Free tool

Filler word counter

Count the ums, uhs, likes and you-knows in anything you say or write. Paste a transcript, or speak and let your browser transcribe. Free, instant, no signup.

Filler words

0

0

total words

0.0

per 100 words

--

per minute

Counting them is the easy part

Fillers appear where planning happens, so they move when your structure changes, not when you try to suppress them. Wellspoken listens to a short recording, finds where yours cluster, and drills the pauses that replace them.

Find where your fillers cluster

Questions

  • How many filler words is too many?

    There is no clean threshold, and a handful of fillers in spontaneous speech is completely normal. What listeners notice is clustering. A few scattered through a five minute talk pass unheard, while six in one sentence pulls attention away from your point. The per-100-words figure above is the number to watch over time, not any single count.

  • Is saying um actually bad?

    Less than most people fear. Research on hesitation treats um and uh as conventional signals that a pause is coming rather than as errors, and listeners largely tune them out at low rates. They become a problem when they replace the pauses that would otherwise give an audience time to follow you.

  • Why does the speak mode find fewer fillers than I expect?

    Browser speech recognition is built to produce clean, readable text, so it discards many hesitation sounds before they ever reach the page. That makes the live mode useful as a rough read and unreliable as a measurement. Pasting a transcript from a real recording gives an accurate count.

  • Which words does this count as fillers?

    Four groups: hesitation sounds such as um, uh and er; discourse fillers such as like, you know and I mean; qualifiers such as basically, actually and literally; and sentence crutches such as so, right and well. Multi-word phrases are matched first, so you know is never also counted as know.

  • How do I stop using filler words?

    Replace them with silence rather than trying to suppress them. Fillers appear where planning happens, so the durable fix is knowing what your next point is before you start the sentence. Pausing fully at transitions gives you that planning time and removes the gap the filler was covering.

What counts as a filler

Filler is a broader category than um and uh. The tool above tracks four groups, because they behave differently and respond to different fixes. Hesitation sounds tend to vanish once you get comfortable pausing. Qualifiers are a writing habit that survives into speech and usually disappear the moment you notice them.

The four groups of filler words counted by this tool
GroupWordsWhat it signals
Hesitation soundsum, uh, erm, er, ah, hmm, mmPure stalling while the next phrase is assembled.
Discourse fillerslike, you know, i mean, sort of, kind of, kinda, sortaReal words used as padding between ideas.
Qualifiersbasically, actually, literally, honestly, obviously, essentially, justUsually add nothing to the sentence they sit in.
Sentence crutchesso, right, okay, well, anyway, yeahOpeners and closers running on autopilot.

Fillers are a symptom, not the disease

Every um sits at a point where you needed a moment and did not want to give up the floor. That is why suppression rarely works for long. Tell someone to stop saying um and they usually replace it with a different filler, or with a stretched vowel, because the underlying pause is still needed.

The pattern in your count is more useful than the count itself. Fillers cluster at transitions, when you move from one idea to the next and have not decided yet what the next one is. They also cluster on topics you know least well. Run the same person through a rehearsed introduction and an unexpected question and the second will produce several times as many.

For the research on how listeners actually interpret hesitation, and why silent pauses read so differently from filled ones, see filled pauses versus silent pauses.

Getting a count you can trust

Speech recognition built into browsers is optimised for clean output. It is doing its job when it silently removes your hesitations, which makes it the wrong instrument for measuring them. Expect the speak mode here to undercount, sometimes badly.

A transcript from a real recording is far more reliable, because the disfluencies survive into the text. Meeting tools, podcast editors and most transcription services will export one. Paste it into the box above and the count is exact.

Measure the same way twice before drawing conclusions. A single sample tells you about one conversation, not about how you speak. Three samples across different situations, one rehearsed and two not, give you something worth acting on.

Related tools