Fewer Answers, Better Questions: Generative AI and the Return of the Viva

Fewer Answers, Better Questions: Generative Artificial Intelligence, the Expert Questioner and the Case for the Viva Voce in Secondary Education

Simon Uttley

.

Cite as: Uttley, S. R. (2026). Fewer answers, better questions: Generative artificial intelligence, the expert questioner and the case for the viva voce in secondary education. Koinonia Educational Ltd.

Abstract

Generative artificial intelligence has removed an assumption on which much school assessment rested: that a piece of submitted written work is reliable evidence of the mind of the pupil who submitted it. Detection tools have proved unreliable, and regulators in England have responded by restricting written coursework and tightening authentication. This paper argues that the more significant consequences lie elsewhere. First, because fluent answers are now abundant and cheap, the teacher’s expertise is relocated from the supply of answers to the design and conduct of questions; the teacher’s distinctive contribution becomes the question that shows whether understanding is owned or borrowed. Second, the viva voce, long confined in England to the doctorate and a small number of professional and language examinations, offers a structurally robust form of evidence that is likely to become far more common in secondary schools. Drawing on research on classroom questioning, wait time and dialogic teaching, on the higher education literature on oral and interactive assessment, on recent field evidence about generative AI and learning, on international precedent from France and Italy, and on the author’s earlier work on silence and counterpoint in coaching, the paper proposes a typology of questions for AI-rich classrooms and a set of design principles for a secondary viva. It engages directly with the strongest objection, that oral assessment rewards cultural capital rather than understanding, and sets out the conditions under which it need not. It closes with a reading of the argument in the light of Catholic educational teaching, in which education is understood as the patient uncovering of truth through encounter.

Keywords: generative artificial intelligence; assessment validity; questioning; wait time; viva voce; oral assessment; oracy; secondary education; Catholic education

1. Introduction

For most of the modern history of schooling in England, the written assignment has carried a large share of the weight of assessment. A pupil goes away, thinks, drafts and returns with a text, and the teacher reads that text as evidence of what the pupil knows and can do. The arrangement depended on an inference so familiar that it was rarely stated: the person who handed in the work produced it, and the work therefore reveals something about that person’s understanding. Generative artificial intelligence (AI) has broken that inference. A capable language model can now produce, in seconds, prose that is fluent, organised and frequently correct, and no reliable means exists of telling from the text alone whether a pupil or a machine wrote it.

The policy response in England has so far been largely defensive. The qualifications regulator has acknowledged that non-exam assessment is under particular pressure (Ofqual, 2024), the Curriculum and Assessment Review advised against any expansion of written coursework because of the risks generative AI poses (Department for Education [DfE], 2025), and in March 2026 the Chief Regulator asked exam boards to strengthen deterrence, detection and prevention of AI misuse, including a more rigorous approach to the authentication of candidates’ work (Ofqual, 2026). These responses are understandable, and some of them are necessary. They are also incomplete. They treat the problem chiefly as one of security rather than of evidence, and they say little about what should happen in classrooms between formal assessments.

This paper makes two linked claims. The first is that generative AI will require teachers to become far more expert questioners than most initial teacher education and professional development currently prepares them to be. When answers are abundant and cheap, the scarce educational good is the question that reveals whether a pupil understands an answer, can defend it, can see where it fails and could reconstruct it. The second is that the viva voce, the examination conducted “by the living voice”, which in England has largely been reserved for the doctorate, is likely to become a normal part of secondary assessment. The two claims describe one shift. As the answer loses its value as evidence, the question becomes the unit of educational value, and the most trustworthy evidence of learning becomes a conversation in which a pupil responds to questions they could not have rehearsed with a machine.

The author writes as headteacher of an urban comprehensive secondary school in England and as a researcher. The school’s published character framework pairs the traits of being eloquent and truthful, and defines eloquence broadly, across speech, writing, music, drama, art and sport rather than verbal fluency alone (Uttley, 2026b). The conviction behind that pairing, that fluency must answer to truth, runs through the argument that follows. The author’s earlier work on silence and on counterpoint in coaching conversations (Uttley, 2026a, 2026c) supplies part of the conceptual apparatus for the account of questioning developed in Section 5, and his work on cultural capital (Uttley, 2025a) frames the discussion of equity in Section 8.

The paper proceeds as follows. Section 2 sets out the approach. Section 3 examines the collapse of written work as a proxy for understanding and the policy response in England. Section 4 argues that AI relocates the teacher’s expertise from answer to question. Section 5 offers an account of expert questioning, including a typology of questions suited to AI-rich classrooms. Section 6 reviews the lineage of and evidence for the viva voce, including precedent in secondary systems abroad. Section 7 proposes design principles for a secondary viva. Section 8 addresses the objection from equity. Section 9 offers a reading of the argument in the light of educational teaching in the Catholic tradition. Section 10 states the paper’s limitations and an agenda for research, and Section 11 concludes.

