To reduce filler words without sounding robotic, stop trying to catch every “um” while you speak. Record one minute, mark the fillers afterward, find one repeated cluster, replace only that cluster with a silent planning pause, and retry the same prompt.
The practical method is awareness, context, pause, retry. It is an editorial framework, not a scientifically proven four-step protocol. The research supports looking at planning demands, filler location, and listener response. It does not support a universal target of zero filler words.
Method
Change one planning point
First recording
Retry
What is actually happening when you say “um”
Spontaneous speech is assembled while you deliver it. You are choosing the next idea, retrieving words, shaping a sentence, and monitoring the listener. A filled pause can appear while part of that planning is still underway.
That does not make every filler useful. A repeated pattern can compete with the message. It does mean that a single total is not enough to explain what happened.
Separate five measures:
- Count: How many marked events occurred?
- Rate: How many occurred per minute?
- Type: Were they “um,” “uh,” repeated words, restarts, or another habit?
- Clustering: Did several appear close together or in the same sentence position?
- Context: What were you trying to do when they appeared?
Use rate to compare similar attempts, not to grade yourself against a universal standard. Six fillers in an improvised answer, a rehearsed introduction, and a relaxed conversation do not necessarily mean the same thing.
What research supports about filled pauses
More difficult planning can produce more disfluency
Heather Bortfeld and colleagues analyzed task-oriented conversations. Speakers were more disfluent when they acted as directors rather than matchers and when they described abstract figures rather than photographs of children. The authors linked those differences to increased planning difficulty (Bortfeld et al., 2001).
This supports a useful diagnostic question: Where does your planning load increase? It may be before a name, while choosing an example, or when moving to the next point. The study does not show that every filler has one cause or establish an acceptable filler rate.
Listeners can use disfluency as information
Jennifer Arnold and colleagues found that speakers were more likely to become disfluent before referring to discourse-new objects than familiar ones. In an eye-tracking experiment, a disfluent instruction shifted listeners’ attention toward the new object (Arnold, Fagnano, and Tanenhaus, 2003).
Jean Fox Tree reported a narrower result. In English and Dutch word-recognition studies, hearing “uh” helped listeners recognize an upcoming word, while “um” neither helped nor harmed recognition (Fox Tree, 2001). Our separate article explains why “uh” and “um” did not behave identically in that research.
Martin Corley and Robert Hartsuiker later found that “um,” a silent delay, and a tone could all improve recognition of the next word relative to no delay. They argued that extra processing time may explain at least part of the effect (Corley and Hartsuiker, 2011).
These studies do not show that fillers improve every speech. They show that listeners do not always process a hesitation as meaningless noise, and that more than one mechanism may be involved.
The outcome depends on the listening task
Scott Fraundorf and Duane Watson tested recorded stories. In two experiments, stories containing six filled pauses were recalled better than fluent versions. Duration-matched coughs did not produce the same benefit (Fraundorf and Watson, 2011). Our related article explains what the recall study found and what it did not prove.
A 2024 study by Zachary Houghton and colleagues tested silent pauses rather than “um.” Among 277 listeners, pause presence and location affected fluency ratings much more clearly than credibility ratings. Pauses inside a clause were rated as less fluent than pauses between clauses, and both were less fluent than no pause (Houghton et al., 2024).
The studies reviewed here used conversations, object instructions, word-recognition tasks, recorded stories, or read trivia statements. None established an ideal filler rate for a pitch, interview, or formal presentation. Apply the findings cautiously to the situation you actually want to improve.
How to reduce filler words naturally
1. Awareness: record before you correct
Choose a realistic prompt. Answer an interview question, explain a current decision, or introduce one presentation section without reading a script.
Record for one minute. Do not stop or restart. Listen back and mark only what you hear clearly:
[um]and[uh]- repeated words, such as
[we we] - abandoned starts, such as
[I wanted, I decided] - “like” or “so” only when the word is functioning as a repeated hesitation
[pause]when a silence seems relevant
Count after you finish. Keep the marked transcript because location matters more than the final number.
2. Context: find one repeated cluster
Describe one pattern in a sentence: “Most fillers appeared before key nouns,” “I said ‘um’ at every transition,” or “I repeated words when I changed direction.”
Choose one cluster, not every imperfection. A specific cluster gives you something to practice. A total only tells you that something happened.
3. Pause: replace the cluster at a clean boundary
Choose a place where silence will not split a phrase in half. Useful planning points often appear before a new sentence, list item, example, or central detail.
Practice one short silent pause there. Do not insert dramatic gaps throughout the answer. Houghton and colleagues found that pause location mattered for perceived fluency, with within-clause pauses rated less fluently than pauses between clauses.
4. Retry: change one thing
Answer the same prompt again. Keep the content broadly similar, but do not memorize a transcript. Focus only on the selected cluster and allow other natural hesitations to remain.
Compare the attempts:
- Did the chosen cluster occur less often?
