Cold calling voicemail scripts that get callbacks
A voicemail is a call, and one FCC ruling decides which rules apply. The scripts, the ring setting that decides whether you reach it, and the regulated line.
Most cold calls reach a mailbox, so the voicemail is not a consolation prize — it is the message, and it is the only artefact of the attempt the person will ever encounter. Two things decide whether it works before the script does. Whether you reach voicemail at all is a ring-duration setting rather than luck. And how the message is delivered decides which rules govern it, because the Federal Communications Commission has ruled that a voicemail dropped straight into a mailbox is a call requiring consent.
A voicemail is a call, and one federal ruling decides which rules apply
There are two mechanically different ways to leave a voicemail, and they are not legally equivalent.
A voicemail left after a live dial is part of the call that produced it. The dial rang the handset, the carrier routed to voicemail, the rep spoke. Whatever rules governed the call govern the message.
A ringless voicemail is delivered to the mailbox without ringing the handset at all, usually by inserting the message through a carrier-side interface. Vendors sold this for years on the theory that a message which never rings is not a call. The FCC disagreed. In a Declaratory Ruling and Order adopted on 14 November 2022 and released on 21 November 2022, in CG Docket No. 02-278, the Commission denied a petition for declaratory ruling filed by All About the Message, LLC and found that ringless voicemail to a wireless number is a call made using an artificial or prerecorded voice, and therefore requires the prior express consent the Telephone Consumer Protection Act demands. The Commission stated that while the ruling arises from that petition, it applies to any entity providing ringless voicemail using the end user's mobile telephone number to direct the message to a mailbox associated with that phone.
The practical reading for a B2B sales team: if you are cold-calling and leaving messages the ordinary way, this ruling is not about you. If someone is selling you a way to deposit thousands of voicemails without dialling, it is exactly about you, and the fact that the handset never rang is not the defence it was sold as.
Calling hours are the other constraint and they are local to the recipient, not to you. The FCC's rule at 47 CFR 64.1200(c)(1) bars telephone solicitations to a residential telephone subscriber before 8 a.m. or after 9 p.m. in the local time at the called party's location. The detail of the consent and registry regime around that sits in what TCPA compliance actually requires.
Whether you reach voicemail at all is a setting, not luck
Before any script matters, the call has to survive long enough to be diverted. That is governed by how long you let it ring, and it is a number somebody configured.
Autocloz exposes it as calling_ring_seconds — labelled "Outbound ring duration before voicemail" — defaulting to 25 seconds, adjustable between 5 and 60. Set it too short and you hang up before the carrier's voicemail diversion fires, so you get neither a conversation nor a message and the attempt is pure waste. Set it too long and you burn dial time on a queue that could be making the next attempt.
The setting lives with the rest of the calling policy on the voice platform settings, next to the calling window and the re-dial cooldown, which is where it belongs because all three decide the same thing: how many attempts a queue produces in a day.
Twenty-five seconds is a reasonable default because most mobile carriers divert somewhere in the twenty-to-thirty-second range, but the number worth checking is your own: place a test call to a phone you control, let it ring, and time the diversion. Then set the value slightly above it.
Note one legal floor that intersects here. The Telemarketing Sales Rule's abandoned-call safe harbour at 16 CFR 310.4 requires allowing the telephone to ring for at least fifteen seconds or four rings before disconnecting an unanswered call. The 5-second end of the configurable range sits below that floor, so where the rule applies to your calls, the low end of the range is not a legitimate setting.
The one script federal regulation writes for you
Most voicemail advice is opinion. This part is not.
The Telemarketing Sales Rule's safe harbour for abandoned calls, at 16 CFR 310.4, protects a seller or telemarketer from liability if it employs technology ensuring abandonment of no more than three percent of all calls answered by a person, measured over the duration of a single calling campaign if less than 30 days, or separately over each successive 30-day period. A call is abandoned when a person answers and the telemarketer does not connect the call to a sales representative within two seconds of that person's completed greeting. And when no representative is available within those two seconds, the safe harbour requires that the seller or telemarketer promptly plays a recorded message stating the name and telephone number of the seller on whose behalf the call was placed.
That last requirement is a script mandated by regulation, and it is worth reading twice because it is narrower than people assume. It is not a voicemail script — it is the message a live person hears when your dialler answered before your rep did. If you run any dial mode that can produce that situation, the recording has to exist and has to contain both the name and the number.
Autocloz carries the abandonment figure as a campaign field, abandon_rate_cap, defaulting to 0.030 — three percent, matching the safe harbour rather than an arbitrary internal target. Setting it higher is a decision to leave the safe harbour, and it should be made deliberately if at all.
