Sign up free

The oral exam is back: what a recorded answer keeps and gives up

What universities and a schools authority have published on vivas and interactive orals since 2023, what research finds, and what a recorded answer keeps.

The EdProcto editorial teamWritten and checked against the research it cites. Last checked . How the guides are made

Guides · Integrity and cheating · · 13 min read

Why is the oral exam back, and can a recorded spoken answer do the same job online?

The oral exam is back because a typed answer no longer shows who wrote it, and a question answered aloud is harder to hand to a chatbot. Research supports a structured oral: set questions, a marking scheme and a second examiner for fails. A recorded spoken answer keeps one question, one clock and a record to re-mark, and gives up the follow-up question.

The oral exam is back. Since chatbots arrived, universities have published guidance on vivas, on defending submitted work aloud and on "interactive orals", and New Zealand's qualifications authority advises school teachers to ask a student about their own work. The reason is plain: a typed answer no longer shows who wrote it.

This guide gives the evidence: what named institutions have published since 2023, what research finds on oral assessment, and what the oral's online form, a recorded spoken answer, keeps and gives up. EdProcto publishes it and makes that form: the spoken answer template, recorded in the browser and marked by the person who listens.

The steps are in two other guides: wording and marking a spoken question and recorded answers for a large class.

Why is the oral exam back since chatbots arrived?

Because an answer given aloud, to a question met on the spot, is harder to hand to a chatbot than a typed one, and institutions have said so in print. We read each page below at its source on 10 October 2026.

  • The University of Auckland (2026) opens its staff page on interactive oral assessment by asking how to test learners in the age of AI. It says the unscripted form makes pre-generated or AI-assisted answers difficult, and that assessors need training and sessions should be recorded for moderation.
  • Dublin City University (2026) defines an interactive oral as a two-way, unscripted conversation on a real-world scenario, and says it is not a viva, which it describes as questions and answers under strict exam conditions. It asks for a rubric designed beforehand.
  • Sotiriadou and Logan-Fleming (2025), of Griffith and Torrens universities, write that interactive orals have reached more than 30 universities since 2015, and that AI has left written and prerecorded assessments more open to impersonation.
  • The University of Westminster (n.d.) tells students that a viva with two academics may be held to verify that submitted work is authentic, and is not an allegation of misconduct.
  • The New Zealand Qualifications Authority (n.d.), writing for schools, lists follow-up conversations among its checks that work is a student's own, such as asking "how did you come to that conclusion?"

A department's wider choices are in assessment design after chatbots.

What does research say about oral assessment?

That an oral is as sound as its design, and an unstructured one marked on impression is the weak form. Where we read only an abstract, we say so.

Is an oral exam valid and reliable?

It can be, when it is structured. Nallaya et al. (2024) screened 2,657 articles and kept 17 studies of oral assessment in higher education. Validity, reliability and the power to reduce integrity breaches all depended on whether the oral was designed, scaffolded and run well (abstract read).

Anastakis et al. (1991) show what structure buys. They gave 23 surgical residents four set scenarios, each with five or six questions and a marking scheme for every question, scored by two examiners independently. The examiners' scores correlated at 0.78 to 0.91 and overall reliability was 0.75, both above the figures for traditional orals (abstract read).

What biases an examiner, and what reduces it?

Delivery does, when the criteria let it. Rowland-Morin et al. (1991) had actors re-enact transcripts of real students' oral answers in two styles: direct eye contact at a moderate pace, or indirect eye contact at a slower one. Seventy-eight surgery teachers rated 255 tapes, believing them real. The same words scored significantly higher in the first style, in every category except decision making (abstract read).

Burchard et al. (1995) used the method with 140 raters: verbal style, dress and content moved the scores, and the rater's age and experience did not. Raters agreed 86% of the time on good performances and 67% on poor ones, so the authors advise a second examiner before a performance is judged poor (abstract read).

What reduces examiner bias, on this evidence: set questions, a marking scheme for each, and a second listener for any fail. Our reading: a sound recording hides dress and eye contact from the marker but not pace or voice, so the rubric must leave those out.