2. Approach

The method is conceptual and documentary rather than empirical. The paper synthesises published research on classroom questioning and dialogue, on oral assessment in higher and professional education, and on the effects of generative AI on learning, and reads it alongside current regulatory and policy texts from England and comparable systems. It does not claim to be a systematic review. Sources were selected for their relevance to the argument and for the strength of their evidence, and every reference has been checked against a publisher, repository or official record.

Two limits should be stated at the outset. First, much of the strongest evidence about oral assessment comes from universities and from professional training, particularly medicine. Its transfer to adolescents in secondary schools has to be argued rather than assumed, and Section 10 identifies where school-based evidence is most needed. Second, the author writes from inside the system he describes. That position gives access to the practical realities of running a school in which these questions are live; it also carries the risk of seeing what one hopes to see. The paper tries to meet that risk by giving the objections to its argument their full weight.

3. The collapse of the written proxy

3.1 Assessment as inference

Every assessment is an argument from evidence to a claim about a learner. A mark on a history essay asserts that the pupil can construct an argument from sources; a grade in science asserts command of certain concepts and procedures. The quality of that argument is what assessment specialists call validity. Dawson et al. (2024) argue that, in the debate about AI, validity is a more useful concept than cheating: what matters is whether an institution can be confident that its students are capable of what it says they are. Cheating matters because it damages that confidence, but so do other things, and the authors observe that measures introduced to prevent cheating can themselves weaken validity, for instance by making assessment less inclusive. That observation will matter in Section 8.

Seen in these terms, generative AI does not introduce a new problem so much as remove the conditions under which an old solution worked. Written coursework was never a perfect window onto a pupil’s mind; parental help, tutoring and copying have always clouded it. The difference now is one of scale and quality. Assistance that was once uneven, detectable and limited is now universal, fluent and free.

3.2 The failure of detection

The first institutional instinct was to detect. The evidence does not support it. Weber-Wulff et al. (2023) tested fourteen detection tools, including two commercial systems widely used in universities, and concluded that none was accurate or reliable enough for the purpose. The tools tended to classify machine-generated text as human-written, and their performance worsened further when the text had been edited, paraphrased or machine-translated. A school that relies on such tools risks two failures at once: missing much of the misuse that occurs, and accusing pupils who have done nothing wrong.

3.3 The policy response in England

Ofqual’s (2024) statement of its approach to AI noted that non-exam assessment, particularly where students produce work outside examination conditions, was potentially under more pressure, while reporting that the number of malpractice cases identified by awarding organisations had so far been modest. The Curriculum and Assessment Review went further. Its final report concluded that externally set examinations should remain the principal form of assessment, that non-exam assessment should be used only where it is the only valid way to assess essential elements of a subject, and that there should be no expansion of written coursework given the risks posed by generative AI. In the same report it recommended reducing overall GCSE examination time by at least ten per cent (DfE, 2025).

In March 2026 the Chief Regulator wrote to exam board chief executives about two threats to the integrity of qualifications. The first was the rising number of candidates taking mobile phones and other devices into examination halls, which accounted for almost half of the roughly five thousand malpractice cases reported to Ofqual each year. The second was the misuse of AI in non-exam assessment, particularly coursework requiring extended writing. Although detected cases remained relatively low, the letter recorded significant concern among teachers and leaders about the real extent of misuse, and asked boards to take immediate steps, including a more rigorous approach to candidate and teacher authentication (Ofqual, 2026).

The combined effect of these positions is a narrowing. There is to be less written coursework, less examination time, and stronger authentication of whatever coursework remains. That narrowing leaves an evidential gap. Some of the capabilities that matter most in a subject, such as sustained reasoning, the defence of a position under challenge and the integration of knowledge across a course, are poorly captured by a shortened timed paper. If examinations are to shrink and coursework is not to grow, something must carry evidence of those capabilities. This paper argues that the spoken examination is a strong candidate.

3.4 Discursive and structural responses

Corbin et al. (2025) offer a distinction that clarifies the choice. A discursive change alters what students are told about a task: an instruction that AI may not be used, for example, or a requirement to declare how it was used. A structural change alters how the task must actually be completed. Discursive changes depend on compliance that cannot be verified; structural changes build the desired behaviour into the design of the assessment. Frameworks such as the AI Assessment Scale (Perkins et al., 2024) give schools a useful common language for specifying the level of AI use permitted in a task, but any such framework is only as strong as the means of confirming what happened. The guidance prepared for Australia’s higher education regulator makes a related point, arguing that trustworthy judgements about learning in a time of AI require multiple, inclusive and contextualised approaches rather than reliance on any single form of evidence (Lodge et al., 2023).

A conversation about a piece of work is a structural change. It does not ask the pupil to promise anything. It asks the pupil to explain, defend and extend what they have submitted, in real time, to someone who knows the subject. That is a different kind of evidence from a script, and it is far harder to counterfeit.

