Rubric criteria for consistent marking

Use this workflow to turn one assignment brief and its learning objectives into a usable marking rubric. It is for teachers who need clear criteria, and for subject leads or assessment coordinators who need staff to apply those criteria in the same way.

The result is a draft rubric, not an approved assessment policy. A teacher or subject lead must check that it matches the course specification, department language and local grading rules before it reaches students.

1. Gather the source material

Put the following in one working document before you ask the model to draft anything:

• The task brief, including the required format, word count or duration where relevant, audience and submission requirements.
• The learning objectives being assessed. Copy their exact wording.
• The mark scheme, level descriptors or curriculum statements already required by your school, exam board or department.
• One or two anonymised examples of previously marked work, if you have permission to use them and they represent the expected standard.
• The proposed mark total or level scale, for example four performance levels or marks out of 20.
• Any non-negotiable rules, such as whether spelling is assessed separately, whether sources must be cited, or whether process work counts.

Remove student names, identifiers, parent correspondence and personal circumstances. You do not need them to create the rubric.

Choose one assessment purpose. A rubric for formative feedback needs room for next steps. A rubric for a final graded task needs a stable mark allocation. Trying to serve both purposes usually creates vague criteria and overlong descriptors.

2. Decide the rubric shape before drafting

Set the structure yourself. Do not ask the model to choose it without constraints.

For most classroom tasks, use three to five criteria and three to five performance levels. Each criterion should test a distinct part of the learning. For an extended history response, for example, you might assess use of evidence, analysis, organisation and subject vocabulary. Do not create separate criteria for points that are already part of another criterion.

Use the table below to select the layout.

If the task needs	Use	Set up
A final numerical score	Analytic rubric	Separate criteria, each with marks that add to the total
A consistent attainment judgement	Level rubric	Shared level labels and descriptors for each criterion
Fast whole-task feedback	Holistic rubric	One descriptor per overall level, with fewer diagnostic details
Feedback before redrafting	Analytic rubric plus next steps	Criteria with a short improvement prompt below the table

For consistent marking, an analytic rubric is usually the safest starting point. It makes the reasons for a mark visible and helps teachers identify where they disagree.

3. Give the model a bounded drafting request

Paste the task brief and objectives, then state the rubric structure. Ask for a table, rather than a narrative explanation. State that descriptors must be observable in the submitted work.

Use this prompt structure:

Draft an analytic assessment rubric for the task below.

Task brief:
[PASTE TASK BRIEF]

Learning objectives:
[PASTE OBJECTIVES]

Required structure:
- Four criteria only.
- Four performance levels: Secure, Developing, Emerging, Not yet evident.
- Marks: 4, 3, 2 and 1 for each criterion.
- Total: 16 marks.
- Use observable evidence from the student response.
- Keep each cell to one or two sentences.
- Make each higher level clearly better than the level below it.
- Do not assess effort, behaviour, presentation or prior attainment unless named in the learning objectives.
- Return a Markdown table, followed by a short list of marking notes.

Department requirements:
[PASTE REQUIRED TERMS, EXCLUSIONS AND COURSE RULES]

If you are unsure which features or settings are available in the model you are using, check the xAI documentation overview. They can vary by version and account setup.

Ask for a second output after the first draft: a mapping table showing which learning objective each criterion assesses. This exposes criteria that sound sensible but do not actually assess the intended learning.

4. Test each criterion against the task

Read the draft one row at a time. For every criterion, ask three questions:

1. Can you point to evidence of this in the student’s actual submission?
2. Does this criterion assess one main thing, rather than several unrelated things?
3. Does it map to at least one stated learning objective?

Rewrite broad labels such as quality, understanding or effective communication. They are headings, not criteria. Replace them with observable actions or features. For example, change shows good analysis to explains how the selected evidence supports the argument and considers an alternative interpretation.

Then check the performance descriptors vertically. The difference between adjacent levels must be clear. A student should not receive a higher level merely because a teacher prefers the phrasing.

5. Check marks, levels and gaps

Add the marks across the rubric yourself. Confirm that the highest possible total is the intended total and that all criteria carry the weight you mean them to carry. If argument matters twice as much as technical accuracy, show that through marks or weighting, not through a longer paragraph.

Look for these common faults:

• A criterion rewards the same evidence as another criterion. Merge them or make the distinction explicit.
• The highest descriptor introduces a requirement that does not appear at lower levels. Add a progression or remove the extra requirement.
• The lowest descriptor describes a student rather than the work. Replace does not understand with gives an unsupported claim.
• A descriptor includes a vague quantity, such as uses several examples. State a range only if your department can defend it as appropriate for the task.
• A criterion tests a requirement missing from the task brief. Either amend the brief before students see it or remove the criterion.

6. Run a short moderation check

Use two or three anonymised samples that sit at different standards. Mark each sample with the new rubric, preferably with another teacher if one is available. Record the selected level for each criterion and one sentence of evidence from the work.

Where markers differ, do not immediately average the marks. Identify the exact cell that led to different readings. Then revise that descriptor. Common fixes include naming the evidence required, separating two skills that were combined, or clarifying whether accuracy, explanation or selection is being judged.

Keep a brief change log with the task name, rubric version date used locally, changes made and reason. This lets the department explain why a rubric changed between cohorts without rewriting its history.

7. Prepare the handover pack

Give teachers one folder or document containing:

• The final student-facing task brief.
• The final rubric in an accessible table format.
• A teacher-only version with marking notes and acceptable evidence examples.
• The objective-to-criterion mapping.
• The moderation notes and any agreed interpretations.

Tell staff which document is authoritative. Do not circulate several near-identical rubric drafts. That is how inconsistent marking starts.

When the draft does not work

If the rubric is too generic, provide one anonymised sample and ask the model to identify evidence that would distinguish each level. If it is too detailed, reduce the number of criteria before shortening descriptors. If it does not match the course requirements, stop editing the wording and return to the learning objectives and required assessment language.

If teachers still mark the same sample differently after revision, treat that as a moderation issue, not a prompt-writing problem. Bring the marked samples to the subject lead or assessment coordinator, agree the interpretation, update the marking notes and rerun the check.