Interactive learning material · M4-T2-MLC1

Designing Transparent Evaluation Criteria and Rubrics with AI

From learning outcomes to expectations that learners understand

Author: Zsolt LengyeliTStudy Hungary Training and Research Centre
Reviewer: Mária HartyányiiTStudy Hungary Training and Research Centre

Learning outcomes lead to observable evaluation criteria and then to performance levels. — Learning outcome → evaluation criteria → performance levels

Learning objective

Clear criteria show learners what good work looks like before they start, and make the teacher’s judgement consistent and explainable afterwards. This material shows how AI can support teachers in designing evaluation criteria and rubrics that are observable, based on learning outcomes and understandable to learners.

By the end of this material, you will be able to

Derive observable evaluation criteria from learning outcomes.
Use AI to draft performance levels and matching descriptors.
Check the draft for fairness and transparency.
Adapt the criteria to your subject, your learners’ needs and the vocational context, while keeping the final responsibility.

Introduction

Why plan assessment in advance?

In vocational education, much of assessment concerns learning products and performances: a weld, an installed circuit, a conversation with a customer or a loaf of bread.

They are hard to judge consistently if the criteria exist only in the teacher’s head. Innovative assessment methods follow two simple rules: the criteria must be planned in advance, and learners must know them as early as the preparation phase.

Writing good criteria and descriptors is demanding and time-consuming work. A language model can propose criteria, draft descriptions for several performance levels in minutes, rewrite them in learner-friendly language and point out weaknesses in a draft.

01

Objectivity

The result should not depend on the person who assesses.

02

Reliability

The assessment tool should produce consistent results.

03

Validity

We should measure what we actually intend to measure.

Evaluation criteria

Start from the learning outcome

A learning outcome states what the learner knows, understands and is able to do after a learning process, in terms of knowledge, skills, and responsibility and autonomy (EQF).

Example: “The learner installs a two-way lighting circuit according to the wiring diagram and the safety regulations.”

From this, we derive features that can be checked in the finished product or during the work.

Each learning outcome is linked to several evaluation criteria, and each criterion to several performance levels. — Each learning outcome is broken down into observable criteria, and each criterion into performance levels.

What do we observe when a two-way lighting circuit is installed?

Correctness of connections.
Safe working practice.
Neat cable routing and workmanship.
Testing before switching on.
Documentation.

Evaluation criteria

What makes a good criterion?

01

Aligned with the curriculum

It follows from at least one learning outcome. Nothing is assessed that was not taught or included in the learning objectives.

02

Observable

It describes something that can be seen, measured or heard in the product or performance, not an inner state of the learner, such as “shows interest”.

03

Distinct

Criteria do not overlap, so the same mistake is not penalised twice.

04

Complete

The criteria cover the essential elements of the outcome, including safety and professional standards.

05

Understandable

A learner at the given level understands what is expected.

These requirements follow Brookhart’s (2013) guidance on rubrics, and also serve objectivity, reliability and validity.

Drafting with AI · The format

Choose an appropriate assessment format

The format depends on the purpose of the assessment. A rubric makes expectations visible through criteria and descriptions of performance quality.

An analytic assessment matrix gives the most information for learning: criteria are in the rows, performance levels in the columns. AI can help to fill in the table.

Assessment matrix with four criteria in rows and four performance levels in columns, with the safety row highlighted. — An analytic rubric: each criterion has its own performance descriptors.

Checklist

Records whether each necessary element is present: yes or no. It suits safety checks and procedural steps.

Analytic rubric

Assesses each criterion separately on several performance levels. It gives detailed information on strengths and areas for improvement.

Holistic rubric

Describes overall quality on a single scale, for example rating an essay from 1 to 5. It is quick and simple, but does not explain individual weaknesses.

Single-point rubric

A practical format for formative assessment. It describes the expected level and leaves space for comments on what falls short of it or goes beyond it.

Drafting with AI · A vocational example

One project, two assessment methods

The park construction project shows how different assessment formats can be chosen for the design and implementation stages.

Design · Single-point rubric

The mentor supports the next steps

The mentor evaluated the learners’ plans using a single-point rubric. Comments and advice written on the plans helped to guide further work.