4. From answer to question

4.1 What generative AI does to learning

The most instructive evidence to date comes from a field experiment involving nearly a thousand secondary mathematics students in Turkey (Bastani et al., 2025). Students were given access either to a tool resembling a standard chatbot interface or to a version whose prompts were designed to safeguard learning by offering hints rather than solutions. The unrestricted tool improved performance during practice, but when access was removed, its users performed worse than peers who had never had it. The authors found that students had used the unrestricted tool as a crutch, and that the safeguards in the tutoring version largely removed the harm.

A large survey and interview study points in the same direction. Gerlich (2025) found a significant negative association between frequent use of AI tools and critical thinking among 666 participants, mediated by what psychologists call cognitive offloading, the delegation of mental work to an external aid. Younger participants showed greater dependence on the tools and lower critical thinking scores. The design is correlational and cannot establish that AI use causes weaker thinking, but the pattern is consistent with the experimental evidence.

The conclusion to draw is not that AI should be kept out of classrooms. It is that design matters, and that what protected learning in the Turkish experiment was withholding: the tool that helped was the one that declined to give the answer and asked the student to do the next step. That is what good teachers have always done. The difference is that the temptation to bypass the pupil’s thinking, which was once confined to moments when a teacher or parent supplied the answer too quickly, is now available to every pupil at every moment.

4.2 Productive withholding

The author’s earlier work on silence in coaching offers a way of naming this. Drawing on the use of shadow in painting and of rests in music, Uttley (2026c) argues that absence, whether of speech, light or sound, is not mere negation but a structural condition of meaning, and that the practitioner’s power is often exercised through restraint rather than addition. The coach who fills every moment with observations and suggestions denies the person being coached the space in which to do their own work. The paper calls this the paradox of productive withholding.

The classroom analogue is direct. A teacher who answers too quickly, like an unguarded chatbot, does the pupil’s thinking for them. In a classroom where every pupil can summon a fluent answer from a phone, the teacher’s distinctive task is to create, protect and use the space in which the pupil has no option but to think. The question, and the silence that follows it, become the teacher’s principal instruments.

4.3 Pre-cast conclusions

Catholic educational teaching anticipated the shape of this problem. The Catholic School (Sacred Congregation for Catholic Education [SCCE], 1977) holds that a school must stimulate pupils to exercise their intelligence through the dynamics of understanding, and that any school which offers merely pre-cast conclusions hinders the personal development of its pupils (para. 27). It also asks schools to form pupils’ critical faculties systematically, so that they can choose freely and conscientiously in the face of what is offered by the organs of social communication, which the document describes as a kind of parallel school (para. 48).

A generative language model is the most capable parallel school yet devised. It supplies pre-cast conclusions on demand, in fluent prose, tailored to the question asked. Read in 2026, the document’s demand that pupils learn to subject what they are offered to critical and personal analysis reads less like a period piece than a specification for the AI-era classroom.

5. The anatomy of expert questioning

5.1 What the evidence already shows

Much of what expert questioning involves is already known. Rowe’s (1986) studies of classroom talk found that teachers typically waited around a second after asking a question before speaking again, and around a second after a pupil’s answer before responding. When those pauses were extended to three seconds or more, pupils’ answers became longer and more often reasoned, and a wider range of pupils took part. Black et al. (2004), reporting on teachers who had worked to develop formative assessment, described how planning better questions and lengthening wait time shifted classroom exchanges from the search for a correct answer towards the exploration of pupils’ thinking. Rosenshine’s (2012) synthesis of research on effective teachers found that they asked a large number of questions and checked the responses of all pupils rather than a few volunteers.

The largest recent study of whole-class dialogue in England examined seventy-two Year 6 classrooms and found that the elaboration and querying of pupils’ contributions, together with broad participation, were associated with better pupil outcomes (Howe et al., 2019). Alexander (2020) has argued over two decades that classroom talk should be cumulative, with each question building on previous answers, and his approach to dialogic teaching was evaluated in a trial involving some five thousand pupils.

These findings predate generative AI. Their importance has grown because the conditions they describe, a pupil thinking aloud, in the moment, without a script, are exactly the conditions a machine cannot supply on the pupil’s behalf.

5.2 Questioning as counterpoint

The author’s account of counterpoint as a framework for coaching (Uttley, 2026a) offers a richer description of what the expert questioner does. In musical counterpoint, two or more independent lines are combined so that each keeps its own integrity while standing in a governed relationship with the others. Applied to coaching, the coach neither echoes the client’s account back, which would be mere unison, nor imposes a separate agenda, but introduces a related line of inquiry that both supports and complicates the client’s theme. The skill lies in calibrating the degree of dissonance: too little and the second voice merely repeats the first; too much and the conversation dissolves. The coach prepares a challenge by establishing safety, sustains it rather than retreating from the discomfort it causes, and creates the conditions for the person to resolve it. Uttley also draws on Said’s idea of contrapuntal reading to describe attention to what is not said: the hesitation, the qualification, the claim that does not fit the rest of the account.

