation and iterative improvement are central to the project. Evidence gathered across the partner institutions will help refine the toolkit and inform case studies and resources for the wider higher education sector.

 

About the Assessment Toolkit

The toolkit offers structured prompts, examples and process-rich assessment approaches to help programme and module teams rethink assessment in the AI era.

It helps move the conversation from “Can students use AI to do this?” to “How can assessment design make learning, judgement and development more visible?”

The Reading version currently covers 27 assessment types across six themes. Each assessment type brings together an overview, SWOT analysis, GenAI considerations, Graduate Attribute links, example briefs, rubrics, formative checkpoints and ideas for feedback and process evidence.

The toolkit is not a universal answer to the challenges created by GenAI. It is a structured conversation starter for reviewing whether an assessment remains valid, inclusive, transparent, sustainable and workable in its context.

 

Use the toolkit to:

  • help review existing assessments;
  • explore alternative assessment types;
  • add formative checkpoints, dialogue or proportionate process evidence;
  • clarify GenAI expectations and what students need to demonstrate;
  • support a programme-level conversation about assessment balance and assurance.

 

Explore and use the toolkit

A SharePoint web version is currently under development for University of Reading colleagues who would like to explore the toolkit further - please check back soon for updates.

Take part in the research

Colleagues who use or review the toolkit may also choose to contribute to its evaluation. Opportunities may include completing a survey, taking part in an interview, or allowing the project team to review relevant assessment-design materials.

Research participation is optional. The Reading project team will explain the available activities, time commitment and relevant consent and data-handling arrangements before you decide whether to take part.

Find out more

 

5. Programme-level assurance in the AI era

Assurance of learning cannot be achieved by trying to make every individual assessment resistant to GenAI. A more sustainable and inclusive approach considers how assessment across a programme provides a credible, varied body of evidence that students have achieved the programme learning outcomes.

The University is developing guidance to help programme teams:

  • identify the most important points at which achievement needs to be assured;
  • consider the balance of supervised, partially observable and independent assessment activity;
  • use multiple and contextualised sources of evidence rather than relying on one final product;
  • make learning processes, judgement and development more visible;
  • balance assurance with inclusion, assessment for learning, student wellbeing and staff workload;
  • use programme mapping and collegial discussion to identify priorities for redesign.

Guidance is in development. Further content and practical tools will be added here as the University’s institutional position and programme-level approach develop.

In the meantime, programme teams can use the Assessment Toolkit as a discussion prompt and draw on existing programme-level assessment guidance. For support with a programme-level assessment conversation, email cqsd-tandl@reading.ac.uk.

Back to the top of this page