Why feedback matters (and matters more) for open assessments

The ability and availability of generative AI tools — which can complete many of our assessments, including quizzes, assignments, reports, reflections and presentations, potentially with minimal student thinking or understanding — means that these assessments can no longer be treated as reliable evidence of students’ capabilities when completed in unsupervised environments. Completion of these tasks and products are, though, still valuable as ways to develop students’ thinking and capabilities, especially when supported by the use of contemporary technologies as they are used in the discipline. For this to happen, students also need feedback to guide them and help them improve. Increasingly, this will include feedback on the quality of their judgements, choices, and thoughtful use of tools such as generative AI alongside traditional academic skills.

The distinction between secure assessments (Lane 1), which provide trustworthy assurance of learning, and open assessments (Lane 2), which are designed primarily for learning, is a clarification of purpose and ensures we can maintain and improve standards. As a result, this categorisation fundamentally reshapes how we should approach and think about marking and its purpose. The shift does not ask markers of open assessment to make weaker judgements, but different ones, because the assurance function has been deliberately relocated to secure assessments elsewhere in the program.

The two lane approach is not an interim response to AI. It recognises that we can only reliably control assessment conditions in supervised environments, and that outside those settings we must design for learning rather than the policing of how AI has been used.

The assessment product is no longer a reliable proxy

This article builds on a recent Needed Now in Learning and Teaching post, What are we going to do about marking now?, which included the statement “the artefact is no longer a reliable witness”. The vast majority (~75%) of our assessments at Sydney are categorised as open, as they should be. Just under half (~48%) of all marks are given to assessments in the open “production and creation” category – written work, presentations, portfolio, journals, creative work and performance. It is pleasing and clear that we value these open assessment types. They also play an important role in developing many of the skills valued by employers, including creativity, critical thinking, and reflection. However, those employers are increasingly expecting our graduates to work effectively with AI tools to generate options, interpret and critique perspectives, iterate purpose and content, and produce polished content quickly. To help our students develop these capabilities, we need to concentrate on the processes they use, including the choice and ways they use tools, not just on the final product.

We’ve long known that we never assess learning directly. We make professional judgements sometimes on direct observation but often based on proxies, drawing on available evidence of understanding and quality (Sadler, 2013). As Boud and colleagues note (2018), we often rely on contextual signals that stand in for learning rather than observing learning itself. What AI has done is expose how fragile those inferences can be when persuasive features such as fluency, structure, and disciplinary language are easy to generate on demand without the student necessarily exercising disciplinary judgement. The “4P framework” (Fawns et al. 2026) focusses on product, process, performance, and practice in the design of balanced assessment.

The risk is turning marking into an exercise in AI detection and suspicion. The two lane model deliberately avoids this trap. As AI tools are very good at producing polished content, marking needs to focus on how well ideas are connected and claims are justified, rather than on surface features of production. In open assessments, markers are not being asked to reconstruct how learning occurred. Their task is to judge the quality of reasoning, decisions, and judgement evident in the work itself and, importantly, to help students see how to improve. It is not their task to infer that a student has or has not learned based on how fluent, generic, or ‘AI‑sounding’ a submission appears.

Open assessments as assessment for learning

Under the Sydney Assessment Framework, open (lane 2) assessments are not designed to independently assure learning outcomes. That responsibility sits with secure (lane 1) assessments positioned at key points across the program. This frees us to concentrate completely on learning and development when we design and mark open assessments. These exist to:

• surface students’ developing understanding,
• normalise the use of tools (including AI) as part of learning and disciplinary practice,
• provide students with actionable feedback on their reasoning, judgement, and use of tools,
• motivate engagement with disciplinary concepts, skills, and dispositions
• signal where work is important by being graded and counting towards the final mark, and
• prepare students for future secure assessments and life after the university.

Once this purpose is clear, marking shifts to thinking about how marks and feedback help students develop a more accurate internal sense of quality, their learning progress, and the gaps in their capabilities. This shift is critical. It reframes setting, marking and providing feedback on open assessments as a developmental act.