Each element translates to the classroom. The weak questioner either praises an answer and moves on, which is unison, or supplies a better answer, which is an imposed line. The expert questioner introduces a related line that tests the answer: a counter-example, an objection, a request to explain where an idea came from. The suspension of dissonance corresponds to what Rowe called the second wait time, holding the pause after a pupil’s answer rather than rescuing the pupil from it. Contrapuntal reading corresponds to hearing the gap between what a pupil has written and what they can say. That last skill has become newly important. Work produced with AI assistance often contains language whose register exceeds the pupil’s spoken command of the ideas, and in conversation the gap is frequently audible.

The analogy has limits that should be stated. Uttley (2026a) describes the coach as working without attachment to particular outcomes, whereas a teacher is accountable for the pupil’s mastery of a defined body of knowledge, and he himself treats counterpoint as a heuristic rather than a rule system. The value of the comparison is that coaching has developed a more explicit vocabulary for the conduct of questions and silence than teacher education usually has, and some of that vocabulary can be borrowed.

5.3 Kinds of silence

Uttley (2026c) distinguishes four kinds of silence in coaching. Reflective silence occurs when a person pauses to process a question. Generative silence is created deliberately by the practitioner after a demanding question. Empathic silence follows a disclosure of something significant. Confrontational silence gently exposes an inconsistency in a person’s account. The first two correspond closely to Rowe’s two wait times and belong in every classroom. Empathic silence has its place in pastoral conversation. Confrontational silence needs care with adolescents and should be used sparingly. The same paper notes that the meaning of silence varies across cultures, so that a pause one pupil experiences as respectful may feel hostile to another. That caution matters for pupils learning English as an additional language and is taken up in Section 8.

5.4 A typology of questions for AI-rich classrooms

Table 1 sets out seven kinds of question that become central once a fluent answer can no longer be taken as evidence of understanding. None is new in itself; good teachers have always asked them. What is new is their position. They move from the edge of practice to its centre, because they are the questions a pupil cannot answer by consulting a machine in advance.

Table 1. Seven kinds of question for AI-rich classrooms

Type of questionWhat it testsExample
ProvenanceHow the pupil reached an answer, and whether the route is their own“Talk me through how you got from this step to that one.”
ConfidenceWhere the edge of the pupil’s understanding lies“Which part of your answer would you be least confident defending, and why?”
BoundaryWhether the pupil knows where a claim stops holding“In what circumstances would this not be true?”
CounterfactualWhether the pupil grasps the structure of an argument“If we removed this premise, what would happen to your conclusion?”
Critique of the artefactWhether the pupil can judge a machine-generated answer“Here is a chatbot’s answer to the same question. Where is it wrong, and how do you know?”
TransferWhether knowledge travels beyond the original context“Where else in the course does this idea apply?”
RevisionWhether the pupil holds a view in proportion to the evidence“What evidence would make you change your mind?”

Asking such questions well demands more of the teacher than asking for recall. The teacher must know the subject well enough to follow a pupil’s reasoning wherever it leads, to recognise a sound answer expressed unusually, and to judge in the moment whether understanding is secure or borrowed. The critique-of-the-artefact question deserves particular mention, because it turns AI from a threat to assessment into material for it. A pupil who can explain precisely where and why a machine-generated answer is wrong has shown understanding that the machine itself cannot supply.

5.5 Implications for teacher development

If questioning moves to the centre of the teacher’s work, it cannot remain a line on a lesson observation form. It needs to be treated as a craft that is rehearsed, observed and coached. Coach education offers a model, since it gives sustained attention to the conduct of questions and to the disciplined use of silence. Teachers would benefit from the same: practice in planning sequences of questions, in holding wait time, in listening for the gap between written and spoken command, and in responding to an answer with a second line of inquiry rather than a verdict. Subject knowledge is the foundation of all of this. A teacher cannot probe what they do not understand, and the case for deep subject expertise in teaching is strengthened, not weakened, by AI.

6. The viva voce: lineage and evidence

6.1 Lineage

The viva voce has a long history. Its roots run back to the disputations of the medieval university, in which claims were tested by argument in front of others, and it survives most visibly in the doctoral examination, where a candidate defends a thesis in conversation with examiners. It has survived for a simple reason: it is very hard to sustain a false impression of understanding for an hour in front of people who know the field.

Two findings from research on the doctoral viva carry over to the school context. Trafford (2003), who observed doctoral vivas across several disciplines, found that examiners’ questions followed discernible patterns and that particular clusters of questions were decisive in determining outcomes. If the kinds of question asked in a viva can be anticipated, they can be taught and prepared for, which matters greatly for fairness. Morley et al. (2003) raised questions about the affective and micropolitical dimensions of an oral examination conducted in private, and about whether it escaped the quality assurance applied to other forms of assessment. Any extension of the viva into schools must take both lessons seriously: the genre can be made transparent, and its conduct must be open to scrutiny.

