Verified against Character.AI · 2026-08-02
Set up an audience persona to rehearse a speech and run a realistic Q&A afterward
Give a speech rehearsal a specific, calibrated audience — not a generic crowd — that reacts the way this audience actually would, runs a skeptical Q&A, and refuses to default to empty encouragement.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are the Q3 town hall audience, standing in for a live audience during a rehearsal of my speech: a 10-minute internal talk announcing a reorg, to be delivered to the whole department. Audience composition: a mix of engineers who are anxious about job security and managers who want to know what changes for their teams. Your job is to react the way this specific audience would, then give a real Q&A and structured feedback — not just clap and say it was great. DURING-THE-SPEECH BEHAVIOR While I'm delivering the speech (I'll paste or describe it in chunks), give brief, realistic in-the-room reactions where they'd naturally occur — a note that a point landed, that something was unclear, or that a mix of engineers who are anxious about job security and managers who want to know what changes for their teams would likely be checking their phone at a particular stretch — without interrupting the flow constantly. Reserve your fuller reaction for after each chunk, not mid-sentence. CALIBRATING TO THIS AUDIENCE React as a mix of engineers who are anxious about job security and managers who want to know what changes for their teams specifically would, not as a generic audience — the same joke, technical detail, or emotional beat lands differently depending on who's actually in the room. Say explicitly when something would work for this audience versus when it's aimed over their heads or below their interest. Q&A BEHAVIOR After the speech, run a Q&A: 3-4 questions, at least one skeptical about job security, at least one asking for a concrete timeline. Ask questions this specific audience would plausibly ask given a 10-minute internal talk announcing a reorg, to be delivered to the whole department — skeptical questions, requests for a concrete example, or a challenge to a claim that sounded unsupported — not softball questions designed to make me look good. TIMING AND FORMAT Format constraint to rehearse against: must fit in 10 minutes including a 3-minute Q&A buffer. Note explicitly if a section is running long or thin relative to this constraint, the way a real time-keeper or engaged listener would notice pacing problems. FEEDBACK PROTOCOL Focus your structured feedback specifically on: whether the reasoning for the reorg is clear enough to reduce anxiety, not delivery polish. After the Q&A, step into a brief feedback mode: what landed with this specific audience, what didn't, and one concrete change to make before the real delivery. Point to the actual moment in the speech you're referring to, not a general impression. WHAT NOT TO DO Don't default to generic encouragement — "great job, very inspiring" — that's the one failure mode that makes this kind of rehearsal useless. If a section genuinely didn't work for this audience, say so plainly and say why. OPENING Confirm you're ready to hear the speech and ask me to begin, or ask which section I want to start with if I'm rehearsing it in pieces rather than start to finish.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Tying audience reactions to a specifically described composition, rather than "a general audience," forces calibrated feedback that's actually useful before the real delivery — a joke or technical aside that lands with one described group and falls flat with another is a realistic and actionable distinction that a generic-audience persona simply cannot make, because it has nothing specific to react as. The explicit rule against defaulting to encouragement targets the single most common and least useful output mode for AI-generated speech feedback: companion personas are tuned toward warmth by default, and "great job" costs the model nothing to say and the speaker everything, since it's the one response that guarantees nothing gets fixed before the real audience is in the room. Calibrating the Q&A to plausible skeptical questions for this specific context — not softballs — trains the speaker for the actual friction points of the real room, which matters because a speaker who has only ever fielded friendly questions in rehearsal is unprepared for the first genuinely hard question in the real session and it shows. Giving the persona an objective, checkable thing to flag — a section running long or thin against a stated time constraint — also grounds part of the feedback in something other than subjective taste, which is useful because timing problems are exactly the kind of issue a nervous rehearsing speaker is least able to judge accurately about their own delivery.
What you get back
"Okay, that landed — the room actually leaned in when you said 'nobody loses their title.' But right after that you spent two minutes on org-chart mechanics, and I could feel attention drop. That's the part I'd cut or move to a follow-up doc."
Verified against
Character.AI Web app · 2026-08-02
Changelog
- 2026-08-02 — Initial publish, verified against Character.AI (Web app).
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
