How to Stop Saying Um: 6 Fixes That Actually Work

Why do we say um at all?

An um is a stall while your brain finds the next word — and it holds the floor while you stall. The linguists behind Language Log put it simply: "uh" announces a short delay, "um" announces that a longer delay is coming, and both tell the listener to wait rather than jump in. That is the mechanism. It is not a defect in you; it is how spoken language handles gaps.

So the goal is not zero fillers. It is that the stall stops eating your sentences. Steven Cohen's advice at Harvard Extension School comes down to the same three words: pause, think, answer — because "when you use a filler word such as 'um,' you are thinking verbally."

Why can't you hear your own ums?

Because while you speak, all your attention is on the content. The um is invisible from inside the sentence; you only hear it from outside. That is why "just notice it and stop" fails, and why counting afterwards works: count the fillers in your own transcript, once a week, instead of policing yourself live.

HBR's advice for presenters is the same in spirit: record your practice, find your crutch words on the playback, and pair each one with an action so it becomes noticeable. The noticing happens in playback, not in the moment.

Try it: Read a short story aloud for sixty seconds, record it, then read the transcript and circle every um, uh and you know. Do not fix anything yet. You are just finding your baseline number.

Fix 1 — replace the um with a closed-mouth pause

The physical version: lips together, breathe through the nose, one beat, then the next word. The mouth has to be closed because an open mouth invites a sound — a silence with your mouth open becomes an um by reflex. A closed pause reads as thinking; an open pause reads as stuck.

Give the pause a count at first: say the period, breathe, then start the next sentence. It will feel theatrical for about a week. What you are rehearsing is the replacement, not the pause itself — once the swap is automatic you stop counting.

Harvard Extension's framing helps here: fillers cluster in two hot spots, the start of a statement and the gap between ideas. Put the closed-mouth pause in exactly those two places and you have covered most of the ground. How pauses work in a spoken story covers the difference between a strategic pause and a hesitant one.

Fix 2 — slow the sentence before the hard word

Most ums sit in front of a name, a number or a technical word. The stall is not random; it is queued behind a word you are not sure of. Slow down two words before it and the stall has somewhere to go — the difficulty arrives inside a deliberate slowdown instead of inside an um. "Her name was — Venetia Burney" said slowly beats "her name was, um, Venetia Burney" at the same speed, and takes no longer.

This is also where rate feeds fillers: if you are speaking fast, the only way to buy a half-second is a sound. If you have room in the sentence, you can buy it with silence. Our page on how fast you speak, and why it feeds the ums has the numbers.

Fix 3 — say the plain word instead of searching

"The thing you stir with" beats four seconds of um. Substitution is faster than recall, and the listener builds a picture faster than you can find a noun. Choose the plain word, keep the sentence moving, and let the exact term come or not. The audience almost never notices that the precise word was missing; they only notice the hole where a word was being dug for.

This is the same move radio tellers use: describe the object, don't name it, when the name is fighting you. It works in a second language too — the plain word is the word you already own.

Fix 4 — plan the first line and the last line only

The two places people stall hardest are the opening and the landing, because those are the two places with no momentum to carry you. Give each one a fixed sentence you have said out loud before. Plan nothing else word for word — recited middles sound flat, and the middle will look after itself once the entry and exit are solid.

Fix 5 — practise in 60-second takes, not whole speeches

Short takes keep the failure small enough to repeat. Take one section, take it once, listen once, fix one thing, take it again. Three takes, one fix each — you will fix more in three minutes of retelling than in twenty minutes of polishing a full run-through. The full routine is in how to practise in short takes on your own.

The reason it works is attention. In a full run-through you are managing the whole arc, so the ums slip past. In a sixty-second take there is nothing to manage but the words, so every um is audible — and fixable on the very next take, while the section is still in your mouth.

Fix 6 — count, don't judge

Rate per 100 words, once a week, same story. That is the whole practice. The arithmetic the filler word counter uses is fillers divided by words, times 100 — so 9 fillers in 62 words is 14.5 per 100. Judge the trend, not the day: this week's number against last week's, on the same passage.

Here is what that looks like on a real story. The source text is the opening of the story of the girl who named Pluto:

In March 1930, an 11-year-old girl named Venetia Burney was eating breakfast in Oxford, England. Her grandfather was reading the newspaper and told her some big news: scientists had found a new planet, far away. It did not have a name yet.

A weak retelling, with the fillers marked:

"So um in 1930 uh there was this girl um in Oxford and like her grandfather was reading the paper you know and he said um they found a planet uh and like she said Pluto um because uh Roman myths you know."

Count: 9 firm fillers (um ×4, uh ×3, you know ×2) and 3 "like", which our tool counts separately because "like" has real uses. The take is 62 words, so: 9 ÷ 62 × 100 = 14.5 firm fillers per 100 words, plus 4.8 possible from the "like"s. Check the sum against the tool.

The same retelling re-taken, with pauses marked as |:

"In 1930 | a girl in Oxford | her grandfather is reading the paper | he says they found a planet | she says Pluto | Roman myths | god of the underworld."

0 firm fillers, 5 pauses, 27 words. Shorter and cleaner. And it has lost the grandfather's reaction — the man who loved the idea and sent it to a professor, which is the part that makes the story warm. Clean is not automatically better; the next take puts the reaction back in, without the ums.

That honest note is the real lesson. A filler count is a measurement, not a score of how good the story was.

What about "like", "you know" and "basically"?

These are a different animal from um and uh. Um and uh are pure stalls; discourse markers like "like", "you know" and "so" often do real work — softening, checking in, signalling a conclusion. A large transcript study by Laserna, Seih and Pennebaker analysed these five fillers separately for exactly this reason: they are not one behaviour.

The three-question test for a "like":

1. Can you remove it and the sentence still means the same thing? If yes, it was a filler.
2. Is it making a comparison ("it was like a wall")? If yes, keep it.
3. Is it quoting a thought ("she was like, why not Pluto?")? If yes, keep it.

"You know" and "basically" pass the same first test. Cut the ones that stand where a pause would.

Is it worth cutting all of them?

No. HBR has made the case for fillers as well as against them: a deliberate "so" before a conclusion, a "look" before a disagreement, or an um that tells the listener a real answer is coming — these hold attention rather than leak it. NPR's piece on the same question lands in the same place: fillers can signal that you are not finished, which is sometimes exactly the signal you want.

The target is not zero. It is a count you chose on purpose, measured in a transcript, moving in the right direction week over week.