- Did the pause sit at a cleaner boundary?
- Did the words after it begin more clearly?
- Did the second attempt still sound like you?
The goal is more control, not a performance scrubbed of every sign of thought.
Before and after: fix the planning point
First attempt
Last Saturday I went to, um, the garden center because I needed, um, a larger pot for, um, the lemon tree on my balcony. The assistant suggested, um, a terracotta one, and I bought it because it drains better.
The fillers cluster before key nouns and noun phrases. Before retrying, reduce the idea to three anchors: garden center, larger pot, lemon tree. These are prompts, not a script.
Natural retry
Last Saturday I went to the garden center because I needed [pause] a larger pot for the lemon tree on my balcony. The assistant suggested a terracotta one, and, um, I chose it because it drains better.
The retry still contains one “um.” That is fine. The distracting cluster changed, and the pause now gives the speaker time before the central detail.
Try this one-minute filler word practice
- Record a one-minute first attempt to one realistic prompt.
- Mark each filler, repeat, restart, and relevant silent pause.
- Circle one repeated cluster or location.
- Choose one clean pause point that could replace only that cluster.
- Record a second attempt to the same prompt.
- Compare the words immediately after the selected planning point. Do they begin more cleanly?
One retry will not necessarily change a long-standing habit. Repeat the exercise with similar prompts and check whether the same cluster returns.
Why too much self-monitoring can sound stiff
Speech already requires planning and monitoring. In one experiment, 24 participants described visual networks at normal and faster rates. Under time pressure, they detected and repaired errors more quickly, while detection accuracy did not change (Oomen and Postma, 2001).
That study did not test filler-word coaching and does not prove that self-monitoring makes people robotic. The practical inference is narrower: adding “catch every um now” to an already demanding speaking task may pull attention away from the next idea.
Review after the attempt. During the retry, hold one planning point in mind.
When this advice does not fit
Leave an occasional filler alone when it does not form a distracting pattern or interfere with understanding. Practice in the context that matters because a relaxed conversation does not reproduce an interview or prepared talk.
Do not count every “like” or “so” automatically. Both can perform ordinary grammatical or conversational work. This guide also does not diagnose or treat a speech, language, or medical condition.
Frequently asked questions
Are filler words always bad?
No. Filled pauses can reflect planning and can affect how listeners process what follows. They can also distract when they form a noticeable pattern. Judge the type, location, clustering, and context rather than treating every “um” as damage.
How many filler words are too many?
The research reviewed here does not provide a reliable universal number. Use your count or rate to compare similar attempts. A pattern is worth addressing when it repeatedly competes with the message or appears in a situation you need to handle better.
Why do I use more fillers under pressure?
Time pressure can change how quickly speakers monitor and repair their speech, and more difficult planning tasks can produce more disfluency. That does not prove that every person uses more fillers under pressure. Record the same type of pressured prompt several times and inspect your own pattern.
Do pauses make me sound confident?
Not automatically. A pause can provide planning time, but placement matters. A short pause between ideas may sound natural, while an interruption inside a phrase may reduce perceived fluency. Use silence to support clear phrasing, not as a trick that is supposed to manufacture confidence.
Should I eliminate “like” as well as “um”?
Only when “like” is the repeated habit you want to change. It can also function as a verb, preposition, quotation marker, or conversational signal. Mark the specific use and location before deciding that it is a filler.
Practice with a manual tally
The free speaking.app Filler Counter is a manual tally tool. Play your first recording and tap when you hear a filler, then repeat the process for the retry. It does not require a microphone, record your audio, or decide automatically what counts.
Keep the attempts separate. The useful result is knowing which pattern changed, where you paused, and whether the retry still sounded natural.
Sources
- Bortfeld, H., Leon, S. D., Bloom, J. E., Schober, M. F., and Brennan, S. E. (2001). Disfluency rates in conversation. Language and Speech, 44(2), 123-147.
- Arnold, J. E., Fagnano, M., and Tanenhaus, M. K. (2003). Disfluencies signal theee, um, new information. Journal of Psycholinguistic Research, 32, 25-36.
- Fox Tree, J. E. (2001). Listeners’ uses of um and uh in speech comprehension. Memory & Cognition, 29(2), 320-326.
- Corley, M., and Hartsuiker, R. J. (2011). Why um helps auditory word recognition. PLOS ONE, 6(5), e19792.
- Fraundorf, S. H., and Watson, D. G. (2011). The disfluent discourse. Journal of Memory and Language, 65(2), 161-175.
- Houghton, Z., Kato, M., Baese-Berk, M., and Vaughn, C. (2024). Task-dependent consequences of disfluency. Applied Psycholinguistics, 45(1), 64-80.
- Oomen, C. C. E., and Postma, A. (2001). Effects of time pressure on speech production and self-monitoring. Journal of Psycholinguistic Research, 30, 163-184.
