English filler words are sounds or phrases such as um, well, like, you know, and I mean that help speakers hold a turn, plan the next words, soften a response, or show approximation. They are normal features of spontaneous speech. The goal is not to remove them all, but to keep useful ones and reduce automatic repetition.
If you have ever heard a recording of yourself and thought, “Why did I say um seventeen thousand times?”, good news: your English is not collapsing. Spontaneous speech is messy because people plan while they talk. Research on natural conversation treats pauses and fillers as part of that real-time planning and interaction, not simply as verbal rubbish. If your bigger problem is stopping, restarting, or losing the whole sentence, that belongs to how to stop hesitating when speaking English. This guide has a narrower job: choosing what to do with the little words and sounds that appear while you think.
For the broader fluency path, see how to speak English fluently.
What filler words do
Filler words can buy planning time, help a speaker keep the floor, soften a response, signal uncertainty, or help repair what was just said.
The label filler word is useful for learners, but it hides an important distinction. Um and uh are usually discussed as filled pauses. Expressions such as I mean, you know, and like can behave as discourse or pragmatic markers with recognizable conversational functions. In one natural-language study, those forms even separated into two groups: uh/um as filled pauses and I mean/you know/like as discourse markers.
| Type | Typical job | Example |
|---|---|---|
| Filled pause | Signal a delay while planning | “I’d choose… um… the earlier train.” |
| Lexical hesitation device | Buy time while keeping a natural turn | “Well, I’d need to check first.” |
| Discourse/pragmatic marker | Adjust, connect, soften, or frame the message | “I mean, the plan is fine; the timing is the problem.” |
Clark and Fox Tree’s corpus-based work on uh and um argues that speakers can use them as signals of an upcoming delay, including while searching for a word or deciding what to say next. Bortfeld and colleagues also found that disfluency rates changed with planning difficulty and conversational role. In other words, a filler can be evidence that your speaking system is doing work—not proof that your English is “bad.”
Neutral conversational fillers
Neutral conversational fillers are most useful when they do a small social or planning job without taking over the sentence.
Start with three high-value forms: well, you know, and I mean. They are common in spontaneous conversation, but they are not interchangeable.
A: Did you like the new restaurant?
B: Well, yeah. I mean, the food was great, but it was, like, really loud.
- Well gives B a soft start instead of jumping straight into the answer.
- I mean signals an adjustment: B is adding the part that makes the first answer more accurate.
- Like makes the description looser and more conversational here. It is optional; deleting it would not change the core message.
You know often invites the listener to connect the idea with shared knowledge or an inference. I mean often warns that an adjustment or clarification is coming. Those are real conversational jobs, not empty decoration. Fox Tree and Schrock’s analysis of spontaneous speech is especially useful here because it shows why counting every you know or I mean as a mistake would be crude.
Actually can also appear in the same conversational territory, but it often has a clearer job than a filler: correction or redirection.
A: So you’re leaving on Friday?
B: Actually, Saturday morning.
Annotation: Actually is not merely buying time. It corrects the listener’s assumption.
You can hear the same overlap in the live Friends S1E1 Discourse Markers guide, where well, so, you know, I mean, and actually are tied to specific scene functions. That episode page is a supporting example; F10 owns the general filler-word decision.
Stalling phrases
Stalling phrases are short expressions that openly buy thinking time and are often cleaner than repeating um when you need more than a split second.
The trick is to choose a phrase that fits the setting. A casual conversation can tolerate a direct “Give me a second.” An interview needs something more purposeful. A presentation usually benefits from a quiet pause rather than narrating every moment of thinking.
| Context | Useful stalling phrase | Why it works |
|---|---|---|
| Casual | “Let me think.” / “Give me a second.” | Direct, relaxed, and honest. |
| Interview | “Let me think of the clearest example.” | Buys time while showing what you are selecting. |
| Presentation Q&A | “There are two parts to that.” | Creates planning time and gives the answer a structure. |
A stalling phrase should not become your new filler habit. If every interview answer starts with “That’s a great question,” the phrase stops helping and starts sounding automatic. Use it when you genuinely need a moment.
