Um, uh, like — how many filler words is actually too many?

There's no universal threshold, and any fluent speaker uses some — the honest answer is that it's a question of density relative to the length of what you said, and the penalty for having some fillers is intentionally gentle because eliminating them entirely isn't the goal.

30-second version: sayit tracks a specific set of markers — um, uh, er, like, you know, literally, basically, actually, so — counted case-insensitively with punctuation stripped, and the two-word "you know" matched as one unit. Filler density can shave up to a fifth off your fluency score, but the penalty is deliberately capped and gentle, because some filler use is completely normal in real speech and a score that punished it harshly would be scoring against how people actually talk.

Why fillers matter less than most learners assume

A small number of fillers, spread naturally through a longer answer, rarely bothers a listener — they're part of how spoken language actually works, buying a speaker a half-second to find the next word without going silent. What actually damages intelligibility and confidence is a high density of fillers clustered together, especially at the start of sentences or right before a word the speaker is unsure about, because that pattern signals hesitation in a way a listener notices consciously.

This is also why the fluency score caps how much fillers can cost you — the cap exists specifically so that a mostly-clean take with one or two natural fillers doesn't get treated the same as a take where every third word is "um." Density, not presence, is what's actually being measured.

Which words actually count

The exact list sayit tracks: um, uh, er, like, you know, literally, basically, actually, and so. A few of these are context-dependent in real usage — "so" can be a genuine connector ("so, given that,") rather than a filler, and "like" can be a real comparison word ("it felt like rain") rather than a hesitation marker. The detector counts the word regardless of that distinction, which is a reasonable trade-off for automatic detection but means the raw count is a signal to interpret, not a perfect transcript of every hesitation you made.

What actually reduces the count

The two changes that reduce filler density fastest aren't about suppressing the urge to say "um" — they're about giving yourself somewhere else to put the pause:

• Replace the sound with silence. A short, deliberate pause carries the same "I'm thinking" signal as "um," without registering as a filler at all. This feels unnatural at first because silence feels exposed in a way a filler sound doesn't — that discomfort fades with practice.
• Plan the next clause, not the next word. Fillers cluster most at clause boundaries, when a speaker has finished one idea and hasn't yet formed the next. Practicing longer, planned chunks — three or four words decided in advance — reduces the number of boundaries where a filler has room to appear.

Drill it

Record a 30–60 second freeform answer to a simple prompt — "describe your morning routine" works well — rather than a scripted sentence, since fillers mostly show up in unscripted speech. Check the filler count and list against the length of the response: three fillers in a 20-second answer is a very different density from three fillers in a two-minute one.

How do I know it's improving?

Track density (fillers per minute or per hundred words) rather than the raw count, since a longer, more confident answer will naturally contain more words and possibly more fillers in absolute terms while still being an improvement in density. A count that drops while your speaking length grows is real progress even if the raw number looks similar.

Try it

Open sayit free — no install, no card — and record a short freeform answer to any question. Check the filler list and count against the fluency score, then try the same prompt again with a deliberate silent pause standing in for your usual filler.