How do you write a call script Chloe can actually run?

A script that sounds great in a Google Doc can fall apart the second a real person says something unexpected. Chloe doesn't improvise around a bad script. She runs it.

Writing for an AI voice agent isn't the same as writing talking points for a human rep who can read the room. Chloe needs branches, not bullet points — a plan for what happens after each answer, not just a list of questions to ask. Getting this right is most of what separates a script that works from one that just sounds fine on paper but falls apart on the first real, messy conversation.

Most first drafts read like an FAQ instead of a conversation. That's the tell that it hasn't been stress-tested against the messy, non-linear way real people actually answer questions on the phone, where nobody sticks to a script and half of every answer contains something you didn't plan for.

What's different about writing for Chloe versus a human rep?

A rep fills gaps with instinct. Chloe fills gaps with whatever you gave her, or with nothing at all. Every question needs a next step attached to every likely answer — yes, no, and the maybe that's actually a no in disguise once you dig one layer deeper into what the lead actually meant.

This means writing more branches than feels necessary at first. A question like 'what's your timeline' isn't one question — it's a question with three or four realistic answers, each pointing somewhere different in the call. Map those branches out before you ever put Chloe on a real lead, because discovering a missing branch mid-rollout means fixing it after it's already happened dozens of times.

How should the qualification questions be structured?

Order them from disqualifying to nice-to-know. If budget or timeline rules someone out, ask that early — not after five minutes of small talk that wastes both their time and yours. Chloe should be able to end a call quickly and politely when the answer is clearly no, instead of dragging a lead through questions that were never going to change the outcome.

This ordering matters for a reason beyond efficiency. A lead who's clearly not a fit shouldn't sit through ten more minutes of questions that were never going to change the outcome. Ending gracefully and early protects your brand as much as it protects your time, and it's the kind of small detail that separates a well-built script from a lazy one that just asks everything in whatever order felt natural to write.

What tone should the script actually sound like?

Write it the way a real person talks, with contractions and short sentences — not the stiff, over-formal copy that makes AI calls feel like AI calls. If it wouldn't sound normal coming out of your best rep's mouth, rewrite it until it does, even if that means several passes before it lands right.

Read every line out loud before you finalize it. Scripts that look fine on a page often sound stilted the moment they're spoken, and that gap is exactly what makes a lead realize they're talking to an AI in the wrong, jarring way instead of a natural one.

How should Chloe handle answers that don't fit the script?

Give her a fallback — a way to acknowledge an unexpected answer and either loop back to the next relevant question or flag the call for human follow-up. Without this, an unanticipated answer can stall a conversation instead of gracefully moving past it and continuing to gather the information you actually need.

A simple fallback line, something like acknowledging what was said and redirecting to the next relevant question, covers most of the situations a rigid script would otherwise fumble. It's a small addition to the script that prevents a disproportionately large share of awkward, stalled calls later on.

How long should a good script actually be?

Long enough to cover every realistic branch, short enough that the call doesn't feel like an interrogation. Most solid qualification scripts run somewhere between five and eight core questions, with a handful of follow-up branches attached to each one. Beyond that, calls start to feel like a form being read aloud instead of an actual conversation between two people.

How do you know the script is actually working?

Listen to the first twenty calls. Not the summaries — the actual transcripts. You'll hear exactly where leads get confused or where Chloe's follow-up question doesn't match what they just said. Fix those spots one at a time. A script is never done after the first draft, no matter how much time went into writing it.

Treat the script as a living document for at least the first month. Small adjustments compound quickly — a script that's been tightened against fifty real conversations behaves very differently than the version you wrote before Chloe ever made a single call on a real lead.

Want a script that sounds like your best rep, not a robot?

We write Chloe's scripts and qualification logic based on how your actual sales conversations go, then test and tighten them against real calls.

Book a call with RevPilot →

‍