When a spoken answer is the right test
Ask for a spoken answer when the job is spoken. Sales, customer service, coaching, interview practice and many trade and care roles depend on explaining something out loud, clearly and without notes. A written answer tests a different skill. If the competency is "can explain the return policy to an upset customer", a recording of them doing it is closer to the real thing than an essay about it.
Mark what was said, not how it looked
The fairness risk with recorded answers is that markers react to things the rubric does not mention: confidence on camera, accent, the room someone recorded in, the quality of their microphone. The simplest control is to mark from a transcript. Reviewers read the words against the rubric, and can play the recording only when the transcript is unclear. If tone or delivery genuinely matters for the role, make it an explicit criterion with its own levels, rather than letting it leak into every mark.
Set up the question properly
- Give the scenario in writing, so nobody is marked down for mishearing the question.
- Say how long the answer should be, and set a time limit that matches.
- Decide how many recording attempts are allowed, and enforce it, so every candidate has the same chance.
- Offer audio-only unless you are assessing something visible.
- Use the same rubric for every candidate, with levels written before any answers come in.
Decide what happens to the recordings
Recordings of people’s faces and voices are personal data, and video can reveal more than you need. Decide before the first answer comes in: do you need the recording after the transcript exists? If the transcript is what you mark, deleting the recording once it has been transcribed reduces what you hold. Whatever you choose, set a retention period, tell candidates in plain words, and apply it automatically.
Keep a person signing off
Whether a colleague or an AI tool drafts the marks, a named assessor should review them against the evidence, change anything they disagree with, and release the result. Record every change and who made it. For spoken answers this matters even more, because a transcription error can turn a good answer into a poor one on paper.
How Scorafy handles spoken answers
Scorafy accepts audio and video answers, transcribes them on the server, and marks the transcript against your rubric, quoting the words behind each mark. You can let candidates choose audio-only, limit recording attempts, and switch on deletion of the recording at submission once the transcript exists. An assessor reviews and releases every result. The details are on the GDPR page.
Common questions
- How do you mark a recorded spoken answer fairly?
- Transcribe it, then mark the transcript against a rubric that says what a good answer contains. Marking the transcript means the score reflects what the person said, not their appearance, background, camera quality or nerves on video. A person should review the marks and release the result.
- Can AI mark audio and video answers?
- AI can transcribe the recording and draft marks for the transcript against your rubric, quoting the words behind each mark. Treat that as a first draft for an assessor to review, not a final result, especially for spoken answers where transcription errors are possible.
- Should candidates record audio or video?
- If the assessment is about what someone says, audio is usually enough, and it collects less personal data. Video only adds value when you are genuinely assessing something visible, such as a demonstration. Letting candidates choose audio-only for spoken questions is a sensible default.
- What should happen to the recordings after marking?
- Decide before you start. If the transcript is what you mark, you may not need to keep the recording at all once the transcript exists. Whatever you choose, set a retention period, tell candidates, and apply it automatically rather than by hand.