Use AI for Speaker Notes, Rehearsals and Q&A Prep
People spend their entire AI budget on slides and none on delivery, which is backwards: a room forgives plain slides, but not a fumbling answer to the obvious question. The deck wins attention; the Q&A wins decisions. And AI is arguably better at the delivery side — notes, rehearsal, Q&A prep — than at the slides themselves, because its unique capability here is simulating the audience you don't have yet: asking the hostile question at zero cost, repeatedly, the night before.
This article covers all three jobs: notes that sound like you, rehearsal against a machine audience that grades you, and the Q&A bank that turns the scariest ten minutes of any presentation into the prepared one. None of it takes more than ten minutes per job, and the whole set runs the evening before the meeting.
Notes that sound like you
Here is my deck's content, slide by slide: [paste outline or slide text]
Write speaker notes for each slide: 60-90 seconds of talking points,
first person, conversational, each ending with a one-line transition
to the next slide. Match this writing sample to my voice: [paste a
paragraph you wrote — an email or doc is fine].
The writing sample matters more than it sounds. Without it, you get a TED-talk narrator — "So, why does this matter?" — and nobody talks like that before their second coffee. With it, you get phrasing and rhythm you can actually say out loud without cringing, which is the real bar for notes: not "well-written", but "sayable". The transition line at the end of each note is a small ask with a big payoff; most presenters lose the thread between slides, and a scripted bridge costs one sentence.
Decide the depth by stakes, not by habit. A full script is right for webinars, high-stakes rooms, and any situation where you'll be nervous enough to lose structure — the script is scaffolding, and there's no shame in scaffolding. Bullet cues are right for everything else, because a script read aloud sounds like reading, and rooms forgive plain slides faster than flat delivery. The workflow that gets you both: generate the full script first, then compress — "Compress each note to a maximum of 5 cue lines" — and rehearse from the cues with the script as the backup in your bag, physical or digital.
Two-level notes for mixed rooms
When the audience mixes levels — executives and engineers in one room, or new hires and veterans — notes can carry the depth the slides dropped. Ask for a second layer: "For each slide, add an 'if asked, go deeper' line: the one technical detail, number, or example I'd bring up if someone pushes." Now the slides stay clean for the general audience, and you have the depth loaded for the specialist who tests you. This replaces the worst improvisation in presentations: the deep-dive answer you construct live, out loud, while the room watches you build it.
Rehearse against a machine audience
Paste the deck content and notes, then: "You are a skeptical [CFO / technical lead / procurement officer]. Ask me the five hardest questions about this deck, one at a time." Answer out loud — actually out loud; thinking the answer is not rehearsing it — then ask for the grade: "Score each answer: what was weak, what dodged the question, and what would a real [CFO] have asked next?" Then have it re-ask the weak ones. Two rounds of this beat rehearsing in your head, which always goes well and therefore proves nothing.
Structure the whole rehearsal as three rounds. Round 1, the hard questions: the hostile-persona exercise above; your goal is to discover which slide's argument you can't yet defend, while the cost of discovery is zero. Round 2, pacing and attention: "Where in this talk does the audience's attention most likely die? What would you cut or condense?" — AI is blunt in ways colleagues are not; colleagues will say "it was great", the machine will tell you slide 7 is a swamp, and it's usually right. Round 3, the Q&A drill: run the bank below at speed, answering in under thirty seconds each, because brevity under pressure is a skill, not a trait.
One more prompt worth its slot: "Grade my speaker notes: where do I sound rehearsed, where do I sound unsure, and which slide's notes don't match what the slide claims?" The last one catches the drift that happens when slides get edited after the notes were written — which is every deck, eventually.
Build the Q&A bank before the meeting
Based on this deck, list the 10 questions the audience is most likely
to ask, including 3 hostile ones. For each: the question, a
2-sentence answer, and one line for what to say if I don't know
("I'll follow up by ...").
The hostile-three matter most, because hostile questions aren't about information — they're about whether you flinch. An answer you've already said once comes out steady; the same answer constructed live comes out defensive. For a pitch, change one word and widen the net: "questions and objections from investors: pricing, moat, competitor X, switching costs" — objections are just questions wearing armor, and the bank disarms both the same way. For internal meetings, ask for the political variant too: budget impact, headcount implications, "what happens to my team" — the questions people think but don't ask are the ones that decide the meeting after it ends.