How anxious does an oral make students?

More anxious than a written paper, on average. Laurin-Barantke et al. (2016) surveyed students at one German university. In the 96 complete responses, test anxiety was higher for oral than for written exams, and only the oral kind rose with social anxiety (abstract read). A practice question that does not count is the cheapest answer.

Who can an oral put at a disadvantage?

Students who stammer, students who are deaf or hard of hearing, and students speaking a second language.

  • A stammer. Daniels et al. (2011) surveyed 328 instructors at two US universities. Knowing more about stuttering went with more positive attitudes to students who stutter, and the instructors asked for more information on accommodating them (abstract read).
  • Deaf or hard of hearing. The Australian Disability Clearinghouse on Education and Training (n.d.) advises alternatives to assignments built on interviews, extra time in examinations, and questions signed for students who prefer it.
  • A second language. Roberts et al. (2000) observed 24 half-hour orals of a UK medical membership exam and concluded that candidates from ethnic minorities and those trained abroad may meet hidden difficulties: the oral judged a way of talking along with the knowledge. They advised training examiners and publishing sample questions with a range of answers.

Agree any adjustment beforehand: more time, a typed version, or a conversation in person. Your institution's policy and the law where you teach decide; nothing here is legal advice.

What does a live oral cost in marking time?

Ten to fifteen minutes a student. Dublin City University (2026) plans about 15 minutes for each interactive oral, ten to talk and five to mark, and at most 20 in a day. The University of Auckland (2026) gives 10 to 15. A class of 30 needs five to seven and a half hours of appointments.

What does a live oral show that a recorded spoken answer cannot?

A live oral shows how a student answers the next question, the one nobody could prepare. A recorded spoken answer gives that up. In return every student gets the same question and clock, the class sits at once, and each answer is a record.

Live oral Recorded spoken answer
Follow-up The examiner asks again None, so the prompt must ask for the reason
Same for everyone Only with questions set beforehand Yes
Time 10 to 15 minutes each, by appointment One sitting in EdProcto, marked afterwards
A record Only if the session is recorded Every answer, to hear again or pass to a second marker
A script Hard to keep up under questioning Possible

Can a recorded answer be read from a script?

Yes, and the marker may not hear it. Canagasuriam and Lukacik (2025) gave 245 people a recorded job interview. Those assigned to read ChatGPT's answers word for word, or to adapt them, scored considerably higher on content and overall than those without it. Ratings of delivery did not differ, though ratings of honesty were lower (abstract read).

An interview is not an exam, but the lesson carries: a recording can be rehearsed or read from a screen. What reduces that is the question, the clock, the room, and a live follow-up with a few students drawn at random.

How do you write an oral question that cannot be read from a script?

Ask for something only this student has, at a moment nobody could plan for. Three patterns help.

  1. The student's own example. "Give an example of opportunity cost from your own week, and say what you gave up." A script written beforehand holds the definition, not the week.
  2. A walk through their own working. "Take Monday's sheet. Talk me through your working for question 4, line by line. If you now see a slip, say where." The sheet was written in class, in front of you.
  3. A prompt seen only at the moment. A graph, a source or a worked solution with one error appears with the question, under a clock too short to type it elsewhere and still answer.

None of the three is a lock: with a second device on the desk, a chatbot answers an unseen prompt in seconds. When the mark matters, sit the recording in your own room.

What should a rubric for a spoken answer reward?

Content, reasons, the student's own material, and whether a listener can follow. This four-row rubric is ours: a working tool built on the studies above, not a validated instrument.

Row 2 1 0
Right Correct and complete Partly right Wrong or missing
Reasoned Gives the why at each step Results without reasons No reasons
Their own Uses their own example or working, or the prompt shown, in detail Mentions it in passing Could be about anyone's work
Clear Understood on first hearing Understood with effort Not understood

Accent, pace, hesitation and a restarted sentence are in no row, so they cost nothing. An answer scoring under half goes to a second listener, as Burchard et al. (1995) advise for poor performances.

What does a recorded oral look like for a class of 30?

