When No One Hears the Recording: Typed Versus Spoken Input in App-Processed 360-Degree Feedback

AI-native multi-source feedback platforms can accept speech as raw input: the employee records their feedback into the app, the system transcribes, analyses and consolidates it, and the assessed person receives a written report. No individual recording is ever delivered to anyone. This severs the usual link between input modality and delivery modality. The standard advice that feedback is best delivered rich, ideally as a conversation, is advice about delivery, and it is silent about a recording that no human will ever receive. What employees actually prefer under these conditions is, to our knowledge, unstudied in the 360-degree feedback context.

We asked 56 working professionals a single standardized scenario question: your company runs a 360-degree campaign about a colleague you work with; you can type your feedback or record it as audio; the audio is only processed by the app, and the assessed person receives a consolidated written document, never your recording. Every respondent received the same wording, the same clarification, and one minute to think. 46 of 56 (82%) chose writing; 10 (18%) chose audio.

The reasons are the finding. Writing-choosers cited control over wording and revision before submission, time to think, discretion in shared workspaces, discomfort with their own recorded voice, and unwillingness to leave voice recordings (identifiable, quasi-biometric data) in an employer's system. Audio-choosers cited speed, typing burden and more natural expression, several reasoning explicitly that "the app processes it anyway." Almost none of the reluctance referenced the removed listener. Removing the human listener removes almost none of the resistance to speaking: the barriers sit in the capture moment and in the stored artifact, not in the delivery.

The split replicates survey-methodology findings in a new domain. Offered a choice on narrative open-ended questions, roughly 78–83% of respondents answered by text in a recent experiment, and 94% chose text in an earlier panel study. A theme-by-theme comparison finds no interview theme that contradicts that literature, while two of our highest-ranked themes, self-presentation under assessment stakes and aversion to one's own recorded voice, do not appear in its catalogue of reasons at all. The same literature finds spoken answers longer, more spontaneous and richer in personal experience: the modality that yields the best raw material for a processing pipeline is the one four out of five respondents avoid. We state design implications: text as default, voice as an option, visible transcription with editing before submission, and deletion of raw audio by default.

We are explicit about what this study is and is not. It is a single-question, stated-preference exploration with a convenience sample, coded by a single rater who is also a vendor in this product category. It establishes that the preference exists, that it is strong, and what respondents say drives it. It does not establish behavior in a live campaign, and we specify the study that would (Section 10).