Should You Tell Students an AI Is Running Their Course Evaluation? The Disclosure Question

Bottom line up front: When an AI conducts or analyses a course evaluation, institutions face a choice they rarely articulate: tell students plainly that they are interacting with an AI, or let them assume otherwise. From 2 August 2026, the EU AI Act removes the discretion for the interaction itself — Article 50 requires that people be informed when they are dealing with an AI system, unless it is obvious. But disclosure is more than a compliance box. Handled well, telling students "an AI is conducting this interview, here is why and what happens to your words" is a trust-building act that improves the data; handled badly or omitted, it becomes the thing that quietly poisons your evaluation programme. This is a governance question every quality-assurance office adopting conversational evaluation needs an answer to now.

What Article 50 actually requires

The EU AI Act entered into force in 2024 and phases in over several years. Article 50 — the transparency obligations — starts applying on 2 August 2026. Crucially, these obligations are not limited to "high-risk" AI systems; they attach to specific situations regardless of risk classification. One of those situations is directly relevant here: providers and deployers of AI systems intended to interact with people (chatbots, virtual assistants, conversational agents) must ensure those people are informed that they are interacting with an AI, unless that fact is already obvious to a reasonably observant person. The European Commission's guidance indicates the notice must come at the latest at the first interaction — for a conversational system, at or before the start of the conversation, and it can be satisfied by a clear written statement up front.

A conversational AI interviewer asking a student about their course is squarely within scope. So the baseline is settled: from August 2026, a student talking to an AI moderator in the EU must be told. The interesting questions are everything around that baseline — and they are where institutions either build trust or squander it. (Whether the analytic use of feedback for personnel decisions tips a system into the AI Act's high-risk regime is a separate, important question we examine in our piece on the AI Act and high-risk student feedback.)

Disclosure is not one decision — it is four

"Tell the student it is an AI" is the floor. A credible disclosure posture answers four distinct questions, only the first of which Article 50 strictly mandates:

1. That it is an AI. Required. State it plainly at the start, in plain language, not buried in a privacy policy.
2. Why an AI is being used. Not strictly required, but decisive for trust. Students reasonably wonder whether an AI is there to surveil, to cut costs, or to genuinely listen. Saying why — consistent, unbiased probing; the ability to follow up on every answer; no tired human moderator on response 400 — reframes the AI as a feature, not a threat.
3. What happens to their words. This is where transparency obligations meet the GDPR/AVG. Students need to know who can see their responses, whether and how anonymity or confidentiality is protected, how long data is retained, and whether outputs feed consequential decisions — the substance of anonymity versus confidentiality under GDPR in course evaluation.
4. What the AI does and does not do. Honest scoping — it summarises themes, it does not assign your grade; a human reviews decisions — heads off the automation fears that erode candour.

Treat only the first as the requirement and you will be compliant and still distrusted. Treat all four as the standard and disclosure becomes the foundation of a programme students actually engage with.

"But won't telling students it's an AI make them clam up?"

This is the strongest objection, and it has to be taken seriously rather than waved away. The worry: if you announce "you are talking to a machine," students will disengage, give shorter answers, or refuse to be candid — and you will have complied your way into worse data.

The evidence is more reassuring than the fear, though it is not unanimous. There is a substantial research literature showing that people often disclose more sensitive information to a computer or AI interviewer than to a human, because the perceived judgement and social desirability pressure drop — the absence of a human audience can make people more honest, not less. That is the optimistic case, and it is well grounded. But honesty also points the other way: some students distrust AI, some worry about being recorded or de-anonymised, and a poorly explained AI can absolutely depress participation. The two facts are reconcilable: candour depends less on whether there is a human and more on whether students believe the channel is safe and worth their time — which is precisely what good disclosure establishes. The question of whether students will be honest with an AI interviewer turns on trust, and concealment is the worst possible trust strategy: the downside of a student discovering an undisclosed AI — feeling deceived by their own institution — is far more corrosive than the downside of telling them up front.

A second objection deserves a direct answer: isn't mandatory disclosure just regulatory boilerplate that changes nothing? No. The AI Act sets a floor for the interaction, but it does not dictate the quality of disclosure, and it does not by itself address the GDPR questions students care about most. An institution that meets the letter of Article 50 with a grudging one-liner is leaving the trust dividend — and the data-quality dividend — on the table.

The transparency-as-advantage reframing

The instinct to minimise disclosure treats it as a cost. For evaluation specifically, it is closer to an asset. Course evaluation lives or dies on candour and participation; both are functions of trust; and trust is built by telling people the truth about how their feedback is collected and used. An institution that is visibly transparent about its AI — what it is, why it is there, what it does with your words, what it will never do — is making a credibility claim that legacy survey tools never had to make and rarely earned. Disclosure done well is not the tax you pay to use conversational AI; it is part of why the AI works.

How Koji approaches it

Koji for Education is built on the premise that transparency is a feature of the instrument, not a disclaimer bolted onto it. Students are told clearly, at the start of a conversational interview, that an AI moderator is conducting the session — satisfying the Article 50 baseline by design rather than as an afterthought. The why is honest and student-facing: a standardized, bias-aware AI moderator probes every answer with the same consistency, removing the human-moderator inconsistency and fatigue that distort traditional interview-based evaluation, and it can follow up on response number 400 as attentively as response number one. On the data side, the pipeline is GDPR/AVG-compliant and built for European norms, with clear handling of anonymity, confidentiality and retention, so the "what happens to your words" question has a real answer. The AI is honestly scoped — it conducts interviews and produces thematic analysis and quality scoring for human review; it is not a black box handing down verdicts, which is the design posture that keeps automation bias out of committee decisions. The result is the future of student feedback as conversational, transparent interviews rather than anonymous-but-opaque survey extraction. The same engine and the same transparency discipline run on the main Koji platform for customer and user research, where being upfront about the AI is equally central to participation.

From August 2026, telling students an AI is running their evaluation stops being a choice about whether. It becomes a choice about how well — and the institutions that treat disclosure as a trust-building feature, not a compliance chore, will be the ones whose students keep talking. Koji is built for that posture from the first message.

Transparent, AI-native course evaluation built for the EU. Explore Koji for Education.