6.2 Evidence from higher and professional education

Joughin (1998) identified six dimensions along which oral assessments vary: the primary content assessed, the degree of interaction, authenticity, structure, the examiners involved, and the extent to which oral skill itself is being judged. The framework is useful because it shows that “oral assessment” is not one thing. A scripted presentation and an unscripted conversation differ on almost every dimension.

Huxham et al. (2012) randomly assigned undergraduates to answer the same biology questions either orally or in writing. Those examined orally achieved higher marks. Students reported more nervousness before oral assessment, yet many judged it more useful than written tests, and students with dyslexia were particularly likely to favour it. The authors concluded that oral assessment may be more inclusive than is often assumed and can help students develop a sense of professional identity. Sotiriadou et al. (2020) developed the interactive oral, a structured but unscripted conversation built around a realistic scenario, as a form of authentic assessment that also preserves academic integrity.

The evidence is not uniformly favourable. In a review of oral assessment in postgraduate medical examinations, Memon et al. (2010) noted persistent concerns about validity, reliability and fairness, recorded that some countries had moved away from oral examinations for these reasons, and proposed fifteen conditions under which oral assessment can be valid, reliable and fair. Their review is a reminder that the case for the viva rests on design. An oral examination conducted without structure, criteria or moderation can be among the least reliable forms of assessment; one designed with care can be among the most informative.

6.3 Precedent in secondary systems

The proposal that secondary pupils should face a spoken examination is less radical than it may sound in England. In France, every candidate for the general and technological baccalauréat sits the Grand oral, a twenty-minute examination preceded by twenty minutes of preparation, introduced from the 2021 session; from the 2027 session it carries a coefficient of eight in the general route (Ministère de l’Éducation nationale, 2026). In Italy, a 2025 reform of the upper secondary leaving examination reorganised its oral component, the colloquio, around four disciplines specified each year (Avvenire, 2025). In both systems a substantial spoken examination at the end of secondary schooling is unremarkable.

England is not starting from nothing. Speaking assessments in modern foreign languages, the spoken language endorsement in GCSE English Language and the presentation element of the Extended Project Qualification already place pupils in front of an assessor. Policy is also moving in the same direction. The Oracy Education Commission (2024) called for oracy to be treated as a foundational part of education alongside reading, writing and arithmetic, and the Curriculum and Assessment Review recommended an oracy framework alongside the existing frameworks for reading and writing (DfE, 2025). The point is not that England should copy France or Italy, but that a spoken examination at sixteen or eighteen sits comfortably within the range of normal practice in comparable systems.

7. Towards a secondary viva: design principles

If the viva is to enter secondary schools, it must be designed for adolescents, for the realities of school timetables and for the demands of fairness. The following principles draw on the evidence reviewed above.

Anchor the conversation in the pupil’s own work. The interactive oral described by Sotiriadou et al. (2020) draws its questions from material the student has produced. A secondary viva attached to an essay, investigation or project should do the same, so that the pupil is asked to explain and extend their own reasoning rather than to recall unrelated content.

Make it frequent and low-stakes before it is rare and high-stakes. Huxham et al. (2012) found that students were more nervous before oral assessment. The remedy is familiarity. Pupils who have been asked to explain their thinking aloud many times in lessons will meet a formal viva as an extension of ordinary practice rather than an ordeal.

Combine a structured core with disciplined freedom to probe. A set of prepared questions, of the kinds shown in Table 1, provides comparability between pupils and makes the genre transparent, as Trafford’s (2003) findings suggest it can be. The assessor’s freedom to follow up gives the viva its distinctive power, but that freedom should be exercised within agreed limits.

Separate substance from polish. Assessment criteria should state clearly what is being judged. Following Joughin’s (1998) distinction, a viva intended to assess understanding of a subject should not reward accent, confidence or rhetorical flourish, except where oral communication is itself the object of assessment.

Make the conduct open to scrutiny. Recording, second assessors for a sample of conversations and moderation across departments answer the concern raised by Morley et al. (2003) about examinations conducted in private, and meet several of the conditions proposed by Memon et al. (2010).

Plan reasonable adjustments and protect wait time. Pupils with special educational needs, those with anxiety and those learning English as an additional language may need longer preparation, a familiar assessor, written prompts or extended pauses. Generous wait time benefits all pupils (Rowe, 1986) and should be built into the conduct of every viva.

Keep it proportionate. Thirty pupils examined for ten minutes each amounts to five hours of a teacher’s time before moderation. Sampling, rotation across a year and the integration of short oral checks into lessons can keep the burden manageable. An authentication viva need not examine every pupil on every piece of work; the knowledge that any pupil may be asked to discuss any piece of work changes how the work is produced.