The learners’ plan for a park construction project. — Park construction: design. Source: Sára Ekert, agricultural engineer teacher and mentor.

Implementation · Assessment matrix

The committee considers several criteria

The construction was evaluated by the mentor and two specialist teachers. Criteria included the accuracy of the dimensions, compliance with the plan and the safety of the equipment.

The matrix used two levels: outstandingly performed / performed.

The completed park construction project. — Park construction: implementation. Source: Sára Ekert, agricultural engineer teacher and mentor.

Drafting with AI · The prompt

Give the AI the information it needs

The five-step framework: task, context, references, evaluation, iteration. Adding one good and one weak example of a descriptor also helps to clarify expectations.

Task: tell the AI what to create.
Context: explain who it is for: age, year and prior knowledge.
References: provide the learning outcomes, exam requirements, task description and, if useful, an earlier rubric.
Evaluation: critically review the draft.
Iteration: refine it through more precise requests.

Example prompt for a rubric draft

“You are an experienced vocational teacher of electricians. Create an analytic rubric for a practical task: second-year apprentices (aged 16–17) install a two-way lighting circuit on a training board. Learning outcomes: [paste]. Criteria: correctness of connections, safe working practice, cable routing and workmanship, testing and documentation. Use four performance levels. Write descriptors that describe observable features of the work, use parallel wording across the levels and are understandable to the apprentices. Mark any criterion that must be met in full regardless of the other scores (e.g. safety). Present the rubric as a table. Then list three weaknesses of your own draft.”

Drafting with AI · Performance levels

Descriptors should show the quality of the work

Three to five levels are typical. An even number, for example four, removes the tempting middle option. The wording should change in parallel from level to level.

“Good”, “weak” or “three mistakes” alone does not explain the quality of the work. Example: the appearance and uniformity of the weld bead on a practice piece.

Level 1

Not yet sufficient

The width and height of the bead vary strongly; undercut, porosity or lack of fusion is visible; the piece does not meet the requirements.

Level 2

Developing

The bead is mostly continuous, but its width varies noticeably; there are minor surface defects at the start and end of the weld.

Level 3

Meets the standard

The bead is continuous and even in width and height; the start and end are properly filled; no surface defects are visible.

Level 4

Exceeds the standard

As level 3, and the bead is uniform along its whole length, matching the reference sample. The learner can explain how they controlled travel speed and torch or electrode angle.

Review · Critical questions

Review the first draft critically

AI is also useful as a critical reader if we ask neutral questions rather than leading ones.

“Isn’t this a good rubric?” invites agreement. “What are the weaknesses of this rubric?” invites analysis.

Expand the review areas and use the questions to examine your own draft.

Checking the rubric for alignment with the outcomes, clarity, fairness and agreement between assessors. — Check alignment, clarity and fairness.

Review · Trying it out

Trying it out and making it transparent

Consistency

Try the rubric on sample work

Try the rubric on two or three anonymised or AI-invented sample products of different quality.

Ideally, two colleagues score them independently. Where they disagree, the descriptor needs more work.

Transparency

Discuss the criteria with learners

Learners should see and understand the criteria before they start. AI can produce “I can…” statements or a self-assessment checklist.

It is even better to discuss the criteria with learners, who can then use them for self- and peer assessment.

The teacher’s responsibility

AI drafts, the teacher decides

AI does not know your learners, your workshop, the exact exam requirements or the expectations of local employers unless you provide them. Even then, it may misinterpret them.

Its descriptors can sound convincing while being technically wrong or unrealistic.

The approved criteria are shared with learners before the task.

AI drafts a rubric; the teacher reviews, corrects and approves it, then shares it with learners. — Professional judgement and responsibility remain with the teacher.

The teacher always decides

Whether the criteria are professionally correct and suit the subject and the vocational context.
Whether the levels are realistic for the learners’ year and prior knowledge.
How points or levels are converted into a grade, and which criteria are non-negotiable.
When the rubric is revised, for example based on the experience of its first use.

