AI tutor for higher education: opt-in, bounded and switchable off

An AI tutor for higher education should be opt-in, limited to approved content and the learner's own work, switchable off, and never a student profile.

An AI tutor for higher education is worth switching on only if the learner controls it, it draws on approved content, and it never turns into a profile of the student. On NUADU the closest module is an optional study companion: started by the learner, bounded to their recent work and approved content, open about the data it uses, and switchable off, with deletion on request. Beside it sit a practice route drafted after a result and pronunciation feedback for practice. Here are the boundaries, who holds each switch, and where the teacher decides.

What should an AI tutor for higher education be allowed to see?

The learner's own recent work and content the programme has approved, and nothing else. An AI tutor is software that helps a learner practise between taught sessions, and the more it sees, the closer it comes to a profile. NUADU draws the line at the source: the study companion is bounded to the learner's recent work and approved content, and it never draws on sources outside that set.

The same boundary runs through the modules around it. The practice route after a result is drafted from approved content, for the rubric criteria where the learner scored lowest. Resource search looks only across approved learning content and finds the segment of a reading, video or document that addresses a concept; the educator decides what to assign. A learner is not handed material the programme has not reviewed.

Optional, learner-controlled study companion — never pervasive profiling.

Who switches an AI tutor on, and off?

Two parties, in order: the institution, then the learner. The institution decides whether the module runs at all, because each AI module can be enabled per assessment, per programme or per institution, or left off entirely. That first switch is where responsible AI adoption in higher education begins. The learner holds the second: the companion starts only when the learner opts in, and the learner controls when it is on.

Control means more than an on button. The learner sees the data the companion uses, can switch it off and can request deletion. The practice route is proposed, not imposed: the learner chooses whether to follow it, and no step is mandatory. Both modules are opt-in, so whether either one runs in a deployment is the institution's choice before it is the learner's.

Does an AI tutor have to profile students?

No. Profiling, as the GDPR defines it in Article 4(4), is automated processing of personal data used to evaluate personal aspects of a person, in particular to analyse or predict aspects such as their interests or behaviour. A tutor does not need a profile of the student to be useful. On NUADU the study companion does not profile the learner, the practice route is not a profile and is never presented as labels, and resource search never profiles the learner.

What the modules offer instead is narrow on purpose: the practice route covers the criteria where a score was lowest, from approved content, after a result is released. Analytics follow the same rule at programme level — individual signals go to a tutor, never onto a label — and how to identify students who need support without labelling them gets the same answer on the institution's side.

Where does the teacher come in?

At every module, by role. For pronunciation feedback, teachers decide whether to act on it. For resource search, the educator decides what to assign. Long-form review sends structural signals from the draft stage to the supervising educator, and it is designed for formative feedback, not summative grading. The practice route follows a result a marker has signed.

That is the difference between a bounded companion and a tutoring system that quietly builds a model of each student and reports back. NUADU's modules do less, by design. They propose a practice route or find a passage, and a named person decides what happens next.

For a language centre, the natural home is practice between taught sessions. A learner receives a result and may opt into a practice route for the criteria where the score was lowest. Pronunciation feedback indicates which sounds drift from a target and offers practice prompts, in formative practice rather than high-stakes scoring, and the live demo of pronunciation feedback shows it on its own. When speaking counts towards a result, how speaking skills are assessed is a marker's job, at audio level.

In higher education the companion appears in the learner line of the assessment platform for higher education, beside rubric-anchored feedback, the marker's name on the result and a documented appeals process. The AI in assessment page lists each of these modules with its human decision and its hard limit.

Frequently asked questions

What are AI tutors?

Software that helps a learner practise between taught sessions: explaining, suggesting practice, giving feedback on attempts. In an institution the useful version is bounded. It works from approved content and the learner's own work, and a teacher decides what to act on.

Are AI tutors good?

Good for what is the institution's question, and no vendor should answer it with a promise. What can be stated is the test a tutor should pass before it is switched on: bounded to approved content, controlled by the learner, open about the data it uses, never a label, with a teacher deciding what happens next.

Are AI language tutors good?

For pronunciation practice, a narrow tool is easier to trust than a broad one. NUADU's pronunciation feedback is phoneme-level and comes with practice prompts; it is used in formative practice, not high-stakes scoring, and teachers decide whether to act on it.

How do you protect student privacy when using AI?

Limit what the tool sees, let the learner switch it off, and keep the documentation. On NUADU the companion is bounded to the learner's recent work and approved content, and the learner can switch it off and request deletion. The documentation pack, including a DPIA template, is available on request, and EU data residency is the default, with other regions on request.

→ Talk to us about your assessment workflow (30 min)