Use the bank live, not just in prep. Bring it printed; glancing at notes during Q&A is normal and reads as prepared, not weak. And learn the three honest forms of "I don't know", because the bank will not cover everything: "I don't have that number — I'll have it to you by tomorrow" (specific, dated), "That's outside what we tested, here's what we do know" (bounded honesty), "That's a real gap, and it's on the risk slide because of that" (owned). All three beat improvising a number, every time, in every room.
Timing without a stopwatch
Speaking pace runs around 130 words per minute for presentation delivery — slower than conversation, faster than you'd guess from text. The math is mechanical: a 12-minute slot needs roughly 1,500 words of notes; a 15-minute slot with slides at 60 words each needs about 900 spoken words plus transitions. Rather than count, ask: "My slot is 12 minutes. Trim my notes to about 1,500 words and mark two slides I can skip if I run long."
The skip slides are the part people forget, and they're the difference between ending on time and ending mid-sentence. Choose them by rule: cut decorations first — background, history, the second example — and never the ask. A talk that runs over and skips its conclusion is a talk that didn't happen; the room got the story and not the decision, which means the meeting will be repeated, expensively, by email.
Webinar and virtual specifics
Cameras change three things. Energy drains faster, so a script keeps pacing alive where bullet cues would let you drift — this is the one format where the full script is the default, not the exception. Silence feels longer on a call, so seed the Q&A: pose the first question yourself from the bank ("A question I often get is…"), because an unseeded webinar Q&A is forty seconds of dead air that nobody volunteers to break. And slides carry more load, since the deck is the only visual channel — bigger type, one idea per slide, and nothing that requires you to narrate what the audience can't see clearly at 720p.
A worked Q&A bank
For the sales-deck example from the workflow article — cutting two lead channels — the bank came back roughly like this, abridged to the five that mattered:
- "Referrals can't scale — you're cutting volume with no replacement source." Answer: volume drops about 30% for one quarter; the referral incentive doubles in the same month, and the last time we raised it, referrals grew 40% in a quarter. Hostile, and the honest answer includes the risk — it might not repeat, which is exactly why the plan has a checkpoint at week six.
- "What happens to the two people who manage those channels?" Answer: they take over the referral program and the CRM cleanup; nobody's headcount is cut, and that's slide seven.
- "Why is cost per deal triple on purchased lists if the target says 1.5×?" Answer: the target's denominator excludes onboarding cost; slide two shows both calculations — and yes, this should have been flagged in Q2. Hostile, and it deserves the "you're right" the room is already thinking.
- "Can we just lower the list quality bar instead?" Answer: we tested that in August — cheaper leads converted worse, net cost identical; the data is in the appendix.
- "What does the pipeline look like in November?" Answer: 22% lighter; marketing front-loads two campaigns to cover the gap, dates on the transition slide.
Build time: eight minutes with the bank prompt, four more checking the numbers against the CRM. The meeting's Q&A ran eleven minutes and produced zero surprises — which is the measurable outcome of the whole exercise: not "better answers", but no questions you're hearing for the first time.
Notes during the meeting itself
Notes have a second job once the talk starts: capturing what the room actually asked. Keep the Q&A bank printed and write the real questions next to the predicted ones — the misses are worth more than the hits, because they're tomorrow's deck "must include" lines. Two questions in a row on the same gap means that gap moves to slide two next time, not to the appendix; a predicted question nobody asked means the slide that answered it gets demoted into the notes. This is the retro habit from the workflow article applied to delivery, and it's why the second talk on the same material is always stronger than the first: the room wrote its own brief, and you took dictation.
Handling interruptions mid-talk
Live rooms interrupt, and a script without interruption handling breaks at the first raised hand. Three lines to have ready. The parking lot: "Good question — it's on slide seven; hold me to it if I don't get there," which keeps the talk's spine intact. The legit detour: "That changes the frame — let me answer it now and we'll skip the transition slide," which is what the skip slides exist for. The time-box: "That deserves its own conversation — can we take it offline with whoever owns it?" Only the third needs care: it must name an owner and a time, or the room hears evasion instead of discipline. Rehearse all three once; they're the difference between an interruption being absorbed and an interruption derailing the last ten minutes.
