What listeners actually hear
Fillers don't hurt until you cross a threshold - and "uh" can help
Laske, M., et al. (2024). Do speech disfluencies matter? JABA. See also Fox Tree (2001), Memory & Cognition; Clark & Fox Tree (2002), Cognition.
Filler words are the thing speakers obsess over most, and the evidence says they are mostly fine. Below about five per minute - roughly one every 25-30 words - there is no measurable penalty. The harm concentrates higher up, around twelve per minute. And "uh" is partly functional: it signals an upcoming delay and can actually help listeners process what comes next. Coaching toward zero fillers fights a feature, not a bug.
Why "uh" is not the enemy#
Fox Tree and Clark showed that filled pauses carry information: "uh" tends to flag a short delay, "um" a longer one, and listeners use those cues to allocate attention. A speech sprinkled with the occasional "uh" sounds human; a speech scrubbed perfectly clean can sound rehearsed or robotic. The goal is to thin fillers out, not eliminate them.
Data table
| Item | how much to weight it |
|---|---|
| "um" / "uh" (weight these more) | 70 |
| "like" / "so" / "you know" (weight these less) | 40 |
What it means for Speech Away#
We raised our tolerance deliberately. The coach does not flag a clean-enough take, does not chase "no ums," and weights "um/uh" above "like/so." When fillers are high, the prescription is the remedy the research points to: replace the filled pause with a one-second silent pause - which buys the same thinking time and reads as confidence instead of searching.