Before you start

Bring a brief and someone to review it.

Generative AI has exposed a weakness in assessments that rely on a polished take-home artefact as the only evidence of learning. Teachrity fits alongside the course as a practice and assessment layer: faculty define the task and rubric, learners explain their reasoning in a voice-led session and receive immediate feedback, and faculty decide how that evidence is used. This guide shows how to do that without promising an AI-proof exam.

Your practice kit
  • A course learning outcome and the evidence it requires
  • A faculty-owned task and observable rubric
  • A plan for access, feedback, moderation and appeals
The method

8 moves to practise

  1. Start with the evidence the learning outcome requires

    Generative AI has not made every university assessment useless. It has made it harder to infer who understands a subject from a polished submission alone. Begin with the course outcome: should the learner explain a concept, apply a method, interpret evidence, make a decision or defend a recommendation? Name the evidence that would demonstrate that learning. A written report may still be right for some outcomes; an oral explanation, demonstration, portfolio or combination may be better for others. The goal is valid evidence of learning, not a task that claims to be impossible to use AI on.

  2. Separate practice from the assessment decision

    Give learners a place to rehearse the kind of reasoning the course values before a consequential assessment. Practice can use examples, hints and immediate feedback so learners can notice a gap and try again. The assessment then uses a fresh prompt, changed assumption or unfamiliar case to see whether the learner can transfer what they learned. Tell students which uses of generative AI are permitted in preparation, during practice and in the assessment. Do not imply that a practice score is automatically a final grade.

  3. Use a voice-led task when speaking reveals the intended skill

    A voice-led task can ask a learner to explain a decision, respond to a follow-up or apply an idea to a changed situation. Teachrity fits alongside the course as its practice and assessment layer: faculty create the scenario and rubric, the learner speaks through the task and receives immediate feedback, and faculty review the evidence. A typical Teachrity voice-led session is designed to take about 20 minutes. Treat that as a workable unit to pilot, not a universal optimum or a substitute for every exam. Some outcomes need longer work, written evidence or observed performance. A spoken interaction adds evidence of explanation and response; by itself it does not prove identity, prevent outside assistance or make an assessment secure.

  4. Let faculty define the rubric before the conversation

    Faculty should decide what counts as a strong answer, write criteria in observable terms and test the rubric on sample responses. Keep subject accuracy, use of evidence, reasoning and application distinct where the learning outcome requires them. Describe levels with concrete anchors rather than vague labels such as “excellent communication.” Assess delivery only when communication is itself part of the outcome. Accent, speed, confidence, eye contact or a particular speaking style should not stand in for subject knowledge. The AI can map feedback to the faculty rubric; academic staff retain authority to interpret that evidence.

  5. Make immediate feedback specific, useful and provisional

    Useful feedback points to something the learner said or did, connects it to a criterion and suggests one next move. For example: “You identified the cash-flow risk, but did not explain what evidence would change your recommendation. On the next attempt, state the assumption you would test.” Immediate feedback can make another practice attempt possible while the reasoning is still fresh. It should not be presented as infallible. Ask learners to question feedback, let faculty inspect the evidence behind it, and avoid treating automated comments as a final academic judgement.

  6. Build moderation and a route to challenge errors

    Before a pilot, have more than one qualified reviewer score a small set of responses against the draft rubric. Discuss disagreements and revise criteria that lead to inconsistent judgements. During use, review a planned sample, flagged or unclear attempts and learner appeals. State who can adjust a score, how a decision is recorded and when a human must review it. An AI system may support repeatable practice and organize evidence; it should not quietly become the sole authority over a grade, progression decision or qualification.

  7. Plan access, fairness, privacy and local policy

    Oral assessment is not automatically fairer for every learner. Consider disability accommodations, speech or hearing needs, language proficiency, connectivity, quiet space, assistive technology and equivalent ways to demonstrate the same outcome. Explain whether audio or transcripts are recorded, who can access them, how long they are retained and which service providers process them. Align the task with the institution’s academic-integrity, accessibility, privacy and assessment policies. Do not use a general-purpose “AI detector” or accent inference as proof of misconduct.