Rehearsing in the real slot
The machine rehearsal checks content; the room rehearsal checks physics. Stand up for at least one full pass — sitting rehearsal hides the energy drop that standing reveals, and pacing feels different on your feet. Record one pass on your phone and play it back: the recording catches the filler sounds, the slide you rush, and the ten-second silence after the ask that feels eternal from inside and reads as confidence from outside. Then rehearse against the real clock once — the actual twelve minutes, not "about the same length" — because the trim you estimated at 1,500 words is a guess until a real pass proves it fits the slot. If the honest pass runs fourteen minutes, cut the skip slides now, in private, instead of live in front of the room.
Notes for demo talks
When part of the talk is live product, the notes need a layer the slides don't show: what happens when the demo fails. Script three things per demo segment — the one-sentence setup ("watch the top right; this is the queue rebuilding itself"), the failure line ("the demo gods are real — the screenshot looks exactly like this"), and the return path (which slide you fall back to). Decide the fallback rule before you're on stage: one retry, then screenshots, without apology theater. Rooms respect a failed demo handled in five seconds more than a polished demo that never breaks, because the five-second recovery proves you've done this before — which, for the audience, is the actual product being sold in a demo talk.
Notes and seniority: what changes with experience
Junior presenters should over-prepare: full cues for every slide, the complete Q&A bank, the timing math done properly — because the failure mode at that stage is losing structure under nerves, and structure on paper is the cheapest cure. Senior presenters should under-prepare: cue lines only, because the failure mode at that stage is sounding rehearsed, and rooms read authenticity as competence. The intermediate stage is the dangerous one — enough experience to skip the prep, not enough to improvise well — and it's where most mid-career presentations quietly die. The honest self-test: if you can give the talk's argument from the title list alone, cues are enough; if you need the bullets to remember the order of the argument, you need the full notes. Neither level is a virtue; they're gear for different terrain.
The panic protocol
Every presenter blanks eventually, and the recovery is a protocol, not courage. Step one: stop talking and look at the slide — the slide text is your outline, and ten seconds of silence reads as emphasis while ten seconds of "um, so, where was I" reads as panic. Step two: name the next takeaway out loud — you memorized the titles as the argument during the outline pass, so the titles are the rail to grab. Step three, only if needed: skip forward — "the detail here is in the appendix; the point that matters is slide eight," which uses the deck's own structure to disguise the restart. Design the notes so this stays rare: one slide per page, 16-point type minimum, cue lines rather than paragraphs, because panicked eyes search badly. And rehearse the protocol once on purpose — blank deliberately in rehearsal, then practice the three steps — so the real version is a rerun instead of an improvisation.
Team presentations: shared notes and handoffs
Multi-presenter decks fail at the seams, so the notes need a seam protocol. Write handoff lines into both presenters' notes: the outgoing speaker ends with a sentence that names the incoming topic ("cost is only half the story — Anna has the churn data"), and the incoming speaker's first cue restates it. Rehearse the handoffs twice, minimum: they're the moment the audience decides whether the team prepared, and they're the moment the clicker changes hands, which is its own comedy without rehearsal. Keep one shared notes file per deck, not per presenter, so the Q&A bank and the timing plan have a single home. In the meeting itself, the non-speaking presenters are the Q&A bench — they track the bank and signal, with a note or a raised pen, when a question lands on their section. Teams that rehearse seams sound like one brain; teams that don't sound like a playlist.
The day-before checklist
- Notes exist in the depth the stakes require — script or cues — and they match the current slide text.
- Rehearsal ran at least Round 1 (hard questions) and Round 2 (attention mapping); the weak answer got fixed.
- The Q&A bank is printed or on a second screen, with the three honest "I don't know" forms ready.
- Notes trimmed to the slot's word budget; two skip slides marked with a dash in the margin.
- Tech check done on the actual setup: projector or call, remote or clicker, notes visible to you and not the room.
The pattern across all of this: prepare the questions, not just the answers. The deck convinces, but the Q&A is where the decision usually happens — and it's the cheapest part of the entire presentation to prepare with AI, because the audience it simulates is exactly the audience you're worried about.