Solutions
Give your interviewers someone to practise on
Interviewing is said to be learned only by doing it. The AI plays the candidate, so it can be practised as often as you like before the real thing.
Challenge
There is no way to train an interviewer except on the job
Training an interviewer means either sitting in on someone else’s interviews or going back over your own. Both need a real one, and you cannot bring in a candidate to practise on: when it goes badly, it costs them their time and you your reputation. So people get good at it across a run of real candidates.
- You cannot bring in a candidate just to practise on
- Sitting in on someone else’s interview never gives you a turn
- How deep an interviewer gets varies from one to the next
- The interviews that went worst are the ones least written up
01
Build the candidates
Several personas, set up in advance
02
Send the practice URL
The interviewer picks who to face
03
Roleplay
The interviewer asks; the AI answers as the candidate
04
Feedback generated
Question by question, straight away
05
The real interview
The interviewer
The three shaded steps can be repeated as often as you like. Deciding what to work on, and facing a real person, stay with the interviewer.
Solution
Fix the candidate, so one run can be compared with the next
The AI candidate is built from a background and a set of motivations, and then stays fixed. Set up several and the interviewer picks who to face; face the same one twice and they can see for themselves what changed. When the run ends, feedback arrives straight away, quoting what was actually said.
- The candidate is built once and stays fixed, so runs can be repeated
- Several personas to choose from, picked at the time
- Feedback goes to the interviewer, straight away
- Nobody else’s time is spent, so it is fine to get it wrong
- Before practice, the AI can talk through the job details or how the interview runs
Roleplay result (excerpt)
Overall 58 / 100
Depth of “why”10 / 20
The reason for moving was probed, but the real one never surfaced
Talk-time balance8 / 10
The candidate had room to speak, though you spoke a little too much
Worth working on next
The reason for moving was not probed deeply enough
The candidate said “Well, part of it was that it looked difficult to do where I am now.”
What came next “I see. Could you tell me about what you do at the moment?”
A question that would have worked “Where specifically did you feel that?”
Related pages
More detail, by topic
Gaps between interviewers come not only from how much they practise but from how they read the criteria.
When interviewers read the criteria differently, the same answer earns different scores →
Use gaps between human and AI scores to find where readings split, and fix them.
Set the criteria first and narrow the spread between interviewers.
Frequently asked questions
Will the AI just give you what you want to hear?
How the candidate is built decides that. A candidate who answers everything straight away is no use as practice, and in a real interview little comes out until there is some rapport. Holding back until there is, or not being entirely sure what they want yet, can be written into the persona. How difficult to make it, while keeping it useful as practice, is something we work out with you.
How do we set up the candidates?
From a background, a set of motivations and what they are looking for. You can build them yourself or we can do it with you, so how much you hand over is something to talk through. Set up several and the interviewer chooses which to face, and their details are shown on the way in, so there is a moment to read them first. You can build a persona from a real interview record, as long as names and the like are replaced first.
Do you look at expression and tone of voice?
What comes back today is about what was said. Expression and tone are something we want to cover, and they are asked for often. Because we keep non-verbal signals out of evaluation in hiring itself (the same position the EU AI Act and New York City’s rules point to), we are working out how they should be handled in practice sessions.
Where can they practise?
A browser is all they need. It works on desktop, smartphone and tablet, with no app to install. Where speaking out loud is awkward, they can answer by typing instead.
Who should start with it?
The people who interview most, and the people about to start. Run a few of them through it to check the steps and the time it takes, then widen it from there.
Who can see the feedback?
By default, the person who practised. Beyond that, you can name who else can see it, person by person.
How much does it cost?
Pricing works differently here than for hiring interviews: it follows the number of people practising and how often they do it. Tell us roughly the scale and we will put a quote together.
Show us how you train your interviewers today
We will show you where the AI interview fits into it. Free, about 30 minutes.