Can ChatGPT grade essays? Ask who signs the grade

Can ChatGPT grade essays? Ask who signs the grade. Defending a grade takes your rubric applied by criterion, a named marker's sign-off and a record of both.

The short answer to “can ChatGPT grade essays” is that the question stops too early: whatever a chat window returns, an institution has to defend everything around the mark — whose rubric was applied, who signed the grade, what was overridden and why. An appeal needs a decision record, not a conversation. NUADU's AI is built around that record: it applies your rubric, not a generic model's opinion, and a named educator decides before any output reaches a learner. Here is the argument, set out for the assessment committee that asks.

Can ChatGPT grade essays with a rubric?

Whatever tool drafts the mark, ask whether it is tied to a criterion of record, a named marker and a recorded change. On NUADU the rubric comes first and travels. The same rubric is used in authoring, marking and feedback.

With AI assistance on, essays and speaking receive draft scores and rubric-anchored explanations, and the marker reviews, overrides, signs off. That is what AI-assisted marking with the marker's sign-off means in practice: the AI applies your rubric and surfaces evidence, and the decision stays with a named person.

The difference shows the first time someone questions one essay. A rubric-anchored draft can be read criterion by criterion, overridden with a recorded reason and explained to the candidate. A mark can be audited only when there is a criterion of record to hold it to and a named decision to review. The AI capabilities page puts it bluntly — aligned with your rubric, not a generic LLM dressed up.

A grade you can't explain is a grade you can't defend.

Is ChatGPT marking accurate enough for a result that counts?

Accuracy is the wrong place to stop, for any model. NUADU's AI Transparency Notice says of its own models that they can mis-score, miss errors or give irrelevant feedback, and that no output is definitive without human review. For a result that counts, the question is who reviews each output before it stands.

It is as plain about the other limits: models may treat some dialects, non-native speakers or writing styles less fairly, which NUADU monitors, adjusting prompts and rubrics and sharing monitoring results on request; and English inputs are the most reliable.

None of that argues against AI assistance in marking. It argues for a named marker between the model and the grade, and for measuring agreement. On NUADU, agreement rates between AI and markers, between markers and across cohorts are visible to QA, so drift can be spotted and addressed.

Who signs the grade?

A named educator, and the roles around that signature stay separate. An assessment office keeps the person who drafts apart from the person who decides, and both apart from the appeal. Whatever tool drafts the mark, ask whether those roles stay separate.

  • The drafter: with AI assistance on, a module drafts scores and rubric-anchored notes; it drafts, it does not decide.
  • The marker: reviews, overrides any score where needed and signs off; AI never makes the final grade decision.
  • The second rater: AI can act as a documented second rater inside your existing second-marking policy; the marker is still the marker.
  • The feedback author: feedback is assembled against the rubric for the marker to edit before release, and the learner sees the marker's name.
  • The contest: the candidate can contest any released grade through your appeals process, and the audit trail supports both the override and the contest.

The last two roles are where rubric-based feedback earns its place. A learner can act on feedback tied to a criterion, and contest it, because it carries a name. The rubric marking workflow shows where the override and the sign-off sit.

What will an appeal ask for?

The record of who reviewed the draft, what they changed and why. On NUADU every AI output, override and approver is captured in an end-to-end audit trail, so a QA office, a DPO, an appeals panel or an external auditor can trace a decision to a named human. An appeal needs to see who decided, not only what was typed.

The law points the same way. Under Article 22 of the GDPR, a person has the right not to be subject to a decision based solely on automated processing which produces legal effects concerning them or similarly significantly affects them. That is why the record matters as much as the mark: on NUADU the marker reviews every result, may override it and signs it, and the platform records who decided and when.

The stakes are classified, too. The EU AI Act names AI systems intended to evaluate learning outcomes as high-risk in Annex III, point 3(b), and NUADU's notice places its own assessment capabilities in that category when institutions use them. Whatever the quality of a draft, the record of who reviewed it is what an appeal, an audit or an accreditation panel will ask for.

Can teachers use ChatGPT to grade?

That is the institution's policy to set, and the test is the same for any tool: can the grade be explained criterion by criterion, with a named marker's signature and a record an appeal can read? None of this says a general assistant has no place in a university. It says that assessment is where a decision about a named learner has to be explained later, and the tool has to be built for that moment.

NUADU also has a “ChatGPT vs a dedicated assessment platform” comparison, on request, alongside the documentation pack. For any vendor, the check is whether each AI module fits one sentence that names its task, the person who decides and its limit. NUADU's approach to AI in assessment is written that way, and a named educator at every decision point is what that sentence looks like when an institution audits it.

→ Bring a sample of marked work and your QA officer (30 min)