Interviewer: Tell me about a time you changed your mind after getting new information.
Weak habit: “Um, so, I, you know, I guess there was this project…”
Cleaner: “Let me think of the clearest example. [silent pause] Last year, I changed our launch plan after user testing showed…”
Annotation: The silent pause is doing useful planning work. It does not need to be filled just because another person is waiting.
Fillers that signal uncertainty or approximation
Some filler-like expressions soften certainty or make a number, description, or judgment deliberately approximate.
Like is the obvious example, but it has several different uses in English. Alexandra D’Arcy’s corpus-based work traces multiple grammatical and pragmatic functions of like across English, which is a useful antidote to the idea that every like is meaningless teenage noise.
- “It took like twenty minutes.” → approximately twenty minutes.
- “It was kind of awkward.” → the judgment is softened.
- “There were about thirty people.” → an approximate number without a casual filler feel.
- “I guess the second option is safer.” → the speaker presents the conclusion with less certainty.
These choices are not equal in register. “It cost, like, fifty dollars” can sound perfectly ordinary in casual conversation, while “approximately fifty dollars” fits a formal report and sounds ridiculous over pizza. In an interview, “about fifty customers” is usually safer than “like fifty customers” if the number is only approximate.
Do not remove uncertainty that is part of your meaning. If you are unsure, language such as probably, roughly, about, I think, or it seems may make the answer more accurate. The goal is clarity, not fake confidence.
Register and dialect differences
Filler use changes with situation, community, speaker, and variety of English, so one universal “correct filler list” would be misleading.
Even the way filled pauses are written varies. English transcripts may show uh, um, er, or erm; British corpus conventions have traditionally used forms such as er/erm, while uh/um are common in American English and increasingly familiar more broadly. Treat the spellings as rough written representations of speech sounds, not as pronunciation laws.
Register matters too. Research on natural speech finds that disfluency patterns vary with task, role, topic, and setting. That is exactly what you should expect: chatting with a friend, answering an interview question, and presenting quarterly results create different planning pressures.
| Context | Good default | Example |
|---|---|---|
| Casual conversation | Keep useful softeners and repair markers | “Well, I liked it. I mean, I’d go again.” |
| Job interview | Use a short pause or purposeful stalling phrase | “Let me choose one example. [pause] In my last role…” |
| Presentation | Prefer deliberate pauses; keep markers that show logic | “Actually, the second option costs less. [pause] So, I recommend option B.” |
In that presentation example, actually and so survive because they do jobs: correction and conclusion. The silent pause gives the audience processing time. If formal speaking is your main target, continue with public speaking in English for non-native speakers and English presentation signposting. If you are tempted to solve every fluency problem by talking faster, read how fast should you speak English? first.
When fillers become distracting
Fillers become distracting when the listener starts noticing the repeated habit more than the idea you are trying to communicate.
There is no honest universal rule such as “more than five fillers per minute is bad.” Context changes the judgment. Public-speaking research also suggests that how a filled pause is realized can affect how noticeable it is, so a raw count cannot tell the whole story.
Instead of chasing a magic number, listen for these patterns:
- Same filler, same position: nearly every sentence starts with “So…” or “Basically…” even when it adds nothing.
- Filler chains: “Um, so, like, you know…” before the actual answer begins.
- Inside-phrase interruption: fillers repeatedly split units that would be clearer together: “the customer um support team.”
- Replacement without improvement: you stop saying um but now say you know every few seconds.
- High-stakes mismatch: a casual filler pattern follows you unchanged into an interview or presentation where a short silent pause would be cleaner.
Casual: “Well, I mean, we could go Saturday. It’s, like, less busy then.”
Interview: “[pause] I chose Saturday because demand was lower and we had more staff available.”
Presentation: “Saturday had lower demand. [pause] So, we moved the maintenance window to Saturday.”