Two recorded answers each, about two and a half hours of marking, and a live follow-up with five students. The example is invented: a maths class of 30, a week after a sheet on linear equations worked in class.

  1. In EdProcto, add two spoken answer questions worth eight points each, with the four rows as each one's rubric. Question 1 is the walk through question 4 of the student's own sheet.
  2. Type question 2 as the prompt: "A student wrote 2(x + 3) = 14, so 2x + 3 = 14, so x = 5.5. Say which step goes wrong and why, then give the right answer."
  3. Set "Time limit for each question" to "Set", "Seconds a question" to 120 and "Minutes overall" to 6. Switch off "Allow going back to change an answer", leave "Whole-quiz retakes" at "None", and set "Show the score" to "When released".
  4. Run one practice question the lesson before, so every browser has asked for its microphone.
  5. Sit it in two halves of 15, with headsets, in your own room.
  6. Mark with the rubric, then ask five students drawn at random to talk one answer through with you.

The time, on our assumptions: five minutes a student (three of listening, one and a half to score, half to open and save) makes 150 minutes, and five follow-ups of four minutes add 20. Live orals for all 30, at Dublin City University's 15 minutes, would take seven and a half hours.

How does EdProcto run a recorded spoken answer?

EdProcto is the best tool for the oral's online form because it supplies the parts of the structured oral that Anastakis et al. (1991) tested: one set prompt for every student, your marking scheme beside every answer, and a recording that can be played to a second listener.

The shapes. Among the templates, the category "Voice and performance" has five ready shapes. Spoken answer is "A recorded answer, assessed for what it says." With it come Pronunciation, "Say the word or phrase; the recording is assessed."; Read aloud, "Given text read aloud; fluency, accuracy, pace."; Presentation, "A longer spoken piece on a topic, recorded."; and Sing or play, "A vocal or instrumental performance, recorded."

The student. The student presses "Start recording" and the browser asks for the microphone. The clock reads "Recording 0:42". "Stop recording" brings "Saving the recording…", then a player to hear the take, and "Record again" replaces it while the question is open. A recording still running when the question's clock moves the student on, the overall time ends or the quiz is handed in is stopped and kept by itself. One that nears the 25 MB the server takes is stopped and kept too, with a line saying so. The microphone is released whenever a recording stops.

The marker. The recording is kept with the attempt. On the grading page a recorded answer shows "to grade" and a player, with your "Model answer" and "Rubric" under it, a "Points" box that takes half marks, "Note for the student" and "Save my grade". "Ask Procto AI" appears on typed open answers; a recorded answer is marked by the person who hears it.

The conditions. "Minutes overall" and "Seconds a question" set the clocks. "Whole-quiz retakes" says whether a student may sit it again, and "Hours between retakes" how long they wait. Add the integrity measures the stakes need.

What you tell students. A voice is personal data. State your school's own retention period for recordings and who will listen, and tell students before they record.

The plans. Tutor opens all 111 templates, the recorded ones among them, for 100 students a quiz; Educator takes 500 and adds the full analysis; Free has the core templates and 30 students a quiz. See pricing. A recording is evidence of what was said, never proof of who prepared it, and the score is yours to release.

Quick questions

Is an interactive oral the same as a viva?

No. Dublin City University (2026) calls the first a free-flowing conversation on a scenario and the second questions and answers under exam conditions.

Is a recorded spoken answer as secure as a live oral exam?

No. It can be rehearsed or read from a screen, and Canagasuriam and Lukacik (2025) found scripted ChatGPT answers scoring higher. Tie the prompt to the student's own work, keep the clock short and use your own room.

Should accent or fluency count in an oral assessment?

Only where the language is the subject. Rowland-Morin et al. (1991) found delivery moving the marks for identical words, so keep it out of the rubric.

Can EdProcto run an oral exam online?

EdProcto runs its recorded form: the spoken answer template records in the browser, keeps the recording with the attempt, and sets it beside your rubric to mark by listening.

References

Integrity

Five levels, from none to a full proctoring stack, each measure switched on by itself with its own consequence.

Light on Free, Strict on Educator

See the five levels

Read next