What do AI students need from marking criteria in the UK?
Published Apr 22, 2024 · Updated Mar 10, 2026
marking criteriaartificial intelligenceAI students notice quickly when marking criteria feel vague, inconsistently applied, or disconnected from the work they want to do after graduation. The pattern in NSS open-text, analysed using our NSS open-text analysis methodology, shows this is a persistent problem, not a niche complaint.
Across the National Student Survey (NSS, the UK-wide final-year student survey) open-text, the marking criteria theme draws about 13,329 comments, or 3.5% of 385,317, with 87.9% negative and an index of -44.6, indicating widespread concern about how criteria are presented and applied. Within Artificial Intelligence, a subject area used for like-for-like sector comparisons, marking criteria accounts for about 7.5% of comments and is even more negative at about -54.0, while feedback is the largest assessment topic at about 9.0% and -44.8. Students often praise teaching staff at +34.3, a pattern echoed in AI students' views on teaching staff and learning impact, so the issue is less about expertise and more about whether standards are clear, timely, and relevant. These patterns point to practical fixes in fairness, feedback speed, and assessment design.
How do AI students define fair marking?
For AI students, fair marking starts with transparency and consistency. They want grading standards tied to learning outcomes, explained early, and applied the same way across assessors. Use the student voice to surface where rubrics and standards are not landing as intended. Publish annotated exemplars at key grade bands, release criteria with the assessment brief, and run marker calibration with a shared sample bank. When students can see the benchmark before they submit, they spend less time second-guessing the process and more time improving their work.
Why does timely feedback matter in AI?
In AI, feedback matters most when students still have time to use it. Quick, detailed comments help them correct misunderstandings, strengthen technical reasoning, and approach the next assignment with more confidence. Delayed feedback turns a learning tool into a historical record. Staff in AI courses should aim to return feedback within one to two weeks of submission, so students can act on the guidance while the work is still fresh, which mirrors wider evidence on what makes good feedback. Systems that support efficient turnaround and feed-forward keep the learning loop open and improve outcomes.
What does unambiguous marking criteria look like?
Unambiguous criteria leave less room for guesswork. Use checklist-style rubrics with concrete descriptors, weightings, and common error notes. For innovative and technical AI work, exemplars that show "what good looks like" are especially valuable because students often solve the same problem in different ways. Provide a short "how your work was judged" summary with each grade, referencing the rubric lines applied. This supports consistent marking, helps students align their effort with expectations, and reduces remark and query volumes.
How should criteria stay consistent across courses?
Students cannot judge their progress confidently when similar modules appear to use different standards, a concern that also appears in computer science marking-criteria feedback. Standardise criteria across modules where learning outcomes overlap, and explain any intentional differences up front. A brief walk-through of criteria with the assessment brief, a short Q&A, and shared "what we agreed" notes from marker calibration all signal consistency. Track recurring queries in an FAQ on the VLE to close the loop. Consistency across courses turns criteria into a reliable guide rather than a moving target.
How should assessments align with industry requirements?
Assessments feel more credible when students can see a clear line to professional practice. AI students want projects and problem-solving tasks that mirror real-world challenges. Review assignments with external input so they reflect current practice and tooling. Regular consultation with industry partners, alongside explicit mapping of criteria to employability skills, ensures relevance and helps students translate feedback into stronger portfolios, project write-ups, and interview examples.
Where does auto-marking software fall short?
Auto-marking can speed up routine checks, but it can also miss the nuanced, creative, or atypical solutions that characterise AI coursework. Over-reliance creates a risk that published criteria and lived judgement drift apart. Use automated checks for clearly defined elements, but maintain substantive human oversight for complex work so the grade reflects performance against the rubric and sustains trust in the process.
How does lecturer behaviour shape student trust?
Students notice when lecturers are visible, approachable, and transparent about standards. Ambiguities in criteria, especially when they feel arbitrary or inconsistent, increase stress and deter dialogue. Regular calibration, openness about how criteria are interpreted, and proactive offers of clarification during timetabled sessions all build confidence. Teaching staff are often well regarded already, and making the application of standards as visible as the teaching itself strengthens that trust.
What should providers do next?
Start with the changes students can see. Publish annotated exemplars, use unambiguous rubrics, release criteria with the brief, and offer short feed-forward opportunities. Maintain prompt feedback cycles and provide a short "how your work was judged" note with each return. Standardise where learning outcomes overlap, highlight any intentional differences, and maintain a VLE FAQ that addresses recurring student queries. Monitor NSS open-text and internal channels to target modules and cohorts where tone on criteria and feedback trends most negative, so improvement work stays focused.
How Student Voice Analytics helps you
Student Voice Analytics shows how sentiment on marking criteria and assessment moves over time and by cohort, site, or mode, with drill-downs from provider to school, department, and programme. It enables like-for-like comparisons for Artificial Intelligence against the wider sector and by demographics, so you can target cohorts where tone is most negative. You can export concise, anonymised summaries for programme teams and boards with ready-to-use tables and year-on-year movement, making priorities and progress straightforward to share and act on.
Request a walkthrough
Book a free Student Voice Analytics demo
See all-comment coverage, sector benchmarks, and reporting designed for OfS quality and NSS requirements.
- All-comment coverage with HE-tuned taxonomy and sentiment.
- Versioned outputs with TEF-ready reporting.
- Benchmarks and BI-ready exports for boards and Senate.
UK-hosted · No public LLM APIs · Same-day turnaround
Related Entries
- The disconnect on what makes good feedback
- What do AI students say about UK teaching staff and learning impact?
- Do current assessment methods meet computer science students’ needs?
- Do computer science students trust marking criteria?
- NSS open-text analysis methodology for UK HE