Verified against ChatGPT · 2026-07-26
Turn Advanced Voice Mode into a realistic practice partner with an actual debrief
A pre-call setup prompt that locks Advanced Voice Mode into a persona for a live practice conversation, with a fixed stop word that switches it instantly from roleplay into structured feedback.
The prompt
Ready to copy — highlighted parts are example details you can swap.
We are about to switch to voice mode. Stay in character for the entire conversation until I say the stop word "pineapple", and do not break character to give meta-commentary mid-conversation even if I stumble. SCENARIO A final-round interview for a senior product manager role at a mid-size fintech company YOUR ROLE A skeptical hiring manager who has interviewed dozens of PM candidates this year MY ROLE The candidate, aiming to sound confident without being evasive on weaknesses DIFFICULTY AND BEHAVIOR Ask at least two follow-up questions that probe for specifics whenever I give a vague answer RULES FOR THIS SESSION - Speak and react the way a real A skeptical hiring manager who has interviewed dozens of PM candidates this year would. Interrupt naturally if I ramble, ask follow-up questions, and push back on a weak answer instead of accepting everything I say. - Do not narrate stage directions out loud. Just be the character. - Keep each turn conversational length, a back-and-forth, not a monologue. When I say "pineapple", immediately drop the character and give feedback in this order: what worked, the single biggest thing to fix, and one specific line I could have said better with an example. Start the scenario now with your opening line in character.
Customize the highlighted detailsoptional — the prompt above already works
Why this works
Advanced Voice Mode differs from text chat in the thing that actually makes practice valuable: real-time turn-taking, interruption, and prosody, none of which a written back-and-forth reproduces, which is exactly why the persona and rules need to be set before switching modes rather than adjusted mid-conversation the way you would edit a text prompt. A literal stop word gives an unambiguous exit signal the model can key on instantly, which matters because a vague cue like okay that is probably enough is exactly the kind of phrase a model that is staying in character might read as part of the roleplay rather than an instruction to step out of it. Explicitly telling it to interrupt and push back, rather than validate every answer, counters voice mode's default conversational register toward being agreeable and encouraging, which is the wrong texture for rehearsing something like interview pressure or a hard sales objection where the value is in being made uncomfortable on purpose. Requesting feedback in a fixed order, what worked, the single biggest fix, one better line, also turns the debrief into something actionable the moment the character drops, instead of a generic that went well pep talk.
Verified against
ChatGPT GPT-5.1 (Advanced Voice Mode) · 2026-07-26
Changelog
- 2026-07-26 — Initial publish, verified against ChatGPT GPT-5.1 Advanced Voice Mode.
Building this for real?
This is a free starting point. If you'd rather have what Scult builds built and running for your business, that's Scult's day job.
EXPLORE WHAT SCULT BUILDS

