Thinking with AI on a high-stakes presentation

A little context I don't usually share: I taught public speaking at Purdue, and for a short time I was a professional motivational speaker. I don't lead with that when I'm invited to present — it sets expectations I might not always want to meet. But it does mean I'm unusually sensitive to what happens when language leaves the page and enters the room.

When the stakes are high, I never let AI write the first draft of a presentation. The thinking that goes into a storyboard is the presentation. But I do use AI as a thinking partner throughout: before the storyboard to sharpen what I want to say, slide by slide after that, and against a near-final script — particularly the opening and closing, where I'm aiming for a specific emotional resonance.

Three moments in that process deserve special care.

Writing for the ear, not the page.

Language that reads well can fall apart the moment you say it out loud. AI-suggested language tends toward longer sentences, evenly weighted ideas, and no cadence. Audiences hear it and think, "This sounds scripted" — one of the fastest ways to lose credibility with senior audiences listening for ownership, not eloquence.

My filters for any AI-suggested language:

• Shorten sentences until they survive one breath. If you can't say it without pausing unnaturally, it's too written.
• Replace precision with emphasis. Spoken language tolerates approximation. What matters is what you stress.
• Remove setup clauses. "It's important to note that…" adds safety, not strength.
• Keep the version you'd say if interrupted. Whatever you'd still say mid-cutoff is usually the sentence you want.

I read everything out loud once — fast. If I trip over a line, I don't edit it. I replace it.

Transitions: where the audience decides

Most presentations don't fail because the content is weak. They fail because the audience loses confidence in the speaker's control of the narrative — and that almost always happens in transitions.

AI-written transitions are logically correct and emotionally inert: "Now let's move on to…" They signal sequence, not intent. When transitions don't tell the audience what matters now and what not to conclude yet, audiences fill the gap themselves — often prematurely and often incorrectly.

AI optimizes for clarity. Presentations under pressure require direction. So I don't ask it to write transitions. I ask it where meaning could drift: Where might someone jump to the wrong conclusion? Where does this feel like a leap? Where would a skeptic disengage?

Then I write the transition myself — plain, short, more direct than feels "complete." If a transition sounds like navigation, it won't land. If it sounds like discernment, it will.

When polish starts to work against you

Late in preparation, AI starts making a presentation worse. Everything is technically solid — every caveat included, the language smooth and defensible — and nothing feels human. What's missing is choice, risk, and presence.

Stories and analogies restore that. They signal ownership. AI can generate candidates, but it's very bad at knowing which to keep. My test: Could I tell this story without notes? Would someone repeat it inaccurately — and still get the point? If not, it doesn't belong.

At this stage the work isn't refinement. It's subtraction. I keep the phrasing I'd say naturally under pressure and let rough edges stand. AI stays in the room — for feedback, for opening and closing candidates when I know what I want the audience to feel. But naming that feeling is my job. The last decisions are about what to leave unsaid, and those are mine.

Public Service Announcement: No matter how nervous you get, never never ever read your slides.