When I first started at Toastmasters, I recorded my speeches and cringed at the result. A phrase I thought sounded "professional" came out as something robotic. I had been editing myself into someone else. The fear is real: the more you let AI improve your words, the more you risk losing the human voice that makes your speech memorable. The answer is not to avoid AI. It is to use AI as a coach, not as an author.
Why AI can flatten your voice
Large language models are trained to produce fluent, conventional, "safe" prose. That is exactly what makes a draft sound polished but generic. When a model rewrites your speech, it smooths your quirky phrasing, replaces your precise personal examples with plausible-but-impersonal ones, and shifts your natural cadence toward a written rather than spoken rhythm.
The problem is not AI. It is treating the model as the final author. A speech lives or dies by its voice: the way you actually think, the phrases you genuinely use, the stories only you can tell. If the draft loses those, the audience stops hearing a person and starts hearing a press release.
The four-step workflow
I use a simple rule: AI does the heavy lifting on structure and precision. I keep ownership of voice and content. Here is the workflow I have refined through dozens of speeches.
Step 1: Capture the raw truth first
Before any AI intervention, record or draft the speech in your own words. Do not try to make it perfect. Note three things that matter most:
- What is the one thing you want the audience to remember?
- What personal story or example proves it?
- What is your natural way of saying things when you are speaking, not writing?
It does not matter if the raw draft is messy. It matters that you captured your authentic material before a machine rewrites it. This is the part a model cannot produce for you: your actual experience, in your actual voice.
Step 2: Ask for a structural critique, not a rewrite
Bring the raw draft to AI, but tell it what to do. A good prompt is:
I am polishing a Toastmasters speech. Please give me a short critique of the structure: where is the opening weak, where does the story lose focus, where is the call to action unclear? Suggest only where structure or clarity needs work. Do not rewrite the language.
The key word is "suggest only." You want the model to point at problems, not substitute its own writing for yours. This keeps the direction in your hands.
Step 3: Ask for targeted edits, with guardrails
When you do want to improve specific lines, be explicit about what to change and what to leave alone. A strong prompt:
Please tighten these three paragraphs for spoken delivery. Keep my original meaning, keep my personal examples, and keep my casual tone. Do not make it sound formal. Return only the edited paragraphs, nothing else.
Breaking it into small chunks is important. Ask the model to edit one two-minute section at a time. That gives you many small decisions instead of one giant rewrite, and it stays easier to spot where voice is slipping.
Step 4: Audit the voice after every edit
This is the most important step, and the one most people skip. After AI touches a section, read it out loud. Ask:
- Does this sound like something I would actually say?
- Do I still sound like me, just clearer, not different?
- Did I lose the personal details that made the story mine?
If the answer to any of these is no, revise it back. The AI is a microphone, not the speaker. It can make your voice louder and clearer; it cannot invent a voice for you.

