The Anatomy of a Great Voicemail Script (and How AI Writes One in Seconds)

A voicemail that gets results follows a tighter shape than "some structure with a reason and a callback number." The data on what that shape actually looks like is more specific - and in one place, more contested - than most template libraries let on.

The duration budget is non-negotiable

Gong's data, pulled from analysis of hundreds of millions of cold calls, puts the ceiling at 15 to 25 seconds - a hard budget of roughly 55 to 60 words at normal speaking pace. Past 30 seconds, retention drops off a cliff and the message reads as a pitch instead of a note. This is the one point every source agrees on without exception.

Experts genuinely disagree on the opening

The traditional structure, the kind in most template libraries, opens with identification: "Hi, this is [name] from [company]." John Barrows' data-backed alternative argues the opposite - naming yourself and your company in the first three seconds triggers a reflexive deletion, so his framework leads with the reason for the call and defers the rep's name and number to the final few seconds, forcing the listener to process the value before deciding whether to screen it out.

Neither camp disagrees on the middle: no product pitch, no feature list, just one specific, research-backed reason the call is relevant to that company right now. That's exactly what an AI-generated dossier feeds into the script - a talking point grounded in real research, not a templated value proposition that could apply to any prospect.

What both structures look like in practice

Barrows' reverse framework, in his own words, reads roughly like this: "Hi [name], the reason for my call today is that we're working with other [persona] right now to help them address [specific pain], and I thought it would be worth a quick conversation. If you get a chance, give me a call back at [number]. This is [rep name] with [company], [number]." Gong's own "double-tap" alternative is shorter still, built to run under 15 seconds: "Hi [name], I noticed [specific trigger event]. I just sent you an email with the subject line '[X]'. Take a look when you get a chance. No need to call me back." Both are specific, both avoid a pitch, and both would fail without a real researched detail to slot into the bracket.

Neither script works as a fill-in-the-blank template on its own - the bracketed parts are exactly where a generic script collapses into "just checking in" filler if nobody does the work to fill them in with something true.

Why AI is well-suited to a genuinely contested structure

A disagreement this concrete is actually a good fit for AI-assisted generation rather than a reason to avoid picking a side. A human writing scripts by hand effectively has to bet on one camp - Barrows' reverse structure or the traditional opener - and live with that choice across every rep and every persona. Generating the script from a dossier makes it just as cheap to produce either shape, or to try both across a segment and see which earns more callbacks for a specific persona or industry, instead of committing to one expert's opinion company-wide.

The callback ask should probably disappear entirely

Gong's own recommended script skips the callback request and points the listener to an email instead - "no need to call me back."

Direct callback rates on a cold voicemail sit under 5% regardless of script quality. Rather than fight that number, Gong's own recommended structure skips asking for a callback and redirects to email:

• "I just sent you an email with the subject line '[X]'. Take a look when you get a chance. No need to call me back."

That's a deliberate redirect to the channel that's actually going to convert, instead of asking for the one that statistically won't.

Frequency has a hard ceiling too

The one universal rule across every source: leave at most one or two voicemails per prospect. A third pushes reply rates on the whole sequence below the reply rate of leaving no voicemail at all - persistence past that point reads as noise, not diligence.

Why the caller details stay separate from the script

Whichever structural camp a script follows, the rep's own name, company, and callback number shouldn't be baked into the AI-generated text. They stay as placeholders, auto-filled at render time, so a new rep or a changed callback number gets instant customization instead of waiting on the script itself to be regenerated. The research and the structure are reusable across the whole team; only three details actually change person to person.