Oral Assignments: How Teachers Can Collect Spoken Answers Online

A take-home essay no longer tells you who did the thinking. In a 2024 study at a UK university, 94% of AI-written exam answers slipped past markers undetected. Oral assignments online are the cheapest way to get that signal back: you ask students to explain their own work out loud, and you listen.

Detection is losing. Explanation still works. This post covers the research on oral assessment, reusable prompts and a rubric table. It also covers the privacy rules for recording minors, and how to run it all through a voice form.

Why oral assignments are back

Oral examination is old. The viva is how many universities have long examined theses, and language teachers have always made students speak. What changed is that written work lost its default status as proof of learning.

AI-written answers pass as student work

The clearest test so far comes from the University of Reading. Peter Scarfe and colleagues wrote exam answers with GPT-4 and submitted them through 33 fake student accounts across five undergraduate psychology modules. Markers were not told. 94% of the AI submissions went undetected, and on average they earned grades half a boundary higher than real studentsScarfe, Watcham, Clarke and Roesch, PLOS ONE, 26 June 2024. Unsupervised at-home exams, 5 undergraduate modules, 1 UK university. Markers unaware of the test..

Two things make this result hard to dismiss. The answers were submitted essentially unmodified, not edited to look human, and they still outscored real students most of the time. We think any written take-home task now needs a second signal before it carries serious marks.

Explaining is harder to outsource than writing

An oral answer asks the student to produce reasoning in real time, in their own words, about something specific. That is a different task from submitting a polished text. The answer depends on what is in the student's head at the moment of speaking.

This is why we would put the effort into oral tasks before buying detection software. Detectors produce a probability. A two-minute explanation produces evidence you can hear.

A recorded oral assignment is not a live viva

An oral assignment online is a spoken task that students complete remotely and submit as a recording. It sits between a written assignment and a live viva. The student hears or reads a prompt, records an answer in the browser, listens back, and submits.

It is not a live oral exam. A live viva lets you ask follow-up questions, and that interaction is where most of its integrity value comes from. A recording cannot answer back. Be clear with yourself about which one you are running.

The live viva wins on authentication, and we will not pretend otherwise. The recorded version wins on scale and timing: 120 students can each answer at a moment that suits them, and you can listen in batches.

What the research says about oral assessment

The evidence on oral assessment predates generative AI, and that is useful. It tells you about learning and fairness, not just cheating.

Students score higher, and no group was shown to lose out

Mark Huxham, Fiona Campbell and Jenny Westwood ran a comparison at Edinburgh Napier University, published in Assessment and Evaluation in Higher Education. A cohort of 99 biology students was randomly split between oral and written versions of the same questions, and both cohorts studied scored significantly higher in the oral formatHuxham, Campbell and Westwood, Assessment and Evaluation in Higher Education 37(1), 125 to 136, 2012. Cohort 1: n = 99, randomly split into oral and written groups on the same biology questions. Cohort 2: n = 29, sat both formats.. The authors found no evidence that particular groups of students were disadvantaged by the oral tests.

The same students reported being more nervous about oral assessment. Many still rated it as more useful than written work, and a recurring theme was that it felt authentic and professional. The authors concluded that oral assessments may be more inclusive than written ones.

Preparation decides whether orals feel fair

Dublin City University piloted interactive oral assessment in 2020 to 2021, in computing, education, French literature and aviation. Monica Ward and colleagues reported the results in Innovations in Education and Teaching International. The pilot covered 322 students in 7 modules, and 140 answered an anonymous survey.

82% of responding students said the orals were good or very good at encouraging them to act with academic integrity, and 77% said the same about discouraging cheatingWard et al., Innovations in Education and Teaching International 61(5), 930 to 947. Survey of 140 of 322 students (43%). Share rating each feature good or very good.. These are self-reports, so read them as acceptance, not as measured misconduct rates.

The more practical finding is what helped students prepare.

Anxiety is real, and design reduces it

Both studies report nerves. In the Dublin pilot, some students described the orals as stressful because they were unsure what to expect. Huxham's students were more nervous about orals and still valued them. Our reading: anxiety is a design problem first. A published rubric, a practice prompt and the chance to re-record take most of the surprise out of a recorded task.