The validity of open assessment lies in the feedback we give

Feedback is not an incidental secondary feature of open assessments. It is the primary mechanism through which they work as assessments for learning. The much cited work by Sadler (1989, 2010) shows that feedback supports improvement only when students can:

• recognise what quality looks like,
• compare their work against that standard, and
• take action to close the gap.

In the context of generative AI, this includes learning to judge when, how, and to what extent AI use meet disciplinary expectations. Open assessments create the conditions for this iterative cycle. Feedback arrives early enough to be used rather than after the learning opportunity has passed. Only secure assessments carry the burden of assurance, meaning that open assessments can function as spaces where feedback has time to do its work.

Sometimes, this will feel like “marking the AI” and this may be understandably uncomfortable (or even depressing). It is important to mark the quality of the work received against the disciplinary standards – and not to try to detect or judge its provenance. We cannot reliably know whether a student has used AI verbatim, used AI for help, completed the work without AI, or anything in between. If the student has just uncritically pasted straight from their AI tool and it is poor quality, then the feedback can help them learn what the disciplinary standards are so that they can develop an eye for quality. If the work is completely their own or even a result of an attempt to use AI effectively, the feedback can similarly help them improve. The focus and feedback remains on the quality of reasoning and judgement visible in the submission, not on the provenance of the text.

Feedback literacy and learning how to improve

Open assessments are powerful in helping students develop their capacity to interpret feedback, evaluate quality, and use standards independently — that is, to develop what Carless & Boud (2018), Molloy et al. (2020) and others term feedback literacy. Ideally, they:

• expose students repeatedly to criteria and standards, including through rubrics,
• encourage self‑evaluation and comparison to these, and
• support students’ capacities to judge quality themselves rather than just complete tasks and chase marks.

Importantly, these purposes also include learning to critically engage with AI, rather than treating its use as inherently good or bad. From this perspective, marking in lane 2 is not just about the overall quality of a piece of work, but about changing how students understand and judge quality itself. As they enter workplaces where AI use is increasingly expected, these capacities are central graduate outcomes.

Evaluative judgement and learning to judge quality yourself

Tiffin’s recent Needed Now article places strong emphasis on evaluative judgement – the capacity to make informed decisions about the quality of work, one’s own and others’ (Boud et al., 2018; Tai et al., 2018). Evaluative judgement does not emerge automatically. It must be developed deliberately through:

• exposure to exemplars,
• shared language about standards,
• dialogue around criteria, and
• high-quality feedback (Henderson, Ryan & Phillips, 2019).

Open assessments provide the mechanism for students to build this judgement. Marking and feedback, in turn, contribute directly to students’ ability to produce work and to recognise and critique its quality. For subjects where students’ capabilities are going to be assured through a secure (lane 1) assessment, providing opportunities to practice and test their understanding through open assessments is key for developing evaluative judgement – i.e. being responsible for knowing how they are going. For subjects where production and creation dominate, students need to be able to judge how good their own work is and how it reflects their own standards. This will help students develop the ability to judge quality, an important characteristic when working with AI as to not be seduced or content to just accept its output.

If you are a marker or interested in more concrete guidance on how to apply these principles in practice, the companion guide outlines how to approach marking, rubrics, and feedback in open assessments.

Marking and feedback for open assessments in practice

Marking open assessments involves a shift in emphasis, not a loss of rigour. Importantly, all assessments including their marking and feedback need to be aligned with the learning outcomes. Assessment criteria and hence feedback should concentrate on what the students are expected to be able to do and the development and help they need to get there. There is no burden of assuring that learning – or policing the use of AI. In open assessments, if we build in strong constructive alignment, the right opportunities for feedback, and assessments, we incentivise students to develop the responsibility for learning that will be assured in subsequent secure assessments. These connections should be deliberate and clearly signposted to students.