  8. Pilot for learning quality, not a marketing claim

    Start with one course, one outcome and a task faculty already understand. Review completion and access failures, rubric disagreements, the usefulness of feedback, appeals and learner experience. To examine learning, compare a baseline with a separate transfer task or other appropriate evidence; do not assume that a higher automated score proves improvement. Record what the pilot can and cannot establish, then decide with faculty whether to adapt, expand or stop. A credible assessment layer is accountable to evidence and policy, not a promise that technology has solved assessment.

See the method in context

A short oral assessment with faculty-owned criteria

Illustrative script
Brief

A learner is given a fictional working-capital case and a faculty-defined rubric. The task asks for a recommendation and the evidence behind it.

Learner

I would accept the early-payment offer because cash arrives sooner.

AI examiner

What cost or assumption would you compare with the benefit of receiving cash earlier?

Learner

I would compare the discount with the cost of borrowing and check whether the customer reliably pays on the agreed date.

Feedback

You named two relevant factors. In practice, strengthen the recommendation by showing how each factor changes the decision. Faculty review determines how this evidence is used.

Example rubric: observable evidence, not a personality score

CriterionDeveloping evidenceClear evidence
Subject understandingNames a concept but applies it inaccurately or without context.Applies the relevant concept accurately to the facts in the task.
ReasoningStates a conclusion with little explanation of how it follows.Connects the evidence, assumptions and conclusion; responds to a relevant follow-up.
Use of evidenceMakes a claim without identifying what supports it.Points to relevant case facts and distinguishes known information from an assumption.
TransferRepeats a prepared answer when a condition changes.Explains how a changed assumption would affect the recommendation.
Communication, if assessedThe intended audience cannot identify the recommendation or next step.Presents the reasoning in a form the intended audience can follow; accent is not a proxy for competence.
Put the guide to work

Try it. Review it. Change one thing.

Your practice task

Choose one assessment that relies heavily on an unsupervised artifact. Redesign it as a sequence of preparation, spoken explanation, evidence-based follow-up and a fresh transfer task.

What to look for

Check alignment, accessibility, score consistency, learner understanding of AI rules and human ownership of consequential decisions.

Before the next attemptJust for you: nothing here is saved.

Common questions

Has generative AI made higher-education assessment completely broken?

No single description fits every course. Generative AI has exposed limits in relying on unsupervised written artefacts alone. Institutions can respond by aligning assessment with learning outcomes and gathering appropriate evidence through a mix of written, oral, practical and process-based work.

Does a 20-minute voice assessment make an exam AI-proof?

No. A short voice-led assessment can gather evidence of explanation and response, but it does not by itself verify identity or prevent outside assistance. The institution must define the task conditions, security level and role of the result.

Can the AI assign the final grade?

Teachrity can provide immediate rubric-linked feedback for practice. Any use of AI-generated assessment evidence should follow the institution’s policy; a qualified human should remain accountable for consequential academic decisions and have a way to review or correct errors.

Who should write the rubric?

The faculty or subject owner should define criteria from the learning outcome, test them on sample responses and revise unclear descriptors. A platform applies the configured criteria; it should not invent the academic standard.

Are Cornell and NYU examples of this broader shift?

Their public resources describe work on oral assessment and AI-supported oral exams. Those examples are not evidence that either institution uses or endorses Teachrity. See the university sources below for their own descriptions.

How should a university start?

Choose one course outcome, one task and one faculty owner. Pilot with a small group, check access and reviewer agreement, and decide what evidence would justify continuing before expanding.

What does it mean to use Teachrity as an assessment and practice layer?

It means adding repeated, rubric-guided practice and a voice-led evidence point to a course the university already owns. Faculty set the outcome, task and criteria; Teachrity provides the conversation and immediate feedback; faculty and the institution decide how results are moderated and used.

Evidence and context

Research and further reading

Use these primary resources to explore the method and its context. References describe their publishers’ work; they do not imply endorsement or certification of Teachrity.

This guide supports assessment design discussion; it is not legal, accreditation or academic-integrity advice. Institutions remain responsible for their policies, accommodations, data governance and academic decisions.

Choose your next step

Keep the next task focused.

See an oral-assessment use caseAI viva voce and oral assessmentContinue