Oral assessment ideas you can use this week

Good oral prompts ask for explanation, not recall. The strongest ones point at something only that student did, which is also what makes them hard to outsource. Here are prompts grouped by level. Adapt the wording and keep the time caps short.

Primary and middle school

• "Read your favourite sentence from your story aloud. Then tell me why you chose it." Time cap: 60 seconds.
• "Explain how you got your answer to question 3, step by step." Time cap: 90 seconds.
• "Teach a younger child what a fraction is. Use something from your kitchen." Time cap: 90 seconds.

Short, concrete prompts work best at this age. Reading aloud also gives you a fluency sample at no extra cost.

Secondary school

• "Pick one source from your essay. Why did you trust it more than the others?" Time cap: 2 minutes.
• "Argue the side of this debate you disagree with, as well as you can." Time cap: 2 minutes.
• "Describe the mistake you made in last week's lab, and what it taught you." Time cap: 2 minutes.
• In languages: "Describe your weekend in French. Use at least two past tenses." Time cap: 90 seconds.

For language teaching, a spoken answer is the learning outcome itself. We cover that case in running an online speaking test.

University and adult learners

• "Walk me through one design decision in your project, and the alternative you rejected." Time cap: 3 minutes.
• "Explain this result to a client who has no background in statistics." Time cap: 3 minutes.
• "What would change in your argument if this assumption were false?" Time cap: 3 minutes.

Questions about the student's own submitted work are the most valuable kind. Pair a written assignment with a short recorded defence of it, and you test both writing and understanding.

A rubric for spoken answers

Write the rubric before you listen to the first recording. Share it with students before they record, since the Dublin students rated the rubric as their most useful support. Keep it to four or five criteria, all on the same scale.

Notice what is missing: accent, fluency in a second language and confidence. Mark those only when speaking skill is the outcome you teach. Otherwise you turn a test of content into a test of background, which is exactly what the CAST guidance below warns against.

How to run an oral assignment with a voice form

You do not need a learning platform plugin or a separate recording app. A web form with a voice field covers the whole flow, and students answer from the phone they already carry. Here is the process in our own tool, Would You Rather. It carries over to any form builder with a native voice answer.

Practical settings that matter

Students can play back and re-record before submitting. That lowers anxiety, and it also means the recording is a considered answer, not a cold one. Plan your prompts with that in mind.

If students work over several sittings, an in-progress response can be resumed. For a class QR code on the board, see using QR codes with forms. For marking in a spreadsheet, see sending form responses to Google Sheets. A co-teacher or moderator can be added to the form as a collaborator with viewer access.

What the tool will not do for you

Would You Rather stores each recording and links it to the student's other answers. It does not transcribe recordings and it does not grade them. Listening is the marking. A 2-minute cap means roughly 2 minutes of listening per student, so a class of 30 is about an hour before you write any feedback.

That is slower than an auto-marked quiz and often faster than reading 30 essays. We think it is time well spent, because what you hear is harder to fake than what you read.

Accessibility: voice as one option, not the only one

The CAST Universal Design for Learning Guidelines 3.0 ask teachers to offer several ways for learners to show what they know. Consideration 5.1 is to use multiple media for communication. Consideration 5.4 is to address biases related to modes of expression and communication. Consideration 4.1 is to vary and honour the methods for response.

Read together, those support oral assignments for many students and rule them out as the only route. For a student with dyslexia or slow handwriting, speaking can remove the barrier that a written task puts in front of the content. For a deaf student, a student with a speech disability or a student with severe anxiety, a spoken task can become the barrier.

Design choices that keep it fair

• Offer an alternative route in the brief: typed answer, a live conversation, or sign language on video where that is the student's language.
• Let students answer in the language they think in when language is not the outcome. See collecting answers in Hindi, Tamil or any language.
• Give a practice prompt that carries no marks, so the first recording is not the graded one.
• Record the question as audio and show it as text, so students can use either.

Our view is that voice should widen the ways a student can succeed, never narrow them. If one student must type while 29 speak, that is UDL working, not a failure of the assignment.

