AETHER / AI-MATCHED CAPITAL INTRODUCTIONSSAIL FORCE MARKETING / B2B REVENUE & OUTBOUNDIRIS VISION CAPITAL / DEAL FLOW & INVESTMENT FACILITATIONTEDxAshesiUniversity / TEDxLeidenUniversityAETHER / AI-MATCHED CAPITAL INTRODUCTIONSSAIL FORCE MARKETING / B2B REVENUE & OUTBOUNDIRIS VISION CAPITAL / DEAL FLOW & INVESTMENT FACILITATIONTEDxAshesiUniversity / TEDxLeidenUniversity
All insights
Public Speaking · 2026-10-07 · 7 min read

How to Write a Keynote Speech: The 18-Minute Structure I Use

Most talks fail on the page, not on the stage. The spine, time budget, opening lines and rehearsal routine I use to write a talk that lands one idea.

Most talks fail on the page, not on the stage. The speaker gathers good material, arranges it in the order they happened to think of it, then tries to rescue the structure with delivery. Delivery cannot rescue a talk that has no spine, and an audience forgets a talk they cannot summarise.

I have given two TEDx talks, at TEDxAshesiUniversity in 2021 and TEDxLeidenUniversity in 2022. What follows is the process I use now, from blank page to a talk that fits the clock and leaves one idea standing.

Start with the sentence the room should repeat

Before you write a word of the talk, write the one sentence you want a listener to say to a colleague the next morning. Not your topic. Not your title. The claim.

"Leadership in uncertain markets" is a topic. "Your team does not need more certainty from you, it needs a shorter feedback loop" is a claim. A topic gives you permission to say anything, which is why topic-led talks wander. A claim tells you what belongs.

Three tests for a throughline

  • Can it be contradicted? If no reasonable person would argue the opposite, you have written a platitude, not a claim. "Culture matters" fails. "You cannot fix culture with values, only with what you promote for" passes.
  • Does it fit in one breath? Say it out loud. If you run out of air, it is two ideas wearing one coat. Split them and pick the better one.
  • Does it change behaviour? A listener should be able to do something different on Monday. If the only response available is agreement, the talk is entertainment.

Write that sentence at the top of the document and leave it there. Every time you are tempted to add a section, check it against the sentence. Most of what you want to add will fail the check, and cutting it is the job.

A talk is not everything you know about a subject. It is one idea, defended well enough that a stranger can carry it home.

Build the spine before you write prose

TED's own guidance for TEDx events is that talks are typically under 18 minutes, and it suggests 3 to 9 minutes as ideal for a focused idea (see the TEDx rules). Corporate keynotes usually run 20 to 45 minutes. The structure below scales to either, because the proportions matter more than the total.

Most people speak at roughly 130 to 150 words per minute on stage, slower than in conversation because of pauses. So an 18-minute talk is a script of about 2,300 to 2,700 words, and a 40-minute keynote is about 5,200 to 6,000. Knowing that number early stops you writing 9,000 words and then panicking in the final week.

Budget an 18-minute talk like this:

  1. Open, 90 seconds. A concrete moment, then the claim.
  2. Movement one, 4 minutes. Why the obvious answer fails.
  3. Movement two, 4 minutes. The better mechanism, explained so it can be used.
  4. Movement three, 4 minutes. What it looks like when it works, and what it costs.
  5. Close, 2 minutes. The single action, and the image you want left behind.

That leaves roughly two minutes of slack for laughter, applause and pauses. Speakers who plan to the exact second run over, because they never budget for the audience.

Write the first ninety seconds last but guard them hardest

The opening is where you either earn attention or spend the rest of the talk buying it back. Three openings reliably work, and they all have one thing in common: they put the listener inside a specific moment before they ask for anything.

  • The scene. One place, one time, one thing at stake. "It is 4pm in Accra and I am sitting in a car park because I have just lost the account that paid my rent." Specific enough to picture, short enough to move on from.
  • The uncomfortable question. Ask something the room has privately wondered and not said aloud. Then wait three full seconds. The silence is the point.
  • The counter-claim. State the thing everyone in the room believes, then take it apart. This works when the audience is expert and expects to be flattered.