What changed? The casual version keeps soft, interpersonal language. The interview strips out loose approximation. The presentation keeps so because it explicitly connects evidence to the decision.
If the real issue is long pauses, searching, and restarting rather than one repeated word, go back to the wider hesitation guide. F10 is not a cure-all for fluency.
Replace, pause, or keep
The fastest way to improve filler use is to decide whether each repeated form deserves to be kept, replaced, or turned into silence.
| Decision | Use it when | Example |
|---|---|---|
| Keep | The word has a clear conversational job | “I mean, the deadline isn’t the problem; staffing is.” |
| Pause | You only need a moment to plan | “The best example is… [pause] our March launch.” |
| Replace | You need more thinking time or a clean repair | “Let me put that more clearly. The main risk is cost.” |
Here is the test: if removing the filler removes a useful signal, keep it. If removing it leaves the sentence clearer and nothing is lost, try a pause. If you need enough time that silence would feel awkward, use one purposeful phrase.
Notice what this rule does not say: “Delete every well, like, and you know.” Speech with zero interactional texture can sound rehearsed, abrupt, or oddly written. Natural speaking is not an audiobook with a pulse.
Self-audit from a recording
A useful filler audit tracks one repeated form in one short recording and judges its function before trying to remove anything.
Use the Job-or-Habit Audit:
- Record one minute. Answer a normal prompt such as “What is one thing you would change about your work or study routine?”
- Choose one target only. Count um, or like, or you know—not every hesitation at once.
- Mark each occurrence. Write K for keep, P for pause, or R for replace.
- Listen to location, not just count. Did it appear before a difficult word, at the start of every clause, inside a phrase, or before a genuine correction?
- Record the answer again. Change only the P and R cases. Leave the useful K cases alone.
- Compare clarity. Did the second answer become easier to follow without sounding robotic? If yes, keep that change.
First take: “Um, I think working from home is better because I can, um, focus more, but um I still like seeing the team.”
Audit: first um = P; second um = P; third um = P. None is carrying a social or meaning job.
Second take: “[pause] I think working from home is better because I can focus more, but I still like seeing the team.”
Do not turn the audit into a punishment spreadsheet. A count is useful only because it helps you notice a habit. Your target is better control, not a score of zero.
Record a one-minute answer and count one filler
Filler versus discourse marker
A filler mainly manages the act of speaking, while a discourse marker helps the listener interpret how one part of the message relates to another—but the same form can sometimes do both.
That overlap is why word lists get messy. Calling so a filler every time it appears would be wrong. In “So, what happened next?”, it helps move the conversation forward. In “So… um… so… I guess…”, the first few items may be mostly buying planning time. Function matters more than the spelling.
| Form | Possible hesitation job | Possible discourse job |
|---|---|---|
| um | Signals a planning delay | Usually treated as a filled pause rather than an idea-organising marker |
| well | Buys a moment before answering | Frames a response, disagreement, or shift in stance |
| you know | Can hold the floor while the speaker continues | Invites the listener to make or share an inference |
| I mean | Can create time during a repair | Signals an upcoming clarification or adjustment |
| actually | Occasionally gives a soft start | Often corrects or redirects an assumption |
| so | Can become an automatic answer starter | Can mark a result, next step, or follow-up |
| like | Can buy planning time in casual speech | Can mark approximation and has several other pragmatic/grammatical uses |
| silent pause | Buys planning time without a word | Can also separate thought groups or highlight a transition |
F17 owns the wider system of markers that organise spoken ideas. If you want to learn how expressions such as first of all, on the other hand, anyway, or result markers guide a listener through a longer answer, continue with discourse markers for spoken English.
The practical rule: do not ask “Is this word a filler?” first. Ask “What is it doing here?” Keep the words that help the listener, pause when you only need planning time, and replace habits that repeatedly blur the message. Natural speech needs room to think. It just does not need every empty second narrated.
Explore more speaking-fluency guides in Speaking Fluency.