Rehearse the conversation with someone who pushes back, then see exactly where you went reactive instead of responsive.
You already know roughly what you want to say. Almost everybody does. The gap is not knowledge, it is what happens to a good answer when somebody pushes back on it and you are the one who has to keep talking.
Reading a strong answer in your head does not close that gap. Saying it, badly, while someone is unimpressed at you, and then saying it again, does. That is why this is a spoken rehearsal and not a worksheet.
It will feel strange for about ninety seconds. Then it stops feeling like a exercise, which is the point at which it starts being useful.
Not a rating. A count of what actually happened, anchored to sentences you said, and one thing to change.
The suggested wording stays hidden until you have said your own version out loud. Being handed a polished line improves your next attempt and does nothing for the room you are actually walking into.
One conversation cannot tell you how good you are at interviews. It can tell you exactly what happened in that conversation, so that is what you get: counts of observable behaviour, never a grade.
The other person stays in character throughout. No praise, no tips, no breaking off to tell you that was a good answer. A rehearsal that congratulates you is not a rehearsal.
Practice without feedback changes nothing measurable. Feedback you read and then close changes very little either. The rep straight after the feedback is what carries, so the room is built to send you back in.
Your browser turns your speech into text on your own machine. The audio itself is never uploaded, never stored, and never heard by anybody.
Not a recording and not a person listening in. It answers in real time, in a synthesised voice, and it will push back.
The text of the conversation goes to Anthropic, which writes the replies. Those replies go to ElevenLabs to be spoken. Nothing is written to our database, and closing the tab is the end of it.
So a later session can be compared against an earlier one, a single moment from each debrief stays in this browser. Never a transcript, never on our servers, and you can clear it from the debrief itself.
A microphone, somewhere you can speak freely, and ten to fifteen minutes. Chrome, Edge and Safari can do this. Firefox cannot yet, because it has no speech recognition built in.
It works on a phone, and a phone is often the easier option, because the hard part is finding a room where you can say "I am resigning" out loud without explaining yourself.
Two full rehearsals are free, including the debrief and the comparison across sessions. That is the whole method rather than a sample of it.
If the same thing keeps happening across three rehearsals, or the real obstacle is not the sentence but the person you have to say it to, more practice will not shift it. That is worth an hour with someone who has done this before.
Talk it through with a person. No pitch, no pressure.
Have it once here first, where the only thing at stake is finding out what your answers sound like.