Remove Filler Words From a Transcript: Um, Uh, Like

Choose clean verbatim or true verbatim

Transcription professionals use a few standard styles. Pick one before you start editing, and apply it consistently across the whole document.

• True verbatim: every um, uh, stutter, false start and laugh is kept, often with pauses marked
• Clean verbatim: fillers, stutters and false starts go, but wording and grammar stay as spoken
• Edited: grammar is lightly corrected and sentences may be tightened for publication
• For quotes attributed to a real person, stay at clean verbatim unless they approve edits

Keep true verbatim for legal depositions, police interviews, linguistic research and some qualitative studies, where a pause or a self-correction can carry meaning. For podcasts, marketing and meeting notes, clean verbatim is the usual standard.

Start with a cleaner draft in AI Studio

The Whisper models in AI Studio tend to leave out many ums and uhs on their own and add punctuation, so the draft is often closer to clean verbatim than to true verbatim. That saves time, but it also means the output is not a reliable record of hesitations if you need them.

Transcribe on your device, then export TXT for editing in Word or Google Docs, or SRT if the cleaned text will become captions. With SRT files, edit only the words inside each cue and leave the numbers and timestamps untouched; the captions will stay in sync because the timing belongs to the cue, not to individual words.

Paste the raw text into the Word Counter before you start. Knowing the original count lets you see how much the cleanup removed, which is a useful sanity check: if a 5,000-word interview drops to 3,500, you have probably cut real content, not just fillers.

Strip the obvious fillers with find and replace

Pure sounds such as um, uh, er, erm and hmm almost never carry meaning, so they are safe to delete in bulk. In Word, press Ctrl+H or Cmd+H, tick Find whole words only, and replace each with nothing. In Google Docs, open Find and replace, tick Match using regular expressions, and use a single pattern.

A pattern that works in Google Docs and most editors is \b(um|uh|er|erm|hmm)\b,? followed by a space. It catches the word, an optional comma and the trailing space, so you are not left with double spaces or stray commas. Run it once, then search for two spaces in a row and replace them with one.

• Check the start of sentences afterwards; removing Um, can leave a lowercase first word
• Search for doubled words such as the the and I I, which come from stutters
• Cut false starts like we went, we drove only when the second version is clearly the intended one
• Keep a list of the patterns you use so the next transcript takes seconds

Judge the context-dependent words by hand

Some terms are fillers in one sentence and meaning in the next. Deleting them in bulk changes what people said, so review each one:

• Like: I like the plan must stay; it was, like, huge can lose the filler
• You know, I mean, sort of and kind of: cut when they are padding, keep when they hedge a claim
• So, well, okay and right at the start of an answer: usually safe to drop
• Actually, basically and literally: keep when they change emphasis or correct something

When you finish, paste the raw and cleaned versions into the Text Diff tool. It highlights every change, so you can scan for anything you did not mean to remove. If you want AI to smooth the text further, the AI Writer’s Improve and fix grammar mode can help, but it sends text to a cloud AI and may reword sentences, so diff the result again and keep it away from confidential material.

Step-by-step

Common mistakes to avoid

Pro tips

Frequently asked questions