Record yourself speaking naturally for two minutes, transcribe it word for word, and read it back. It is a genuinely humbling experience. The ums arrive every few seconds. Sentences restart halfway through. There are ers you would swear you never made.
You are not a bad speaker. This is simply what fluent human speech looks like written down. Fillers are the sound of a brain composing in real time, and in conversation they are completely invisible. On the page they are excruciating.
The short version: switch on filler word removal and PeekoType strips um, uh, er, ah, hmm and similar sounds as it types, tidying the spacing they leave behind. Real words are never touched, because fillers are matched as whole words only. It is one toggle, and it is the difference between a transcript and a draft.
What gets removed
The list covers the sounds English speakers actually make when thinking, in their common spellings and lengthened forms: um, umm, ummm, uh, uhh, er, err, ah, ahh, hmm, mm, uh huh, mm hmm and friends.
Just as important is what happens afterwards. Pulling a word out of a sentence leaves a mess: double spaces, a comma stranded before a full stop, a gap at the start of a line. Removal without cleanup produces text that is arguably worse than what you started with. PeekoType collapses the leftover spacing and repairs the orphaned punctuation, so what lands is text you could send.
The rule that stops it eating real words
Here is the detail that separates a useful feature from a liability.
Fillers are matched as complete words, never as fragments. So umbrella keeps its um. Error keeps its er. Ahead, ahoy, hummus, summer and every other perfectly ordinary word containing those letters passes through untouched.
A naive implementation that searched for the letters would quietly turn umbrella into brella and error into ror, and you would not notice until something embarrassing went out. Whole-word matching is the boring, correct answer, and it is why you can leave the feature on and forget about it.
When to leave it on
Most of the time. Emails, reports, blog drafts, meeting notes, case notes, letters, messages. Anything where a person is going to read the words and judge you a little by them. Which is nearly everything.
The benefit compounds with dictation specifically, because speaking produces far more fillers than typing ever could. When you type, the pause for thought is silent and costs nothing. When you dictate, the pause has a sound, and the sound gets written down.
When to switch it off
There are real cases, and they matter.
Verbatim records. If you are transcribing a research interview, a witness statement, or anything where the record must reflect exactly what was said, hesitations included, turn it off. A pause before an answer can be meaningful evidence, and it is not yours to tidy away.
Linguistic or clinical work. Speech and language therapists, discourse researchers and anyone analysing how someone speaks rather than what they said will want everything preserved.
Transcribing other people. When you are transcribing a recorded meeting or interview, think about whether the transcript is a record or a document. Records should be verbatim. Documents should be readable.
It is a single toggle in the Smart output panel, so switching between the two takes a second.
What it does not do, deliberately
Filler removal takes out sounds that carry no meaning. It does not edit your writing.
Hedges like probably, roughly, I think and sort of are left completely alone, even though a style guide might cut them. They carry meaning and tone, and removing them would change what you said rather than tidying how you said it. False starts, where you begin a sentence and restart it, are also left in place, because deciding which half you meant requires understanding your intent.
That line is drawn on purpose. Automatic cleanup should be something you can trust without re-reading. The moment software starts making editorial decisions on your behalf, you have to check its work, and then it has cost you more than it saved.
The features it works alongside
Filler removal is one of three things quietly turning speech into prose. Automatic punctuation adds the full stops and capitals. Spoken commands let you place specific marks and line breaks. Filler removal takes out what you did not mean to say.
Run all three together and the gap between how you speak and what appears on screen narrows to the point where you stop thinking about it. That is the real goal: not perfect transcription, but text you can use without a tidying pass.
Try the two-minute test
Dictate two minutes of something real with the toggle off. Then dictate the same thing with it on. Most people are startled by the difference, and mildly startled by how many ums they produce, which is a small price for a much better first draft.
If you are new to dictation, our five-minute setup guide is the fastest way in, and the user guide documents every toggle in the Smart output panel.