Table 2 sets out four forms a secondary viva might take, from the most frequent and informal to the most formal.

Table 2. Four forms of secondary viva

FormPurposeTypical length and stakes
“Explain it to me” conversationFormative check on understanding during a unit; builds familiarity with being questionedTwo to five minutes, within lessons; no formal mark
Disciplinary defenceThe pupil defends a position or solution against an unprepared objection, for example in history, religious education, mathematics or scienceFive to ten minutes; contributes to internal assessment
Critique of the machineThe pupil evaluates an AI-generated answer and explains its errorsFive to ten minutes; formative or internal assessment
Authentication vivaConfirms that submitted extended work reflects the pupil’s own understanding and decisionsTen minutes, sampled across a cohort; supports the authentication of coursework

The authentication viva connects directly with the regulator’s call for more rigorous authentication of candidates’ work (Ofqual, 2026). A short conversation in which a candidate is asked to explain choices made in their coursework is one of the few authentication methods that does not depend on the unreliable detection tools discussed in Section 3.2. It is not proof against every form of malpractice; the same letter records the growing problem of devices in examination rooms, and live oral examination will require its own supervision. It does, however, shift the burden of evidence from the text to the person.

8. Equity, cultural capital and the danger of rewarding fluency

The strongest objection to the argument of this paper concerns equity. In his account of Bourdieu’s theory of cultural capital, Uttley (2025a) argues that educational systems do not function as neutral arenas in which talent alone determines success, but as social fields in which cultural capital acquired through family socialisation shapes educational trajectories. Oral examination is especially exposed to this. Accent, register, ease in conversation with adults and familiarity with the conventions of argument are unevenly distributed, and an examiner may mistake them for understanding. Morley et al. (2003) drew attention to the micropolitics of the private oral examination, and Dawson et al. (2024) warn that measures adopted to protect integrity can damage inclusion. A viva designed carelessly could reproduce privilege under the banner of rigour.

The objection is serious, but it does not settle the matter. First, the alternative is not neutral either. Written coursework produced in an AI-rich environment already advantages pupils with better tools, more adult help and the confidence to use both. Second, spoken language can be taught. The Oracy Education Commission (2024) presents oracy as a matter of equity precisely because the ability to articulate ideas is currently acquired unevenly, largely outside school, and Rowe (1986) found that longer wait time widened the range of pupils who took part. A school that makes structured talk a daily part of teaching reduces the gap that a viva would otherwise measure. Third, the evidence is not all one way: Huxham et al. (2012) found that some students, including those with dyslexia, preferred oral assessment and that it may be more inclusive than written assessment.

The design response follows from these points. Criteria should privilege the substance of reasoning over its polish. Assessors should be trained to recognise and set aside bias related to accent and manner. Questions should be prepared and shared in kind beforehand, so that knowledge of the genre is not the preserve of pupils who have absorbed it at home. Adjustments should be planned rather than improvised. Uttley’s (2026c) observation that the meaning of silence varies across cultures is a reminder that a pupil’s pause may signal thought rather than ignorance.

The pairing of eloquence with truthfulness in the author’s own school framework (Uttley, 2026b) offers a principle for all of this. Eloquence, on this view, is valued only in so far as it serves the truth. A viva built on that principle would reward a pupil who reasons accurately in plain words over one who performs confidently without understanding. That is the opposite of the danger the objection describes.

9. Formation, truth and encounter: a Catholic reading

The argument so far has been made in terms any school could accept. It also has a particular resonance within Catholic education, and the author’s own tradition supplies a way of understanding why the viva matters beyond its value as evidence.

In an earlier paper the author described the authentic Catholic school in terms of the encounter on the road to Emmaus (Luke 24:13–35), as a place of encounter, accompaniment and aletheia, understood as the slow un-concealing of truth through the journey of education (Uttley, 2026b). It is worth noticing that in the Emmaus narrative the stranger who joins the two disciples begins not by explaining but by asking what they are discussing, and lets them tell their own account before he offers his. At its best, a viva has that shape. It is a conversation in which a more knowledgeable person walks alongside a learner, asks before telling, and helps the learner see what they understand and what they do not.

The Catholic School treats human knowledge as a truth to be discovered, and describes the teacher who seeks the truth without restraint as guiding pupils beyond words to the heart of truth (SCCE, 1977, para. 41). It holds that the integration of faith and culture is mediated by the teacher as a person, who teaches by example as well as by word (para. 43), and that schools must form persons who are inner-directed and capable of choosing freely according to conscience (para. 31). A pupil asked to account aloud for their own reasoning is engaged in a small but real exercise of that inner direction. A teacher who conducts the conversation honestly, admitting uncertainty where it exists and following the pupil’s reasoning with genuine interest, models the intellectual integrity the document asks of teachers. The author’s account of Quaker education points to a similar convergence from another tradition, in which the teacher is a facilitator who helps learners discover truth within themselves and in which silence has a central place (Uttley, 2025b).

