Generative AI (GenAI) use is increasing throughout higher education and creating some profound changes in teaching practice. As educators, we know our students are actively engaging with GenAI tools with both positive and negative outcomes. They are keen to be taught and to learn more about its application, especially for their future careers. However, less is known about how we, as academic staff, are using the technology.  

This guidance on AI-augmented feedback is focused on possible ways that AI could be used to enhance or smooth human oversight in feedback workflows.

This guidance is not a template for enabling AI evaluation of student work. Separately, it is also not recommending that AI is used to replace human insight, expertise and judgement. Importantly, it does make clear that we should never upload student work to any AI system without explicit consent and forethought.

The guidance itself is a result of a cross-institutional collaboration between King’s, LSE and the University of Southampton. This guidance takes a neutral stance on the use of GenAI, while encouraging staff to engage in continued professional development (CPD). It offers principles and guidelines to help educators responsibly integrate GenAI into assessment and feedback, recognising both its potential and limitations.

By establishing clear protocols, we aim to support safe experimentation that enhances learning outcomes and mitigates risks. GenAI is seen as a tool to support, not replace, human activity, judgement and oversight. Our ultimate goal is to harness its strengths while fostering trust-based relationships between staff and students.

An emerging picture suggests that while many staff are experimenting with GenAI tools, a significant minority are not or are making an active choice not to engage with the technology (Walker, et al 2025). Those who are using GenAI tools are reaping benefits but may be unaware of the implications of its use for feedback and assessment. The initial focus for institutions was to develop guidance for students. Our next task is to develop staff guidelines to help overcome challenges such as ethical dilemmas, legal ramifications, social implications and concerns about trust, data privacy and security. Without proper guidance, the integration of GenAI into educational practices could lead to unintended consequences that may undermine the integrity and quality of education provided. 

This guidance assumes a neutral position on whether staff should or should not use GenAI, but does encourage staff to engage in continued professional development (CPD). In considering its use in practice, educators need to be able to recognise limitations as well as potential. It aims to provide educators with principles and guidelines for responsibly incorporating GenAI into their assessment and feedback processes. In establishing clear protocols and effective practices, we hope to encourage safe experimentation so our approach will enhance educational outcomes while safeguarding against potential risks. 

GenAI technology is rapidly evolving. We acknowledge that experimental work is taking place to automate feedback processes beyond self-grading and feedback tools used with, for example, multiple choice quizzes. It is possible to use large language models (LLMs) to provide extensive feedback on written work to inform automated grading, but this is not within scope of this guidance as it relies on the uploading of student work which requires ethical clearance and student consent.

The focus of this guidance, therefore, is on AI-assisted feedback and marking. We use the term AI-assisted to describe the incorporation of GenAI tools such as Microsoft Co-pilot, ChatGPT, Claude and Gemini in our practice. These elements are limited by design to ensure sufficient human oversight and ensure no breach of existing institutional policy.

AI-assisted feedback and marking requires human oversight. There are a number of ways in which AI assisted feedback and marking can help academic staff to be more effective. These include:

• Enhancing and clarifying feedback (for example to change the tone or reduce jargon) initially authored by human markers.
• Using AI tools to generate draft rubrics that are aligned with module and programme-level outcomes.
• Aligning human-generated feedback notes to specific rubric criteria for structured and personalised responses.
• Digitising hand-written notes or converting text to spoken feedback to enhance readability, clarity and engagement.
• Capturing, transcribing, and summarising verbal feedback discussions or meetings (for example during supervisory meetings or oral examinations).
• Summarising and synthesising individual marker feedback into coherent collective reports.

Whilst GenAI offers efficiency and scalability in feedback provision, its use raises critical concerns about trust in educational relationships (Liu 2025). Some tools and applications promise immediacy, greater precision, consistency and time-saving benefits (AlBadarin et al., 2023). These affordances are leading to high rates of approval amongst staff and students (Barrett and Pack, 2023).

A recent Artificial Intelligence in Higher Education (AIinHE) survey of more than 6,000 students indicated students valued the feedback that GenAI produces which provides immediate, accessible, understandable, positive, objective responses to many questions. However, on trustworthiness they rated human feedback higher than GenAI feedback, noting that academics provide feedback that is relevant, contextualised, personalised and expert.

Survey results showed that participants question the authenticity of AI-generated comments, perceiving them as impersonal and ethically inconsistent, especially when their own use of GenAI may be discouraged. Trust is also influenced by educators' competence with GenAI and their openness to guiding students through its complexities.

The inconsistency of approach across modules, departments and faculties has the potential to create perceptions of inconsistent positions. To preserve trust, Barrett and Pack (2023) advise educators to prioritise transparency. Emerging research may be worth sharing and discussing with students. For example, Nazaretsky et al., (2024) highlight that students often state they prefer human feedback, particularly after disclosure of the feedback provider. However, those who mistakenly assume AI-generated feedback to be human feedback tend to rate it higher.

