Removing filler words without putting words in your guest's mouth
The goal isn't “zero fillers” — it's a transcript that reads the way the conversation felt. Here's the method, including the cuts that are not safe.
Always safe to cut
- Pure hesitations: um, uh, erm, hmm — they carry no meaning in text.
- Immediate repeats: “the the”, “we can we can” — transcription artifacts of natural speech.
- Orphaned punctuation left behind by the cuts: doubled commas, comma-before-period.
Only safe in the discourse-marker position
Words like like, right, well, actually, basically, you knoware fillers only when they sit alone between punctuation: “, like,” or “. Well,”. Inside a sentence they usually carry meaning:
✂️ “It was, like, a huge shift” — filler, cut it.
🚫 “It looks like a huge shift” — comparison, keep it.
✂️ “Actually, the numbers went up” — throat-clearing, cut is fine.
🚫 “What actually happened was different” — contrast, keep it.
Never cut
- Hedges that are the point — “I think”, “probably”, “in my experience”: removing them turns opinion into claim. That's an integrity problem, not a style choice.
- Characteristic speech — if your guest says “y'know what I mean” as a signature, keep one or two instances; the transcript should still sound like them.
- Anything inside a quote you'll publish as a quote — clean around it, not inside it, or mark edits with brackets.
The 30-second version
Our free cleanup tool implements exactly these rules — conservative by default, discourse-marker position only — and strikes through every removal so you can audit it. Paste a transcript or transcribe audio locally in your browser.