Built for UK and EU desks
Nobody in the room is surprised she is there
An AI in a candidate interview is a serious thing, and pretending otherwise is how products in this category get agencies into trouble. She is named in the participant list, admitted by the host, announced out loud on arrival, and silent for the rest of the meeting unless you have deliberately asked her to speak.
The problem
The tool is easy. The room is not.
A candidate notices a bot in the participant list, and nobody in the meeting can explain what it is.
Recruitment sits at the sharp end of this. Interviews are recorded conversations about someone's livelihood, held under UK and EU expectations of consent, and increasingly under rules about AI in hiring decisions. A generic notetaker sold to sales teams was not designed with any of that in mind — it joins under a personal account name, records by default, and hands you a score with no evidence and no human in the loop. That is a compliance conversation waiting to happen, in your name rather than theirs.
What goes wrong
- A bot in the roster nobody can explain
- Recording on by default in a two-party jurisdiction
- AI output that looks like a decision, not a draft
What it costs
- Candidates dropping out of a process that feels processed
- Client confidentiality raised at exactly the wrong moment
- An interview record that would not survive a challenge
Why it matters more here
- Hiring decisions carry rules a sales call does not
- Agencies carry the client's reputation as well as their own
- The transcript is evidence, whether you meant it to be or not
How we solve it
1
Visible, named and admitted
She appears as Simi followed by your agency's name, with a face rather than a grey bot icon, and she waits in the lobby until the host lets her in. That admit click is the consent moment, and it belongs to whoever is running the meeting.
- Named for your agency, so nobody wonders whose bot it is
- Waits in waiting rooms like any other guest
- Announces herself out loud on arrival
- Removable in one tap at any point in the meeting
- Never joins a meeting nobody invited her to
2
Silent is the ground state, not a setting we enforce
She cannot speak unless the product explicitly asks her to, so silence is a property of how she is built rather than a promise about her behaviour. Three presets decide how much of a participant she is, and the default is the quiet one.
- Silent — listens and writes, never speaks or types in chat
- On request — answers when someone says her name
- Co-pilot — runs your question script and yields the moment a human talks
- Chosen per meeting, with a default for the agency
- The client-facing default is silent unless you change it
3
The human decides, and the evidence is attached
AI output is a draft until a person confirms it. Scores carry the quote they came from, so a decision can always be traced back to something a candidate actually said rather than to a model's opinion.
- AI scores are drafts, visibly, until a human saves them
- Every score carries its evidence from the transcript
- Changing an AI score is recorded as a human override
- Recording off by default — transcript only
- Your retention rules, your tenant, your data
- Nothing sent to a candidate or client without you
The controls
Set for the agency, overridable per meeting.
The rest of the meeting
Consent & control is one moment of it. They only work like this because they are all the same product as your CRM.
Questions recruiters ask
Do we need to tell candidates?
She tells them herself — she is in the participant list under your agency's name and says one line when she arrives to say she is taking notes. Most agencies add a sentence to their interview confirmation email as well, and we would encourage that: the tag-team design only works when everyone in the room knows exactly what is happening, and candidates who understand it are noticeably more relaxed than candidates who spot a bot.
Can she be in a meeting without speaking at all?
That is the default, and it is enforced by construction rather than by instruction — she makes no sound unless the product deliberately asks her to speak. Silent means silent: no audio, no chat messages, no interruptions.
What about AI in hiring rules?
The design is the answer: a human runs the meeting and makes the decision, the AI asks and records, every score is a visible draft until a person confirms it, and every score carries the transcript quote behind it. That is the posture regulators in New York, Illinois and the EU are asking for, and it is how the product was built rather than something bolted on afterwards.
Who can see the recordings and transcripts?
Your people, under your existing access rules, on your records. The meeting belongs to the tenant whose user invited her, and nothing crosses between tenants — even when the meeting had attendees from three different companies in it.
An AI in an interview should be the most visible thing in the room.