Privacy rules for recording minors

A voice recording is personal data, and for children it carries extra rules. This section is orientation, not legal advice. Check your school or district policy first, because it usually sets stricter limits than the law.

United States: COPPA

The FTC's COPPA guidance covers children under 13. It states that audio recordings containing a child's voice count as personal information. A school can consent on behalf of parents, but the FTC limits that to the educational context: the operator must collect the data for the school's use and benefit, and for no other commercial purpose.

India: the DPDP Act

India's Digital Personal Data Protection Act, 2023 defines a child as anyone under 18. Section 9 requires verifiable parental consent before a data fiduciary processes a child's personal data. It also bans tracking, behavioural monitoring and targeted advertising directed at children. The Act lets the government exempt some classes of data fiduciaries and purposes, so check the notified rules for how schools are treated.

A short checklist for teachers

• Collect only a name or student ID and the recording. Skip phone numbers and addresses.
• Turn off anonymous public access; share the link only with your class.
• Tell students and parents what the recording is for, who will hear it, and when you will delete it.
• Give collaborator access only to people who mark or moderate.
• Read the tool's data terms. We make no compliance claim for any tool, ours included; your institution decides.

Where recorded oral assignments are weakest

The evidence above is mostly about live oral assessment. Recorded oral assignments borrow its spirit, not its full strength. Five limits deserve plain statement.

First, a recording proves the voice, not the thinking. A student can ask an AI tool for an answer and read it aloud. Personal prompts and short caps make that harder, and a scripted reading often sounds scripted. It does not make it impossible. If marks carry real weight, sample 10 to 20% of students for a 5-minute live follow-up on their recording.

Second, the studies are small and local. Huxham's randomised cohort was 99 students in one subject at one university. The Dublin figures are self-reported by 140 students who chose to answer a survey. Neither measures cheating directly.

Third, the Reading study tested unsupervised at-home exams in one UK university. Detection rates in your school, subject and age group may be different.

Fourth, listening takes time, and there are no shortcuts in our product: no transcription, no auto-scoring. Large cohorts need a marking plan before launch.

Fifth, some students will find speaking harder than writing. That is the strongest argument against making orals compulsory, and it is a good one. The fix is choice, not abandoning the format.

How to try it this week

Pick one written assignment you have already set. Add a single recorded follow-up: "In 2 minutes, explain the most important decision you made in this piece of work, and why." Publish the rubric with it.

Then compare what you learn from 5 recordings with what you learned from the 5 matching essays. If the recordings told you something the essays did not, extend it to the next unit. If you want more on the mechanics of spoken answers, read why spoken answers run longer than typed ones and how long voice answers usually are.

References

1. Scarfe, P., Watcham, K., Clarke, A. and Roesch, E., A real-world test of artificial intelligence infiltration of a university examinations system: A 'Turing Test' case study, PLOS ONE, 2024. Used for the undetected share and grade comparison of AI-written answers.
2. Huxham, M., Campbell, F. and Westwood, J., Oral versus written assessments: a test of student performance and attitudes, Assessment and Evaluation in Higher Education 37(1), 2012. Used for the randomised oral vs written comparison, inclusiveness and student nerves. Abstract only; we could not open the full text.
3. Ward, M., O'Riordan, F., Logan-Fleming, D., Cooke, D., Concannon-Gibney, T., Efthymiou, M. and Watkins, N., Interactive oral assessment case studies: An innovative, academically rigorous, authentic assessment approach, Innovations in Education and Teaching International 61(5), 2024. Used for student ratings of integrity and preparation supports.
4. CAST, Universal Design for Learning Guidelines version 3.0, Action and Expression, 2024. Used for considerations 4.1, 5.1 and 5.4.
5. US Federal Trade Commission, Complying with COPPA: Frequently Asked Questions, accessed 8 October 2026. Used for the age threshold, voice recordings as personal information, and school consent (FAQ N.1 and N.2).
6. Government of India, The Digital Personal Data Protection Act, 2023, section 2(f) and section 9. Used for the definition of a child and the parental consent and tracking rules.

The weakest point in this source base: the oral assessment studies examine live, examiner-led orals in higher education, so applying them to recorded assignments in schools is our inference, not a tested finding.