One of the hardest assessment problems in the AI era is finding a format that gives us credible evidence of student learning without turning every teacher into an investigator. Live oral exams can help, but they take time to schedule and conduct. Asynchronous oral assessments offer an interesting middle ground: students receive a question, get a short preparation period, and record a brief response without meeting the instructor live.
Pentland, Lowenthal, and Krier examine this format in two university courses. The study belongs within the wider conversation on AI and assessment because it asks whether oral evidence can be collected at scale while preserving student explanation, engagement, and professional communication.
Summary of the Study
The researchers conducted two small studies using a web-based platform developed by the first author. Students accessed questions through their learning management system, revealed each prompt immediately before answering, and recorded time-limited video responses.
Questions were selected randomly from a larger pool, and instructors graded the recordings with question-specific rubrics. In the first study, students in an intermediate accounting course could complete an oral task for extra credit before an in-person exam.
In the second, students in a data analytics course completed graded oral assessments before their midterm and final exams, followed by a survey about preparation and experience.
The first study produced encouraging patterns, but no reliable relationship between oral-assessment participation and later exam performance. The second study found that students earned higher marks on the oral assessments, and performance across the oral and multiple-choice formats was positively related. That result needs careful interpretation.
The oral tasks were open-ended, completed remotely, and graded with rubrics; the comparison exams used multiple-choice questions under in-person conditions. Higher scores on one format do not establish that it measures learning more accurately.
The relationship between formats provides some preliminary evidence that they capture overlapping knowledge, though it is not enough to confirm validity on its own. This is close to the argument behind assessment twins in the AI era: different forms of evidence may strengthen an assessment system when their relationship to the intended learning is clear.
Students reported preparing differently for the oral assessments. They described organizing their ideas more carefully, practising aloud, self-quizzing, and engaging more actively with the content. They also viewed oral responses as relevant to workplace communication.
These are perceptions from a small voluntary survey, so they do not demonstrate deeper learning or improved communication skills. They do suggest that assessment format can change what students do before the exam. That may be one of the strongest reasons to test asynchronous oral work: it asks students to retrieve, organize, and explain knowledge in their own words instead of recognizing an answer from a list.
The limitations are substantial. Both studies came from courses taught at one university, the samples were small, and the first relied on voluntary extra-credit participation. The study did not measure cheating, so its academic-integrity benefit remains a plausible design claim, not an observed effect.
Grading still required the instructor to review every response, and the timed video format may introduce anxiety, technology problems, language demands, and accessibility barriers. The first author also developed the platform, holds equity in a company applying this technology, and has a related patent application pending. The disclosure is welcome, and independent replication with larger, more varied student groups is especially important.
Practical Insights
I can see asynchronous oral assessments working best as one layer of evidence, especially when a course already relies heavily on take-home writing. They do not need to replace essays or written exams. A short oral explanation attached to a larger assignment can help the teacher see whether the student understands the decisions, evidence, and concepts behind the submitted work.
My earlier review of oral exams in an AI world discusses the richer follow-up available in a live conversation; the asynchronous version gains flexibility but gives up that opportunity to probe an unclear answer.
For teachers considering the format:
- Begin with a low-stakes practice round so students can test the technology and become familiar with speaking under a time limit.
- Write prompts that require explanation, application, or justification. A factual recall question gains little simply because the answer is spoken.
- Match preparation and response time to the intellectual demand of the prompt. Complex reasoning needs more space than definition or recall.
- Use a clear, question-specific rubric and share it before the assessment. Grade the quality of understanding and explanation, not accent, polish, eye contact, or performance style unless those are explicit learning outcomes.
- Plan alternatives and accommodations from the beginning. Some students may need extra processing time, another response mode, assistive technology, or a setting that reduces anxiety.
- Treat authorship as a claim that still needs testing. A recorded response may make undisclosed assistance harder, but students could rehearse with AI beforehand or receive help outside the camera frame. Question design, randomization, limited preparation, and comparison with other course evidence all matter.
The study gives us a promising assessment design. Its real value may lie in broadening the evidence teachers collect. When written work, oral explanation, and independent performance point in the same direction, confidence in what a student knows becomes much stronger.
Reference
Pentland, S. J., Lowenthal, P. R., & Krier, K. (2026). Asynchronous oral assessments: Enhancing integrity, engagement, and communication in the AI era. Educational Technology Research and Development. Advance online publication. https://doi.org/10.1007/s11423-026-10677-0