The teacher’s responsibility · Using tools

Tools, data and responsibility for decisions

Rubrics in learning platforms

AI support for rubric design is increasingly built into learning platforms. In Moodle, rubrics are an advanced grading method.

In Google Classroom, Gemini can help draft a rubric from the assignment description where the school’s settings and available features allow it.

Moodle: rubrics · Google Classroom: creating rubrics

Designing criteria requires no learner data

The task description and the learning outcomes are enough to design the criteria. No personal data is needed.

Designing criteria with AI is different from evaluating learners with AI. Annex III of the EU AI Act lists AI systems intended to evaluate learning outcomes among high-risk uses; the specific classification depends on the conditions set out in the Regulation.

Regulation (EU) 2024/1689, Article 6 and Annex III, point 3

Takeaways for teachers

Five takeaways for your next assessment

Start from the learning outcomes!
Every criterion should be observable and traceable to at least one intended learning outcome.
Give context and references!
Give the AI the learners’ level, the task, learning outcomes, exam requirements and examples.
Describe quality!
Ask for descriptors with parallel wording across levels, and check their professional correctness.
Ask for critical analysis!
Use neutral questions to check alignment, clarity, fairness and weighting.
Make expectations transparent in advance!
Share the criteria with learners before the task, and keep the final responsibility for them.

Next step

Try it in your own subject!

Choose a practical task, ask for a rubric draft and try it on two sample products. The practical PDF on the final page includes exercises for individual learning and classroom work with learners.

Self-check quiz

Check your understanding!

Choose one answer for each question, then check your answers. The self-check result is displayed only on this page.

1What are evaluation criteria derived from?The score proposed by the AI.The learning outcomes.The learners’ ranking relative to one another.

2Which format assigns several performance levels to each criterion?A holistic rubric.A yes/no checklist.An analytic rubric.

3Which descriptor shows the quality of the work?The bead is continuous and even in width and height; no surface defects are visible.Good work.No more than three mistakes.

4Which question helps you review the draft critically?Surely every criterion is perfect?Which learning outcome is not assessed by the rubric?Can you confirm that no revision is needed?

5Two colleagues assess the same sample differently. What should happen next?Accept the AI’s score as final.Choose one of the two scores at random.Review and clarify the descriptors that were interpreted differently.

6What remains the teacher’s task when a rubric is drafted with AI?Professional review, approval and sharing the criteria with learners in advance.Only uploading learners’ personal data.Accepting the draft without changes.

Sources

References and further reading

Professional sources for this material

Brookhart, S. M. (2013): How to Create and Use Rubrics for Formative Assessment and Grading. ASCD.
Farkas, É. (2018): Mérés-értékelés kézikönyv. Tanulási eredmények mérése és értékelése a szakképzési mobilitási gyakorlatokban [Assessment handbook]. Tempus Public Foundation.
Council Recommendation of 22 May 2017 on the European Qualifications Framework for lifelong learning (2017/C 189/03).
Redecker, C. (2017): European Framework for the Digital Competence of Educators: DigCompEdu. Publications Office of the European Union. Area 4.1: Assessment strategies.
Regulation (EU) 2024/1689 (Artificial Intelligence Act), Annex III, point 3.
iTStudy Hungary: Innovative assessment methods [Innovatív értékelési módszerek].
Lengyel, Zs. (2026): Concrete applications for content generation. VITA training presentation.

Moodle

Rubrics as an advanced grading method.

Moodle documentation

Google Classroom

Creating rubrics with Gemini support.

Classroom Help

DigCompEdu

The European framework for educators’ digital competence.

View the framework

VITA · Credits

Credits and copyright

Author: Zsolt LengyeliTStudy Hungary Training and Research Centre
Reviewer: Mária HartyányiiTStudy Hungary Training and Research Centre
Project ID: 2025-1-LT01-KA220-VET-000362456
Learning material: M4-T2-MLC1

Attachments

Download the learning materialM4-T2-MLC1-TEXT-EN.pdf Download the practical exercisesM4-T2-MLC1-EXC-EN.pdf

This work may be reused and adapted under the terms of the Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0).

Funded by the European Union. Views and opinions expressed are those of the author(s) only and do not necessarily reflect those of the European Union or the National Agency.