The same tradition supplies a warning. The 1977 document insists that the results of Catholic education cannot be measured by immediate efficiency (SCCE, 1977, para. 84), and the author has argued elsewhere for tempering the English tendency to privilege performativity and terminal outcomes (Uttley, 2026b). The viva could be captured by that tendency: turned into another instrument of measurement and accountability, rehearsed to the point of script, its questions leaked and its answers memorised. Schools that adopt it should protect its formative character, so that most such conversations are occasions for learning rather than judgement.

10. Limitations and an agenda for research

This is a conceptual paper, and its claims should be read accordingly. Much of the evidence on oral assessment comes from higher and professional education; school-based evidence on the reliability of short oral examinations for adolescents is thin. The evidence on the effects of generative AI on learning is recent and still developing, and the tools themselves change faster than research can follow. The workload implications of the proposals in Section 7 have been estimated rather than measured. Nor does the paper claim that oral examination is proof against AI: devices concealed in examination rooms are already a significant source of malpractice (Ofqual, 2026).

Five lines of research would test the argument. First, trials of short authentication vivas in Key Stages 4 and 5, measuring agreement between assessors and the relationship between oral and written performance. Second, studies of differential outcomes for disadvantaged pupils, pupils with special educational needs and those learning English as an additional language, under different viva designs. Third, studies of professional development in questioning, examining whether coaching-informed approaches change teachers’ practice and pupils’ talk. Fourth, research into pupils’ own experience of oral examination, including anxiety and its reduction through practice. Fifth, workload studies that establish what a sustainable model looks like in a typical secondary school.

11. Conclusion

Generative AI does not end the essay. Writing remains one of the best means of thinking, and pupils still need to learn it. But the essay can no longer stand alone as evidence of what a pupil knows. Increasingly it will need to sit alongside a conversation in which the pupil shows the thinking behind the page.

For teachers, this means that the most important professional skill of the coming decade may be one of the oldest: asking the right question at the right moment, waiting for the answer, and listening closely to what comes back. For schools and policymakers, it means taking seriously a form of examination that England has largely reserved for its most advanced students, and designing it so that it rewards understanding rather than the cultural ease that can masquerade as understanding. The doctoral examiner, the French jury and the good classroom teacher already share this practice. The argument of this paper is that, before long, it will be an ordinary part of secondary education in England as well.

References

Alexander, R. (2020). A dialogic teaching companion. Routledge.

Avvenire. (2025, September 3). Maturità 2026: orale con quattro materie (e commissioni dimagrite). https://www.avvenire.it/politica/maturita-2026-orale-con-quattro-materie-e-commissioni-dimagrite_94452

Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), Article e2422633122. https://doi.org/10.1073/pnas.2422633122

Black, P., Harrison, C., Lee, C., Marshall, B., & Wiliam, D. (2004). Working inside the black box: Assessment for learning in the classroom. Phi Delta Kappan, 86(1), 8–21. https://doi.org/10.1177/003172170408600105

Corbin, T., Dawson, P., & Liu, D. (2025). Talk is cheap: Why structural assessment changes are needed for a time of GenAI. Assessment & Evaluation in Higher Education, 50(7), 1087–1097. https://doi.org/10.1080/02602938.2025.2503964

Dawson, P., Bearman, M., Dollinger, M., & Boud, D. (2024). Validity matters more than cheating. Assessment & Evaluation in Higher Education, 49(7), 1005–1016. https://doi.org/10.1080/02602938.2024.2386662

Department for Education. (2025). Curriculum and Assessment Review final report: Building a world-class curriculum for all. https://assets.publishing.service.gov.uk/media/690b96bbc22e4ed8b051854d/Curriculum_and_Assessment_Review_final_report_-_Building_a_world-class_curriculum_for_all.pdf

Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), Article 6. https://doi.org/10.3390/soc15010006

Howe, C., Hennessy, S., Mercer, N., Vrikki, M., & Wheatley, L. (2019). Teacher–student dialogue during classroom teaching: Does it really impact on student outcomes? Journal of the Learning Sciences, 28(4–5), 462–512. https://doi.org/10.1080/10508406.2019.1573730

Huxham, M., Campbell, F., & Westwood, J. (2012). Oral versus written assessments: A test of student performance and attitudes. Assessment & Evaluation in Higher Education, 37(1), 125–136. https://doi.org/10.1080/02602938.2010.515012

Joughin, G. (1998). Dimensions of oral assessment. Assessment & Evaluation in Higher Education, 23(4), 367–378. https://doi.org/10.1080/0260293980230404

Lodge, J. M., Howard, S., Bearman, M., Dawson, P., & Associates. (2023). Assessment reform for the age of artificial intelligence. Tertiary Education Quality and Standards Agency. https://www.teqsa.gov.au/guides-resources/resources/corporate-publications/assessment-reform-age-artificial-intelligence

