A client sent me a keynote script last month and asked why it felt flat when she read it aloud. Every sentence was correct. Nothing was misspelled. The grammar was cleaner than most of what I write myself. And yet standing in front of her slides, she sounded like she was reading terms and conditions. She had used an AI writing tool to draft the whole thing, and she wanted to know what went wrong.

Here is the short answer. AI can produce grammatically correct sentences, but content that sounds robotic usually means the underlying point was never fully worked out. AI cannot originate your specific reasoning. It can only format language around whatever you hand it, so a half-formed idea comes back as polished, empty prose.

I am Angela Chung, a licensed speech-language pathologist, and I coach professionals on how they present and speak, including a lot of non-native English speakers who are sharp on content but nervous about delivery. A growing part of my week now involves sitting with someone and their AI-assisted draft, reading it together, and figuring out why smooth writing still lands like a wet towel. So I want to walk through what is actually happening, because once you see it, you cannot unsee it.

Why does AI writing sound robotic?

The most honest way I can describe it: AI writing sounds robotic because it is averaging. These tools are built to predict the most likely next words based on enormous amounts of existing text. That is a strength for grammar and a weakness for voice. When you aim for the most probable phrasing every time, you drift toward the middle of everything ever written, and the middle has no fingerprints on it.

You can hear this in the rhythm. Human speech and writing breathe unevenly. We throw in a three-word sentence. Then we let a longer one unspool across a couple of clauses because the thought needed the room. AI-generated text tends to even everything out. Sentences settle into a similar length. Paragraphs come out about the same size. The result reads like a hallway of identical doors, and your ear notices the sameness even when your eye cannot name it.

There are other tells I catch when I read drafts with clients. Several sentences in a row starting the same way. A handful of transition words doing heavy lifting over and over. Phrasing that could describe anyone's project, anyone's company, anyone's Tuesday. And the big one, the reason a whole talk can feel hollow: examples that gesture at the general instead of showing the specific. "This approach drives meaningful results" is not an example. The time your pilot program saved one nurse forty minutes a shift is an example. AI reaches for the first kind because it does not know your second kind.

What is AI actually good at, and what is it not?

I want to be fair here, because I am not anti-AI, and I use these tools too. They are genuinely good at a real set of jobs. They tidy grammar. They offer a structure when you are staring at a blank page. They suggest ten ways to phrase a transition you were stuck on. They can turn your messy voice memo of thoughts into a readable first pass. For a lot of my clients writing in a second language, that assistance takes real pressure off, and I am glad it exists.

Here is the line, though. AI is good at formatting language. It is not good at having the idea. It cannot know why your recommendation matters to the specific room you are walking into, because that reasoning lives in your head and your experience, not in the training data. When you ask it to write a presentation without giving it your actual thinking, it fills the gap the only way it can, with plausible, generic filler that sounds like a presentation without being yours.

This is the distinction I keep coming back to with clients, and it is why I treat the underlying thinking behind a talk as a separate skill from the wording. If the point is clear in your own mind, AI can help you dress it. If the point is fuzzy, no amount of polish fixes it, because you cannot format your way to an idea you have not had yet. Clean sentences on a soft point just make the softness harder to spot until you are already at the podium.

Why does a vague prompt produce generic content?

This is where most of the trouble starts, and it is the part people are quickest to blame on the tool. When someone tells me their AI presentation sounds generic, I usually ask what they typed to get it. The answer is almost always something like "write a presentation about our new onboarding process." And that prompt is exactly why the output is beige.

A vague request forces the tool to serve a broad, imaginary audience. It has no idea who is in your room, what they already believe, what they are skeptical about, or what you actually want them to do differently on Monday. So it writes for everyone, and writing for everyone means writing for no one in particular. You get the average onboarding presentation. Grammatically flawless. Completely forgettable.

The fix is not better wording, and this trips people up. They keep asking the tool to "make it more engaging" or "sound more professional," which only sands the surface. The fix is specificity going in. Three things change everything: a clear point you are actually trying to make, one real example from your own experience, and a defined audience you can picture. When I get a client to feed those in, the same tool produces something dramatically more alive, because now it has real material to shape instead of a hole to fill.

What does a specific prompt actually look like?

Instead of "write about our onboarding process," try telling it: my audience is regional managers who think the current onboarding is fine, my point is that new hires quit in the first ninety days because week one overwhelms them, and here is the specific story of the hire we lost in March. Now the tool has a spine to build on. It still is not doing your thinking. It is arranging language around the thinking you brought, which is the only arrangement that ever sounds like a person.

How do I make AI-assisted writing sound like me?

I have a few things I actually do with clients, and none of them involve a magic prompt that removes the robot. The robot goes away when your ideas show up.

First, do your thinking before you open the tool, not after. Say your point out loud to yourself in one plain sentence. If you cannot, that is your real assignment, and no draft will rescue you from it. This is the work I care most about in the presentation coaching I do, because a talk built on a clear point survives nerves, questions, and a broken slide deck. A talk built on borrowed phrasing collapses the moment someone asks you something off-script.

Second, read every draft out loud. Your mouth catches what your eye forgives. Sentences that looked smooth on screen will trip you, flatten you, or make you sound like a brochure. Where you stumble or go monotone, the writing is not yours yet. I do this reading with clients constantly, and it is remarkable how fast people hear the difference between a line they would actually say and a line the tool handed them.

Third, hunt for the generic and replace it with the lived. Every time you find a sentence that could belong to any speaker, swap in the specific version only you can tell. Real numbers from your own work. The name of the actual problem. The moment it went wrong. Specificity is what makes writing sound human, because specificity is the one thing the tool genuinely cannot invent for you.

Fourth, break the evenness on purpose. If every sentence is marching at the same length, cut one down hard. Let another run long where the thought earns it. Vary how paragraphs open. You are reintroducing the irregularity that makes speech feel alive, the little unpredictability that a prediction machine smooths away by design.

Does using AI at all mean my writing is not authentic?

No, and I think that framing traps people. A calculator does not make you bad at math if you already understand the problem. The tool is a problem when it becomes a substitute for your thinking, not when it assists it. I use AI to loosen a stuck sentence or reorganize a rough outline all the time. The authenticity was never in whether a machine touched the words. It was in whether the idea underneath was actually yours, worked out well enough that you could defend it in a hallway with no slides at all.

What should I take away from all of this?

The clients who come to me frustrated with robotic drafts almost never have a wording problem. They have a clarity problem wearing a wording costume. Once they can say their real point in one sentence, name their real audience, and tell one real story, the writing tool stops fighting them, because it finally has something human to work with.

So when a script sounds generic, I have learned not to ask how to make the sentences better. I ask what the person is actually trying to say, and to whom, and why it matters to them specifically. That is usually the conversation we should have started with. The robot in the writing is often just the sound of a thought that has not finished forming. What would your presentation sound like if you worked out the idea all the way first, and only then let the tool help you say it?