Open assessment marking and feedback therefore considers:

• Looking beyond polish. AI‑supported fluency should increasingly be expected, so the focus should shift to conceptual connection, reasoning, choices made, and responsiveness to context (Bruner, 1976; Marton & Säljö, 1976), including how students shape, adapt, and evaluate AI‑generated material.
• Using rubric criteria as learning tools. Criteria should make standards visible and discussable, not just operationalise grading (Sadler, 2013). Where appropriate, criteria can also signal expectations about judicious and critical use of AI within the discipline.
• Focusing on growth and development. Open assessments are ‘assessments for learning’, so try to gauge and encourage the process of becoming a disciplinary expert. Design rubrics to reward reflection, iteration, thinking, processing, and growth, as aligned to the learning outcomes of the unit and program.
• Treating feedback as preparatory. Feedback should deliberately point forward to future tasks, including secure assessments, rather than act as a summative verdict.
• Keeping the program view in mind. Open assessments do not need to do everything. Their strength lies in what they contribute before assurance is required. They should, however, be constructively aligned with the program (i.e. course or major) to maximise relevance and motivation. This should include alignment with how AI is used in disciplinary and professional practice. For example, feedback might explicitly comment on how effectively a student has prioritised ideas, justified decisions, or adapted AI‑generated material to the task context.
• Help students see the alignment. Beyond assuring ourself that our programs are aligned, we also need to help our students see that effort in open assessments will be rewarded in meeting the program learning outcomes and in performing well in secure assessment.

This shift in emphasis for marking also places greater importance on clear, judgement‑focused rubric criteria and opportunities for marker calibration, so that expectations about quality are shared and visible.”

A cultural shift

Generative AI has not broken assessment, but it has forced us to confront assumptions we have long tolerated about what products and artefacts can legitimately tell us. The two lane model responds by shifting responsibility for assurance across programs and by re‑focussing feedback and the development of evaluative judgement as the mechanism of learning within open assessments.

In turn, we need to help students navigate this approach so that they maximise opportunities for learning, feedback and improvement before they meet assurance of learning tasks.

Related resource

• Guide for Markers of Open Assessments

References

Boud, D., Ajjawi, R., Dawson, P., & Tai, J. (Eds.). (2018). Developing evaluative judgement in higher education: Assessment for knowing and producing quality work. Routledge.

Bruner, J. S. (1976). The process of education (Revised ed.). Harvard University Press.

Carless, D., & Boud, D. (2018). The development of student feedback literacy: Enabling uptake of feedback. Assessment & Evaluation in Higher Education, 43(8), 1315–1325. https://doi.org/10.1080/02602938.2018.1463354

Fawns, T., Boud, D., & Dawson, P. (2026). Identifying what our students have learned: a framework for practical assessment validation. Assessment & Evaluation in Higher Education, 1–17. https://doi.org/10.1080/02602938.2026.2620053

Henderson, M., Ryan, T., & Phillips, M. (2019). The challenges of feedback in higher education. Assessment & Evaluation in Higher Education, 44(8), 1237–1252. https://doi.org/10.1080/02602938.2019.1599815

Marton, F., & Säljö, R. (1976). On qualitative differences in learning: I—Outcome and process. British Journal of Educational Psychology, 46(1), 4–11. https://doi.org/10.1111/j.2044-8279.1976.tb02980.x

Molloy, E., Boud, D., & Henderson, M. (2020). Developing a learning centred framework for feedback literacy. Assessment & Evaluation in Higher Education, 45(4), 527–540. https://doi.org/10.1080/02602938.2019.1667955

Sadler, D. R. (1989). Formative assessment and the design of instructional systems. Instructional Science, 18(2), 119–144. https://doi.org/10.1007/BF00117714

Sadler, D. R. (2010). Beyond feedback: Developing student capability in complex appraisal. Assessment & Evaluation in Higher Education, 35(5), 535–550. https://doi.org/10.1080/02602930903541015

Sadler, D. R. (2013). Assuring academic achievement standards: from moderation to calibration. Assessment in Education: Principles, Policy & Practice, 20(1), 5–19. https://doi.org/10.1080/0969594X.2012.714742

Tai, J., Ajjawi, R., Boud, D., Dawson, P., & Panadero, E. (2018). Developing evaluative judgement: Enabling students to make decisions about the quality of work. Higher Education, 76(3), 467–481. https://doi.org/10.1007/s10734-017-0220-3