What a voicemail can do, and what it structurally cannot
Being clear about the job stops you writing a message that tries to do a different one.
A voicemail can establish that a specific named person from a specific named company called for a specific reason, and it can make a later email recognisable. It can create the small obligation of an unanswered message. It can also make the second attempt land as a follow-up rather than as a stranger.
A voicemail cannot hold a conversation, handle an objection, qualify anything, or be skimmed. That last one is the constraint people underrate. Email is scannable; a voicemail is strictly linear and played at real time. A listener cannot jump to the interesting part, so anything before the interesting part is a tax they pay in full before deciding to delete.
Two consequences follow, and they set the shape of every script below. The reason for the call goes first, before your company name, because the reason is what buys the next five seconds. And the number goes at the end and again nowhere else, because a number in the middle interrupts the only pass the listener gets.
Three scripts, by position in the sequence, not by tone
Scripts organised by where they sit, because the same words are right on the first attempt and wrong on the third. The bracketed fields are yours to fill; the length is the constraint.
First attempt — the reason. Around fifteen seconds.
"[Name], it's [your name] at [company]. I'm calling because [specific observable thing about their situation — a hire, a launch, a stack change, a stated goal]. We work with [two named comparable companies or a precise category] on [the narrow problem that follows from it], and I wanted to ask you one question about how you're handling it. I'll send a two-line email so you have this in writing. If it's easier to call back, I'm on [number]."
What makes it work is the second sentence. "[Specific observable thing]" has to be something you could not have said to the next person on the list. If it would survive a find-and-replace of the company name, it is not a reason, it is a pretext, and the listener will hear the difference.
Second attempt — the referenced follow-up. Around twelve seconds, two to four days later.
"[Name], [your name] at [company] again — I left you a message on [day] about [the same specific thing, named again]. Since then [one new fact: something published, something you learned, a result from a comparable customer]. Same question, and I'll leave it at this one. [Number]."
The new fact is the entire point. A second voicemail that repeats the first is evidence you are working a list, and it converts a neutral non-response into a negative impression.
Third and final — the close-out. Around ten seconds.
"[Name], [your name] at [company]. I've left you a couple of messages about [the thing] and I don't want to keep filling your voicemail. I'll stop here. If [the problem] becomes live, my number is [number] and the email's in your inbox. Good luck with [the specific thing]."
Stopping explicitly does two things. It is the only version of persistence that does not damage the brand, and it occasionally produces the callback, because a stated ending is more interesting than a fourth attempt. The copy craft for the live conversation that follows is a different discipline, covered in cold calling scripts that book meetings.
The number: say it the way a person writes it down
A small mechanical section, because this is the single most common failure in an otherwise good voicemail.
Say the number twice, at the end, at dictation pace, grouped the way it would be written in your country's convention. Say it slower than feels natural — the listener is holding a phone in one hand and often writing with the other, and the pause between groups is what lets them catch up. Do not say it at the start as well; a listener who has not yet decided to call back is not writing anything down.
Then make the number worth writing down. If the caller ID they see is a rotation pool rather than a number that reaches you, the digits in the voicemail are the only route back, and they had better ring a phone that a human answers. Autocloz's default caller-ID strategy is country_match, which selects an organisation-owned number in the lead's own country; the alternatives are agent_did, which sticks to the agent's own assigned number, and campaign_did, which uses the campaign's configured number. If callbacks matter more than pickup rate for your motion, the agent's own number is the strategy that makes the voicemail's closing line true.
Pairing the voicemail with the email that follows it
The voicemail-plus-email pairing is standard advice and it is usually implemented in a way that defeats itself. The failure is that the email does not reference the voicemail specifically enough to be recognised, so the recipient receives two unrelated cold touches on the same day instead of one coherent approach from one person.
Three rules that make the pairing work:
- The email arrives the same day, within a couple of hours. Longer and the voicemail is no longer in working memory, which is the only thing the reference is trading on.
- The email's first line names the thing the voicemail named, not the fact that a voicemail happened. "I left you a voicemail" is not a reference; "I called about the two SDR roles you posted last week" is.
- The email is shorter than the voicemail. It exists to be the written record of a spoken message, not to restate the case at length.
Wiring that as a sequence rather than as a rep's habit is the part software should do. In Autocloz a call step and an email step sit in the same sequence with a wait between them, and both pass the same pre-send checks — the do-not-contact list, the campaign's sending window, and the per-channel daily cap. The call channel covers the dialler side; the general shape of stacking channels in one cadence is in multichannel against single-channel outreach.