What does not work: thanking the organisers at length, walking through your credentials, or announcing your agenda. Your introducer already handled your credibility. If you need to establish authority, do it in a subordinate clause halfway through, where it reads as evidence rather than as a request for respect.

Give the middle three movements, not seven points

The most common structural fault I see is a talk with too many sections and not enough argument in any of them. Seven points in 18 minutes gives each roughly two minutes, enough to assert something and not enough to prove it. The audience hears a list, and lists do not persuade.

Three movements, four minutes each, gives you room to do the thing that actually changes minds: take one objection seriously and answer it. In each movement, use the same internal shape.

  1. Claim, in a sentence. Say it before you explain it.
  2. Evidence, in a story or a number. A story you witnessed beats a statistic you read, because you can be cross-examined on it.
  3. Mechanism. Why does this happen? This is the part most speakers skip, and it is the part that makes a talk useful rather than merely agreeable.
  4. Turn. One line that hands the listener to the next movement.

Where you need data, use a real source and say where it came from on stage. A speaker caught rounding up loses the whole talk, not just the number. I would rather stand on one verifiable figure and three honest stories than on a slide of statistics I cannot defend in the corridor afterwards.

Close on an action, not on applause

Endings collapse in predictable ways. The summary ending repeats what the audience just heard and drains the energy. The inspirational ending borrows someone else's quotation at the exact moment you should be using your own authority. The trailing ending simply stops, and the room does not know whether to clap.

A close that works does three things in about two minutes: it names the one action, it makes that action small enough to start this week, and it restates the claim in different words from the opening.

Make the action absurdly specific. Not "invest in your people" but "this week, ask one person on your team what decision they are waiting on you for, and then answer it before Friday." Specific asks get done. General asks get admired.

Rehearse to the clock, not to the page

Rehearsal is where the talk becomes yours, and it is where most people under-invest. My routine for a talk that matters:

  • Ten full run-throughs, out loud, standing. Not reading, not muttering through it in your head. Standing and speaking at performance volume, because your body times the talk differently from your eyes.
  • A timer running on every single one. Record the time at the end of each run in a note. If the spread across your last three runs is more than 90 seconds, the talk is not memorised, it is being improvised, and it will run long on the day.
  • Two runs with a hostile listener. Ask someone to interrupt with "so what?" at any point. Wherever you cannot answer in one sentence, that section is decoration.
  • One run recorded on your phone, watched with the sound off. You will see the pacing, the fidgeting and the place where you retreat behind the lectern. You cannot hear any of that.
  • The final run at the opening and close only. Those two passages should be word-perfect. The middle can breathe.

Do not memorise the script word for word. Memorise the spine: the claim, the three movement headlines, the first line and last line of each movement, and the final action. That is about twelve sentences. Hold those and the connective tissue takes care of itself, which is also why a dropped line no longer derails you.

The failure modes worth naming

Writing for the transcript. Written prose is dense and subordinate. Spoken prose is short and concrete. Read every paragraph aloud while drafting and cut anything you would not say to one person across a table.

Slides as a second talk. If your slides make sense without you, you have written a document and hired yourself to read it. One idea per slide, a few words, and long stretches with nothing on screen at all.

Borrowing someone else's story. Second-hand stories are always thinner in the telling, and the audience can feel it. Use your own material, including the parts where you were wrong.

Speaking to be admired. Attention is borrowed, not owned, and I treat the room's time as something I will answer for. That conviction is practical as much as it is spiritual: it pushes me to cut the clever line that serves me and keep the plain one that serves them.

What to do this week

Open a blank document and write your one sentence. Test it against the three questions above, rewrite it until it survives, then build only the five-block spine underneath it. No prose yet, no slides. If the spine holds without you talking over it, the talk will hold too.

Once the talk is written, the next question is usually commercial rather than creative, and it deserves separate treatment: I have written about how speakers actually get paid, and there is more in the public speaking archive.

Ready for the next step?

Book a free 30-min call with Steven