# Consent in voice and video interviews: what to say and when — Criterio Talent > What informed consent to record audio and video in an interview must contain, when to ask, what to keep as evidence, and what happens when it is refused. URL: https://criteriotalent.com/en/recursos/voz-y-video-con-consentimiento/ --- **Issue 08 · April 2026** # Voice and video interviews: what consent has to say, and when you ask for it Consent asked for after the microphone is on is no longer consent. What it has to contain, when to ask, and what to keep as proof that it was given. Method · Consent and recording Published on 29 August 2026 · 5 min read **In short** It has to say what is recorded, what it is used for, who answers for that processing, how long it is kept and with whom rights are exercised, in the language of the interview and before a single second exists. Evidence of that moment is kept — which version of the text was shown, when it was accepted — because consent you cannot evidence is the same as consent you never asked for. There is a quiet way to break an AI interview: ask for camera permission, open the room, and show the notice while the conversation is already running. Technically the text appeared. Practically, the person accepted something that was already happening, and the evidence produced from there carries a defect of origin that no later report repairs. What follows is how to build that prior moment so that it holds up without turning into an eight-screen formality. This is not legal advice: it is product design on ground where the detail matters, and where each implementation has to deliver its own statement in writing. ## A button is not consent “I accept the terms” informs nobody of anything. For an acceptance to mean something, the person has to be able to answer five questions without leaving the screen, and all five have to be answered in the text itself, not behind a link almost nobody opens. The five points that turn an acceptance into informed consent. An illustrative table, not a legal template. | What is declared | Why it cannot be left out | What is recorded | Audio is not the same as audio and video, and the person decides with that difference in hand. | What it is used for | Assessing this application is not the same as training anything, or reusing it in another process. | Who answers for the processing | The contracting company is the controller; the provider acts on its behalf. Blurring that leaves the person unsure who to complain to. | How long it is kept | A concrete period. “As long as necessary” is not a period, it is a way of not saying one. | With whom rights are exercised | Access, rectification, cancellation and objection need a channel that exists and answers. ## When to ask: before the first second The correct order is single and admits no shortcuts: the text is shown, the person accepts, and only then is browser permission requested and the room opened. Inverting it for technical convenience — because it is easier to build the room first and lay the notice on top — produces recordings that started before permission. There is a simple test for whether a system gets this right: if the person closes the tab right after reading the notice and before accepting, no file of theirs should exist anywhere. If one does, consent arrived late no matter what the screen said. ## Granularity, and when all-or-nothing is the honest option Intuition says splitting consent into independent checkboxes is always better: one for audio, one for video, one for keeping the material. Sometimes it is. Other times it is an elegant way of lying. If the system cannot work without recording — because its promise is precisely that every conclusion can be checked against the recording — then offering a box to decline recording and carry on is offering something that does not exist. The honest option is to say that part is indivisible, explain why, and make the exit unmistakable: decline, and stay in the process by another route. Where granularity is real and worth having: reusing the material for another role, keeping it beyond this process, or showing it to somebody outside the deciding team. Those are genuinely separable decisions, and each deserves its own acceptance. ## What happens when the person declines Something has to happen, and it has to be written down beforehand. A system that goes blank on refusal turns a legitimate decision into a dead end, and the person ends up accepting rather than losing the role — which is the exact opposite of freely given consent. The minimum: confirm the refusal was recorded, say the application is still alive, and explain where it continues. The concrete alternative is defined by the company inside its own process; what the system guarantees is that nothing starts without acceptance and that a refusal does not disappear quietly. The whole experience around that moment is in the [March issue](https://criteriotalent.com/en/recursos/experiencia-del-candidato-con-ia/). ## How to keep evidence that consent was given Consent that cannot be evidenced six months later is, in practice, consent that was never asked for. And evidencing it is not storing a boolean set to true: it is being able to reconstruct exactly what that person read. - The version of the text that was shown, identified — not the text in force today. - The moment of acceptance, with its timestamp. - Which application and which role it belongs to. - That the record is immutable: if the notice changes next month, an earlier acceptance cannot end up pointing at the new text. The common error is versioning the notice but not the acceptance. The text gets corrected on legal advice, and every earlier acceptance starts referring to a document nobody signed. The test is the same as for the interview script: [if you cannot tell which version it ran with, you cannot stand behind it](https://criteriotalent.com/en/recursos/entrevista-estructurada-con-ia/). ## The recording that fails closed, and what it costs If the promise is that every conclusion can be checked against the recording, then an interview without a usable recording cannot be accepted. Not because a file is missing, but because the assessment coming out of it will not be defensible, and nobody will notice until somebody asks. The rule worth implementing is a hard one: if the recording cannot start, or if a voice is missing from it at the end, the interview does not count and is repeated. That has a real cost — someone who already gave their time has to give it again — and the cost has to be met head on, by saying so and making the repeat easy. The opposite cost cannot be paid later. A conclusion without evidence enters the record like any other, looks identical in the report, and only surfaces on the day it has to be supported. How that material is kept and deleted is the subject of the [May issue](https://criteriotalent.com/en/recursos/retencion-y-gobierno-de-datos/). ## What this does not solve Impeccable consent does not make legitimate a use that was not. If the material is going to be used for something the person would refuse if they knew all of it, the problem is not on the screen: it is in the use, and no wording fixes that. Nor does it replace review by people who actually practise law. What this text offers are the product decisions that give such a review something defensible to find; the document that counts contractually is the statement each implementation delivers in writing, and its terms are set by the responsible company, not by the platform. The general position is in [security and data governance](https://criteriotalent.com/en/seguridad/). ## Questions about this issue ### Is showing the company privacy notice enough? The notice says how the company handles data in general. What is missing is what is specific to this moment: that they will speak with a system, that audio and video are recorded, and what happens if they decline. These are usually two texts with two different jobs. ### Can consent be collected in the invitation email instead of in the room? It can be previewed there, and it should be. But the acceptance that counts is the one immediately before recording, on the same device and with the same text: days can pass between the email and the interview, and somebody else may open it. **To keep reading on this site** ### [Solutions](https://criteriotalent.com/en/soluciones/) How it is configured for high volume, technical profiles, leadership, and multi-round processes. ### [Platform](https://criteriotalent.com/en/plataforma/) How it runs the interview, follows up, and cites the evidence behind each conclusion. ### [Integrations](https://criteriotalent.com/en/integraciones/) How your ATS requests the interview and receives the report, with nobody retyping anything. ### [Contact](https://criteriotalent.com/en/contacto/) A 30-minute demo on a real role of yours. **Other issues** **Method · Candidate experience** ### [How to design an AI interview the person does not experience as an insult](https://criteriotalent.com/en/recursos/experiencia-del-candidato-con-ia/) Almost everything that makes an automated interview hateful is a design decision, not a technical limit. Seven decisions, with the uncomfortable one last. **Method · Retention and data governance** ### [Retention of interview recordings: how long to keep them, where, and how to actually delete them](https://criteriotalent.com/en/recursos/retencion-y-gobierno-de-datos/) The period is the easy part. The hard part is deletion that actually happens, leaves a record, and does not block itself — which is exactly what happens when storage expires before the promise does. **Method · Interview design** ### [How to design a structured AI interview that leaves useful evidence and keeps the decision human](https://criteriotalent.com/en/recursos/entrevista-estructurada-con-ia/) Eight design decisions, in order. The first is the one almost everyone skips: you do not ask a model to judge, you ask it to document. **Next step** ## See it with a role of yours on the table. Thirty minutes: an interview is defined from your job post, walked through the way the candidate sees it, and a report is read with its evidence.