The Ums Are Telling You Something

Open almost any transcript and the first thing you want to do is clean it up.

Delete the "um." Delete the "uh." Delete the "like, you know, kind of." Merge the false start into the real sentence. Make it read like someone wrote it.

That instinct is right for some jobs. For others, it quietly throws away the most honest part of the conversation.

Two people said "yes." Only one meant it.

Picture two answers to the same question: "Can we ship this by Friday?"

A: Yes.

B: Um… yeah. I mean — yes, probably. If the review goes fine.

Run both through an aggressive cleanup and you can end up with:

A: Yes.

B: Yes, if the review goes fine.

Same headline. Very different meetings. The first person committed. The second person hesitated, restarted, added a hedge, and attached a condition. Anyone who was in the room heard the difference instantly. The cleaned transcript erased it.

The words survived. The confidence level did not.

What hesitation actually encodes

Filler words and disfluencies are not random static. They tend to show up around specific kinds of moments:

• Uncertainty. "It's, uh, around forty thousand?" is not the same claim as "It's forty thousand."
• Reluctance. A long "well…" before agreeing often means the agreement is soft.
• Searching for a word. The pause can mark exactly where a speaker is unsure about terminology — useful in interviews and research.
• Self-correction. "We launched in March — sorry, April" contains a fact and a fix. Keep only the fix and you lose the clue that dates are fuzzy here.
• Changing course. A sentence abandoned halfway and restarted usually means the speaker decided not to say the first version. That is worth knowing.

None of this is mystical. It is just information that lives in how something was said rather than what was said. A good transcript can keep a surprising amount of it, if you let it.

Verbatim, clean, and the space in between

Transcription people usually talk about three levels:

1. Strict verbatim — every um, uh, repetition, stutter, and false start, often with pauses marked.
2. Clean (or intelligent) verbatim — fillers and stumbles removed, grammar left mostly as spoken, meaning preserved.
3. Edited — rewritten for reading, closer to a summary in the speaker's voice.

There is no universally correct level. There is only the right level for what you plan to do with the text.

The mistake is not choosing clean verbatim. The mistake is choosing it by default, before you know which parts of the conversation will matter.

When you should keep the ums

Keep the messy version when the hesitation itself might be evidence:

• User and customer research. A customer who says "I'd… probably pay for that" is giving you a much weaker signal than one who says "I'd pay for that." Cleaning turns a maybe into a yes.
• Negotiations and sales calls. Pauses before price, hedges around timelines, and walked-back offers are the whole game.
• Decision records. If a meeting "agreed" on something, it helps to see whether the agreement was crisp or reluctant before you build on it.
• Journalism and quotes you might be challenged on. If someone disputes a quote later, the raw version — stumbles and all — is your receipt.
• Language and speech work. Anyone studying how people actually talk needs the disfluencies; they are the data.

When you should strip them

Clean it up when the text is going to be read for content and the speaker's certainty is not in question:

• Publishing a podcast or interview transcript for an audience
• Turning a lecture into study notes
• Pulling quotes for an article (cleaned with care, without changing meaning)
• Captions or subtitles where reading speed matters
• Internal notes where you only need the facts, not the mood

Nobody wants to read a keynote transcript where every third word is "um." That is not honesty. It is just friction.

The order of operations matters

Here is the practical rule: transcribe raw first, clean second.

If your only copy of a conversation is already cleaned, you cannot get the hesitations back. The pause is gone. The restart is merged. The "probably" may have been smoothed away by whoever (or whatever) tidied it up.

If you have the raw transcript, you can always produce a clean version for reading — and still go back to the messy one when someone asks, "Wait, did she actually say yes?"

Think of it like photos: you can always crop later, but you cannot un-crop.

A simple workflow:

1. Get the full transcript of the recording, as close to what was said as possible.
2. Save that version next to the audio or video.
3. Make a clean copy for sharing, publishing, or notes.
4. When a detail is disputed or important, check the raw transcript before trusting the clean one.

A quick gut check

Before you scrub a transcript, ask one question:

Would it change anything if I knew how sure the speaker was?

If the answer is no — a lecture, a recipe, a tutorial — clean away. If the answer is yes — a commitment, a price, a feeling, a claim someone might deny later — keep the ums somewhere you can find them.

Soft close

Filler words are annoying to read and easy to delete. That is exactly why they get deleted too early. But a hesitation is part of the message, and sometimes it is the message.

To The Text turns recordings and video on your iPhone into editable text, so you can keep the full version of what was said and tidy up a copy for everything else.