Overusing Em Dashes Makes Voice Agents Sound Robotic

Follow me on LinkedIn - AI, GA4, BigQuery

Text-to-speech voices usually treat an em dash as a pause. So when your prompt has too many em dashes, the voice agent does not just “read smoothly.” It pauses again and again.

That can make the caller feel like the agent is reading punctuation instead of having a conversation.

Check the demo below:

0:00

/1:08

Why does this matter?

An em dash creates a noticeable break in speech.

That can be helpful when the agent is reading back important information, such as a phone number:

“5-5-5, 1-2-3-4 — right?”

The pause gives the caller a moment to check the number.

But the same pause sounds awkward in short everyday phrases, such as:

“Got it — Michael — how old is he?”

That sentence should sound quick and natural. The em dashes make it sound broken up and robotic.

Use em dashes when the pause helps the caller.

Use them for:

reading back phone numbers, emails, or addresses.
separating important information before a confirmation question.
creating a pause the listener actually needs.

Avoid them for:

short replies like “Got it”.
simple name confirmations.
normal conversation flow.
style or decoration.
places where a comma or period would sound more natural.

Why does this matter?

Use em dashes when the pause helps the caller.

Related Articles:

Himanshu Sharma