There is a particular kind of shame reserved for hearing yourself back. You play a voice memo, a recorded meeting, a message you thought was fine, and there it is: um. Then uh. Then um again, wedged into a sentence you remember delivering with confidence. And something small collapses. You sound uncertain. You sound like you don't know what you're talking about. You sound, most damningly, like someone who has not thought this through.
Here is the uncomfortable inversion: the um is evidence that you were thinking. It appeared precisely because your mind was doing the hardest part of the work. The people who never say it are not more thoughtful than you. Often they are saying something they've said before.
What is actually happening in the gap
Speaking is not one act. Psycholinguists model it as a chain: you settle on what you mean, you retrieve words to carry that meaning, you assemble a grammatical frame to hold them, and only then do your lungs and tongue get involved. Willem Levelt's account of speech production made this staging explicit — conceptualizing, formulating, articulating — and the crucial detail is that the stages run at different speeds. Articulation is fast and mechanical. Formulation is not. Retrieving the exact word for a thing you have only half-decided to say is slow, effortful, and prone to stalling.
So the mouth catches up to the mind and finds nothing waiting there yet. That gap has to be filled or abandoned. Um fills it.
What Herbert Clark and Jean Fox Tree argued, in a paper that quietly upended how researchers think about this, is that uh and um are not accidents leaking out of a broken system. They function as words. They are announcements: a delay is coming. And the two are not interchangeable — Clark and Fox Tree found uh tends to precede a short delay, um a longer one. You are, without deciding to, telling your listener how long to wait.
More striking still is what happens on the other side of the conversation. Fox Tree found listeners recognize the word after an uh faster, not slower. Work by Martin Corley and colleagues, using EEG, found that words following an um produced a reduced N400 — the brain signature of unexpectedness — as if the filler had told the brain brace, something unusual is coming, and the brain had listened. Jennifer Arnold and colleagues showed that when a speaker hesitates, listeners shift their expectation toward the new thing in the scene rather than the familiar one, because we intuitively know that hesitation means the speaker is reaching for something harder to name.
And Scott Fraundorf and Duane Watson found that when a speaker told a story with natural disfluencies, listeners remembered the plot points better than when the story was delivered smoothly.
Read that again. The thing you edit out of yourself is a thing that helps people remember what you said.
Fillers scale with difficulty, not with incompetence
The rate at which people produce um and uh isn't a fixed personality trait. It rises when the topic is unfamiliar, when the sentence ahead is long and syntactically complicated, when you are choosing among several ways to say something, when the stakes are high. Researchers counting fillers in natural speech find them at a rate of a few per hundred words — but that rate climbs the moment the speaker is doing something cognitively expensive.
Which means: your filler rate is closest to zero when you are on autopilot. The salesperson delivering a memorized pitch, the professor teaching the same lecture for the eleventh year, the person telling the anecdote they've told at forty dinner parties. Smooth as glass. And not, in that moment, generating a single new thought.
The fluency you envy is often the fluency of retrieval, not the fluency of thought. You are comparing your first draft to someone else's rerun.
Why trying to suppress them makes everything worse
Here's the trap. Once you decide to eliminate your fillers, you install a second process in your head: a monitor, running alongside the speaking, scanning for the forbidden sound and killing it before it escapes.
But speech production already uses a self-monitoring loop — it's how you catch yourself mid-word and repair. That loop draws on the same limited attention that formulation needs. Add a vigilant filler-suppressor on top and you are spending working memory on policing your own mouth, taken directly from the budget you needed to find the right word.
The result is familiar to anyone who's tried it. You don't say um. Instead you produce a long dead silence, or you say a worse, safer, blander word because it was the one available under load, or you lose the thread of the sentence entirely and restart. You traded a 200-millisecond signal that your listener would have processed without noticing for a genuinely damaged sentence.
This is why the advice "just pause instead of saying um" only works after the underlying fluency exists. A pause replaces a filler cleanly when you already know what comes next. When you don't, the pause is not a pause. It's a hole.
The real cause, and the real fix
If fillers are the sound of formulation running behind, there are only two honest ways to reduce them. Neither is suppression.
Reduce the formulation load. Know the material better. Have the structure in your head before you open your mouth. Rehearse the shape of the argument, not the words — because rehearsed words make you brittle, but a rehearsed structure means every sentence has somewhere to go. Speakers who seem effortless usually have three or four landing points memorized and improvise between them.
Or accept the disfluency at production time and remove it afterward. This is what every polished thing you've ever consumed actually did. The interview you read had its ums stripped by a transcriptionist. The podcast had them cut. The author who sounds so assured on the page wrote a fourth draft. You are the only person judging your first take against other people's final ones.
That second path is enormously underrated, because for most of history it was expensive. Cleaning up spoken language required someone to sit down and do it by hand. So we internalized the belief that speech must be clean at the moment of speaking — a standard we never once applied to writing.
Your next moves
- Record yourself talking for two minutes about something you understand deeply, then two minutes about something you're still figuring out. Count the fillers in each. The gap between the two numbers is not your competence — it's your cognitive load. Seeing it in your own voice is more convincing than any argument.
- Before your next meeting or call, write down three landing points, not a script. Structure is what actually reduces disfluency; memorized sentences collapse the moment someone interrupts.
- Practice replacing the filler with a full breath, but only on material you know cold. Try it on unfamiliar material and you will feel exactly why the advice backfires. That felt difference is the lesson.
- Stop apologizing for hesitation in conversation. When you catch yourself saying "sorry, I'm rambling," you are drawing attention to something your listener's brain had already used as a helpful cue and forgotten.
- Separate capture from polish, permanently. Say the messy version. Fix it after. Never do both at once — that's the whole mechanism.
Where this leaves the blank page
That last move is the one that changes the most, and it's the one hardest to do alone. Speaking a first draft only works if the cleanup is close to free.
Quill is built on exactly that split. You talk — hesitations, restarts, the um you were never supposed to be ashamed of — and it lands as clean text in whatever app you're already in, processed on your device, never sent anywhere. Then one tap rewrites it into the register you actually needed: tighter, warmer, more formal. Your mind gets to do the hard part, out loud, unpoliced. The polishing happens where it always belonged — afterward.
If you've been holding your thoughts back because they don't arrive fluent, that was never the real problem. Try Quill and let the first draft sound like a first draft.