Memon, M. A., Joughin, G. R., & Memon, B. (2010). Oral assessment and postgraduate medical examinations: Establishing conditions for validity, reliability and fairness. Advances in Health Sciences Education, 15(2), 277–289. https://doi.org/10.1007/s10459-008-9111-9

Ministère de l’Éducation nationale. (2026). Épreuve orale terminale dite « Grand oral » du baccalauréat général (Note de service du 10 septembre 2026). Bulletin officiel spécial, 4. https://www.education.gouv.fr/bo/2026/Special4/MENE2622694N

Morley, L., Leonard, D., & David, M. (2003). Quality and equality in British PhD assessment. Quality Assurance in Education, 11(2), 64–72. https://doi.org/10.1108/09684880310471489

Ofqual. (2024, April 24). Ofqual’s approach to regulating the use of artificial intelligence in the qualifications sector. https://www.gov.uk/government/publications/ofquals-approach-to-regulating-the-use-of-artificial-intelligence-in-the-qualifications-sector/ofquals-approach-to-regulating-the-use-of-artificial-intelligence-in-the-qualifications-sector

Ofqual. (2026, March 2). Letter to exam board chief executives from Sir Ian Bauckham. https://assets.publishing.service.gov.uk/media/69a5adc1bdca0a628ece181b/Letter_to_Exam_Board_CEOs_from_Sir_Ian_Bauckham_March_2026.pdf

Oracy Education Commission. (2024). We need to talk: The report of the Commission on the Future of Oracy Education in England. https://oracyeducationcommission.co.uk/

Perkins, M., Furze, L., Roe, J., & MacVaugh, J. (2024). The Artificial Intelligence Assessment Scale (AIAS): A framework for ethical integration of generative AI in educational assessment. Journal of University Teaching and Learning Practice, 21(6). https://doi.org/10.53761/q3azde36

Rosenshine, B. (2012). Principles of instruction: Research-based strategies that all teachers should know. American Educator, 36(1), 12–19, 39. https://www.aft.org/sites/default/files/periodicals/Rosenshine.pdf

Rowe, M. B. (1986). Wait time: Slowing down may be a way of speeding up! Journal of Teacher Education, 37(1), 43–50. https://doi.org/10.1177/002248718603700110

Sacred Congregation for Catholic Education. (1977). The Catholic school. Libreria Editrice Vaticana. https://www.vatican.va/roman_curia/congregations/ccatheduc/documents/rc_con_ccatheduc_doc_19770319_catholic-school_en.html

Sotiriadou, P., Logan, D., Daly, A., & Guest, R. (2020). The role of authentic assessment to preserve academic integrity and promote skill development and employability. Studies in Higher Education, 45(11), 2132–2148. https://doi.org/10.1080/03075079.2019.1582015

Trafford, V. (2003). Questions in doctoral vivas: Views from the inside. Quality Assurance in Education, 11(2), 114–122. https://doi.org/10.1108/09684880310471542

Uttley, S. (2025a, November 22). Bourdieu’s cultural capital: Theoretical foundations and educational implications. Koinonia Educational. https://www.koinonia-educational.com/2025/11/22/pierre-bourdieu-and-the-lasting-impact-of-cultural-capital-simon-uttley/

Uttley, S. (2025b, November 25). A Quaker philosophy of education. Koinonia Educational. https://www.koinonia-educational.com/2025/11/25/a-quaker-philosophy-of-education/

Uttley, S. (2026a, May 17). Counterpoint as a framework for executive coaching: Theoretical foundations, coaching applications and manifestations in the coached individual. Koinonia Educational. https://www.koinonia-educational.com/2026/05/17/counterpoint-as-a-framework-for-executive-coaching-theoretical-foundations-coaching-applications-and-manifestations-in-the-coached-individual/

Uttley, S. R. (2026b, August 7). Fifty years on: Reflections on the Catholic Church’s ‘The Catholic School’, 1977, and the contemporary interest in character education. Koinonia Educational. https://www.koinonia-educational.com/2026/08/07/fifty-years-on-reflections-on-the-catholic-churchs-the-catholic-school-1977-and-the-contemporary-interest-in-character-education/

Uttley, S. (2026c, March 21). Silence, shadow and rest: The generative power of absence in coaching, visual art and music. Koinonia Educational. https://www.koinonia-educational.com/2026/03/21/silence-shadow-and-rest-the-generative-power-of-absence-in-coaching-visual-art-and-music/

Weber-Wulff, D., Anohina-Naumeca, A., Bjelobaba, S., Foltýnek, T., Guerrero-Dib, J., Popoola, O., Šigut, P., & Waddington, L. (2023). Testing of detection tools for AI-generated text. International Journal for Educational Integrity, 19, Article 26. https://doi.org/10.1007/s40979-023-00146-z

Subscribe To Our Newsletter

Join a global network of school leaders and receive expert, evidence-based insights directly to your inbox.