How to stop saying um
"Um" is not a sign that you do not know what you are saying. It is the sound of planning while speaking, which is why it appears most often at the exact moment a sentence turns. The fix is not to remember not to say it — that makes it worse — but to put something quieter in its place: a closed mouth and a breath. Below is how many fillers is too many, where they come from, and four drills that replace the sound instead of suppressing it.
How many is too many
Fillers are counted as a rate, not a total, because a four-minute answer has more room for them than a forty-five second one. Podivm counts them per hundred words, and uses the same three bands the analysis code has always used:
| Rate | What it sounds like |
|---|---|
| Under 2 per hundred words | Rare. Most listeners would not be able to tell you there were any. |
| 2 to 4 per hundred | Regular. Noticeable if somebody is listening for it, invisible if they are listening to the point. |
| Over 4 per hundred | Frequent. The fillers start competing with the words that carry meaning. |
For scale: four per hundred words in a 130-word-a-minute talk is about five a minute, or one every twelve seconds. That is roughly the density at which a listener starts hearing the hesitation rather than the sentence.
What counts as a filler
The count is not only "um". Two separate rates are measured, and the report combines them:
- Fillers: um, uh, erm, ah, eh; "you know", "sort of", "like"; and a sentence that ends on "right", "yeah", "okay" or "huh" as a bid for agreement.
- Hedges: just, really, actually, basically, literally, honestly, obviously, probably, maybe, somewhat, and phrases like "or anything like that". These do not sound like hesitation, but they take the edge off a claim, which is the same move in different clothes.
Both are matched as patterns in the transcript, which has one consequence worth knowing: "like" is counted wherever it appears, so a simile ("it went like a shot") costs you a count even though it is doing real work. The bar is a rate, not a verdict on any single word. If your only fillers are similes, the number will be slightly unkind and the rest of the report will still be right.
Where they come from
Three causes, and they need different drills:
You are planning the next sentence while saying this one. This is the common one. The mouth keeps moving because stopping feels like losing the room. The filler is the sound of the buffer running.
You are holding the floor. "Um" tells the listener not to speak yet. On a phone call it is doing a job. At a podium nobody is going to interrupt you, so the job is already done and the sound is pure cost.
You are softening a claim. "I just think maybe we should probably..." is not a filler problem, it is a conviction problem, and no amount of pause drills will fix it. That one gets fixed by deciding what you actually mean before you stand up.
Four drills
1. The closed mouth. Say a sentence out loud, then stop with your lips together for one full breath before the next one. Do it for a paragraph. The point is to teach your face that a closed mouth is a valid state while speaking. Most people have never practiced it, which is why silence feels like an emergency.
2. Breathe through the gap. Take the same paragraph and make every pause an actual breath in through the nose. A filler and a breath occupy the same slot in the sentence, so giving the slot a job removes the need for the noise. This is the single most effective drill in this guide, and it is the one that survives nerves.
3. One sentence, then stop. Answer a question in exactly one sentence and then say nothing. It will feel like eight seconds; it is usually about two. The drill trains the ending, which is where fillers cluster: people who are unsure how to stop will fill the space after the last useful word.
4. Slow by a tenth. Take your normal pace down by about ten percent and record again. Planning noise drops when there is more time to plan, and ten percent is small enough that nobody in the room will call it slow. If the filler rate falls with the speed, you have found the cause.
Count by recording, not by memory
Nobody can count their own ums while saying them. Ask someone after a talk how many they used and the answer is a guess, usually low, because the sound is invisible from the inside. Record forty-five seconds, then read the transcript and count. It takes four minutes and it is the only version of the number that is real.
One caveat about tooling: a transcript read by a small on-device model can swallow a short hesitation, and a browser's own recognizer can too. Podivm reports the filler bar from whatever transcript it actually got, and the report names the source. A missing reading is reported as missing, not as a clean zero — a zero would be the most flattering possible lie.
What not to do
Do not practice not saying "um". An instruction phrased as a prohibition puts the word at the front of your attention, and the sentence you are trying to say has to compete with it. Practice the pause instead. The pause is a thing to do, and a thing to do is easier to remember under pressure than a thing to avoid.
Find out how many you actually say
Forty-five seconds of talking, counted per hundred words and compared with the setting you pick. The take stays in the tab, and the first cycle is free.
Start a 45-second take