How To Write A Speech

Most speeches are just paragraphs spoken out loud, and that is why they fail.

I have written corporate keynotes, memorial tributes, and product launch addresses. The mechanics are the same across every single one. You are not trying to be eloquent. You are trying to get a specific reaction from people who are standing there, looking at their phones, and thinking about lunch. Every word you write needs to earn its place or it gets cut. Start with the one sentence you want people to remember when they walk out the door. Not a theme. A single line. If you cannot state it in fourteen words, you do not know what you are trying to say yet, and you will waste two hours writing something nobody remembers. Determine your audience's actual state of mind before you write a single word of the body. This is where most people screw up. They write for who the audience is, not for what the audience is experiencing. A room full of engineers who just got told their project is delayed does not need inspiration. They need clarity on what changed and what they should do tomorrow morning. Writing the wrong emotional register into a speech is the fastest way to lose a crowd, and you will not notice it on the first draft because every sentence sounds fine in isolation.

Once you have the core line and the audience state mapped out, write the first draft like you are explaining the problem to one person at a coffee shop. Short sentences. Direct statements. No jargon unless the room already uses it. A speech written at conference-room speed reads at stage speed. If you have to read a sentence aloud twice to get the rhythm right during the first draft, rewrite it immediately. The ear catches problems the eye skips over. I once wrote a twelve-minute product announcement for a hardware team. Everything tracked. Then I read it aloud and realized the opening thirty seconds were entirely internal metrics. Nobody in the room cared about our Q3 throughput targets. I cut the first section down to one sentence about why the product exists and moved the specs to the second half. The revised version ran eleven minutes and got actual questions instead of polite nodding. That cut saved the talk.

Structural mechanics most people skip

A speech needs a transition pattern. Not a thesis statement. A transition pattern. Each section should end by pointing toward the next section. When you finish a point, the last sentence should create a question or a gap that the next part fills. This keeps people from zoning out between ideas. Without it, you have a list of topics, not a speech. Openings matter, but not the way speakers think they do. Do not start with a joke unless you are a trained comedian. Do not start with "Thank you for having me." Open with the problem, the shift, or the claim. Everything before that is preamble and drains attention. The opening line should land within three seconds of you starting to speak. Closings work the same way. Restate the core line. Then stop. Do not summarize the whole speech. People do not need a recitation of what you just said. Give them the line and step away. A closing that repeats three points reads like a student reviewing notes. It kills the momentum you spent the last ten minutes building.

Length, pacing, and delivery markers

Use roughly one hundred thirty to one hundred fifty words per minute as your baseline when drafting. That is not a hard rule, but it keeps you honest about scope. An eight-minute speech is about one thousand words, not fifteen hundred. When people write long, it is usually because they are unsure of their point and pad the gaps with filler sentences. Filler sounds normal on paper. It sounds exactly like filler the moment you speak it. Mark your draft with delivery cues. Pause here. Slow down. Emphasize this word. Skip this sentence if you lose the room. A script without performance notes is just an essay. The difference between a script and a speech is the intention behind each sentence. You need to know which ones breathe and which ones drive. I keep a habit of reading every draft out loud with a timer and marking breath points. The first read tells me what works. The second read tells me what is awkward. The third read tells me what to cut. Most drafts die on the third read. That is normal. The draft that survives three full read-throughs without major restructuring is the one you can actually perform under pressure.

Edge cases and where the method breaks down

Simultaneous interpretation is the hardest constraint I deal with. If your speech will be interpreted live into another language, you cannot use idioms, wordplay, or tight rhetorical devices. Everything doubles in length during translation, and idioms become nonsense. I solved this by writing a second version of the script optimized for interpreters, not for the live audience. The live version gets the color. The interpreter version gets plain syntax and explicit clause ordering. It takes twice as long to draft, but it prevents the presenter from sounding like they are reading poetry to a room that only understands literal English. Another common failure mode is the hybrid audience. You have executives in the front row and individual contributors in the back. They need different things from the same speech. I handle this by burying the executive takeaway in the first two minutes so they stay engaged, then moving into concrete detail for the rest of the room. If you lead with detail, the leaders check out. If you lead with strategy, the workers tune out. Splitting the payload solves the problem without weakening either side.

Counter-intuitive things that actually help

Write the hardest section last. Not first. Beginners front-load their best material because they are nervous about running out of momentum. That is backward. Save the section you are least sure about for after the rest of the draft is locked. The structure around it gives you a frame to test whether the hard section actually fits. If it does not fit after the draft exists, cut it. A clean speech with one fewer point beats a speech with a weak bridge paragraph in the middle. Also, shorter is usually louder. A room remembers three sharp points better than five soft ones. The cognitive load of listening is real. You are asking people to hold your structure in working memory while you deliver content. More than three main points and people start dropping one without realizing it. They will blame your delivery later, but it was simply too much to track. There is a downside to this approach that nobody mentions. It requires you to know your content well enough to make aggressive cuts. If you are still learning the material, you will struggle to identify what can go. In that case, write the full draft first, then compress it using the three-point rule. The compression pass is where the speech actually gets built. Most people skip it because editing out of pride feels worse than editing out of discipline. It feels worse. It also makes the speech ten times better.

Final notes on execution

Print your draft on paper and read it standing up. Sitting and reading changes your pace. You will catch run-on sentences and tangled clauses you missed on screen. Mark the marks you need while standing. Then rehearse from the marked paper, not from your phone. Phone text is too clean. It hides rough spots. Paper forces you to slow down and notice where your tongue trips. If you follow the core line first, write for the actual audience state, pace to roughly one hundred thirty words per minute, mark your delivery cues, and run three read-throughs, you will produce a speech that lands consistently. The drafts that fail are the ones written for performance instead of comprehension. Keep it plain. Keep it tight. The room will catch up.