To develop trust between staff and students we propose the following principles for AI-assisted marking and feedback: 

1. Staff should explain their approach to, and use of, AI-assisted marking and feedback.
2. Reading and assessing student work should primarily be conducted by humans, except in specific circumstances.
3. Human oversight is essential to ensure accuracy and appropriateness.
4. Where AI-assisted feedback has been appropriately used, challenges to academic judgement will not be accepted, in line with standard university regulations.
5. The use of approved/supported university GenAI tools should be used for AI-assisted feedback and marking.
6. GenAI tools may inform the development of marking criteria and assessment rubrics.
7. Academic staff should not use GenAI tools to detect students' use of GenAI, as these tools are notoriously unreliable.

1. Staff are encouraged to use GenAI tools safely, securely, ethically, and appropriately.
2. Staff should use institutionally-approved GenAI tools that comply with GDPR and relevant privacy regulations. These tools provide enterprise-grade assurances of security, data privacy, and training restrictions for Large Language Models (LLMs). If unsupported GenAI tools are used, staff should explain their rationale and obtain appropriate consent
3. Staff should not include any student details or upload work into any GenAI tools without explicit permission. This prevents personal data from being collected, stored, accessed, and shared without consent.
4. Staff should read and follow institutional and local (faculty/department) guidelines to ensure their approach to AI-assisted feedback and marking processes align with agreed practices. Transparency should start with, and extend from academics to students.
5. Staff should reflect critically on how AI-assisted feedback and marking practices influence student perceptions of authenticity, trust, and fairness. Staff are encouraged to actively seek student feedback on these practices.
6. Staff should provide human oversight at every stage of the approach they are adopting. GenAI can generate incorrect responses based on probabilities and biases. While GenAI is good at summarising information and solving known problems, it can produce simplified or incorrect outputs and combine existing knowledge in new ways.
7. Staff should ensure AI-assisted feedback does not independently determine marks or academic judgements. Marking responsibilities remain strictly with human educators.

The following scenarios follow the above guidelines and offer insights into ways that academic staff can use AI transparently and in an assistive capacity, always ensuring human oversight and judgment remain central.

Scenario A – Scaling feedback while maintaining quality

Lecturer A is responsible for marking over 100 essays within a two-week window.

Conscious of the limitations this workload places on the depth of individual feedback, they adopt a hybrid approach using their university’s approved or supported LLM tool, Copilot.

Without ever uploading student work directly, Lecturer A composes an anonymised summary for each student, noting which marking criteria were met and the approximate percentage achieved for each. They input this summary alongside the official rubric into Copilot, prompting it to generate supportive, criterion-referenced feedback. This feedback is then carefully reviewed, adapted, and personalised before being uploaded to the marking platform.

Students are made aware of this process in advance and shown a demonstration, reinforcing transparency and trust.

Scenario B – Harnessing accessibility features

Lecturer B experiences recurring pain from repetitive strain injury, making traditional typing-intensive marking methods challenging.

To reduce physical strain, they have begun using the voice chat functionality of an AI tool while reviewing assignments. They verbally articulate their comments during the review, ensuring all input remains anonymised and free of identifying details. The AI transcribes the spoken reflections in real time and is then prompted to produce a concise summary, isolating one key strength and one or two developmental areas to support student progression.

Students are made aware of this process in advance and shown a demonstration, reinforcing transparency and trust.

Scenario C – Bridging traditional and digital practices

Lecturer C prefers the immediacy and freedom of handwritten annotation when reviewing student work.

Historically, these handwritten notes were then typed into the feedback platform - a time-consuming duplication of effort. More recently, Lecturer C has started photographing their feedback notes and using an approved LLM to transcribe the content. The LLM is prompted to reframe the handwritten comments into a clear, structured feedback format organised into bullet points under three headings: strengths, areas for development and points for action. This preserves the authenticity of the lecturer’s voice while enhancing clarity and readability for students.

Students are made aware of this process in advance and shown a demonstration, reinforcing transparency and trust.

Scenario D – Calibrating grades to align with feedback comments

A new teaching assistant is enthusiastic about giving students a positive and encouraging experience in formative feedback. This backfires when students note a mismatch between the superlatives in the comments and the low indicative grades given, leading to appeals and disgruntlement among students who assumed their grade should have been higher given the highly positive comments.

In a subsequent formative assessment, the teaching assistant deploys a GenAI tool to input the official marking criteria and rubric against their anonymised feedback comments to validate alignment between the grade awarded and the feedback language. This helps nudge the teaching assistant to moderate their language to match the marking criteria more closely.

Students are made aware of this process in advance and shown a demonstration, reinforcing transparency and trust. In each of these vignettes, GenAI has been used in a responsible and transparent manner that demonstrates the benefits of AI-assisted feedback to students. Taking this approach does not compromise trust, academic integrity or fairness. Augmenting the clarity, depth and consistency of feedback and streamlining grading processes will enhance the quality of the education provided to our students.

Guidance first published in July 2025.