One honest note about how the follow-up email threads. Autocloz's sequence mail sets a Message-ID on each send but does not set In-Reply-To or References headers, so follow-ups group by subject line in most clients rather than threading by the RFC mechanism. Keep the subject stable across the pair if you want them to sit together.
Measuring callbacks when the callback arrives on a different channel
The measurement problem with voicemail is that its success rarely shows up as a returned call. It shows up as an email reply two days later, a LinkedIn acceptance, or a pickup on the third attempt from someone who now recognises the name. If you count returned calls only, you will conclude voicemail does nothing.
Three measurements that survive:
- Contact rate by attempt number, not blended. If attempt three connects at a materially higher rate than attempt one on the same list, the voicemails are doing work even if nobody dialled back.
- Inbound of any kind within 72 hours of a voicemail, across every channel, attributed to the attempt rather than to the channel it arrived on.
- Answering-machine verdicts as their own bucket, kept out of the answered-conversation count.
That third one is worth dwelling on because it is a specific bug class. Autocloz's answering-machine detection now writes the machine verdict onto the call record itself rather than only onto the dialler queue row, because reports read the call record — and a call record with no disposition is treated as answered, which meant a call the system had positively identified as an answering machine was being reported as a live human conversation. If your own reporting has a similar seam between the dialler and the call history, your answer rate is overstated by exactly your voicemail rate.
The verdict vocabulary is worth knowing too. On Telnyx, answering-machine detection is requested with an answering_machine_detection parameter and reported on a call.machine.detection.ended event carrying a result of human, machine, fax or not_sure. Autocloz normalises anything ambiguous to unknown and takes no automatic action on it, deliberately, so the rep decides rather than the detector guessing. When the verdict is machine, the campaign's configured action applies: skip, retry later after the cooldown window, or the voicemail-drop branch.
Autocloz's free plan includes the call channel at 100 dials a day alongside 5 users and 10 mailboxes, with the do-not-contact list and the calling window enforced before every attempt — start free and pair the voicemail with its email in one sequence.
What a voicemail script cannot fix, and what Autocloz does not do
The limits, plainly.
A script cannot fix a wrong number, and number accuracy is the largest single input to whether any of this happens at all. It cannot fix a caller ID that arrives labelled as suspected spam, because the labelling decision is made before the phone rings. And it cannot fix an offer the listener has no reason to care about — a well-delivered irrelevant message is still irrelevant, and the fifteen-second constraint means there is nowhere to hide a weak reason.
The frequency limits are real and worth configuring. Autocloz defaults to a 24-hour re-dial cooldown allowing a single attempt to the same number per window, a per-campaign maximum of three attempts, and automatic addition to the do-not-contact list after seven consecutive failed attempts. Those apply to the direct dialler as well as to campaign modes.
And Autocloz specifically, because this one matters if you are choosing a tool for voicemail. The voicemail-drop feature is a management surface, not an automatic delivery mechanism. You can register audio files, bind a drop to an organisation, agent or campaign scope, and select voicemail_drop as the answering-machine action — and that action records the outcome and stamps the call's disposition, but no carrier integration in the product currently plays the stored audio into the mailbox. In practice the rep leaves the voicemail. Given the FCC's ruling above, an automated drop is also the branch with the most compliance exposure, so the honest position is that the scripts in this post are meant to be spoken. Beyond that: outbound answering-machine detection depends on the carrier, and only Telnyx among the supported providers offers it — FreJun's published REST API documents no answering-machine parameter, so rather than send an invented one that would bill for detection you never receive, Autocloz sends none and the rep hears the voicemail directly. Voice runs through Telnyx, FreJun or DIDLogic, so call quality and number reputation are ultimately properties of the carrier you connect rather than of the software, and the Telnyx integration is where that connection is configured. None of this is legal advice; the gate enforces the rules you configure, and knowing which rules apply to a given recipient stays with you.
Frequently asked
Is leaving a voicemail regulated the same way as a phone call?
A voicemail left after a live dial is part of that call. A voicemail delivered directly to a mailbox without ringing the handset is treated separately, and the Federal Communications Commission addressed it in a Declaratory Ruling adopted on 14 November 2022 and released on 21 November 2022 in CG Docket No. 02-278. The Commission denied a petition from All About the Message, LLC and found that ringless voicemail to a wireless number is a call made using an artificial or prerecorded voice, so prior express consent is required.
How long should a cold call voicemail be?
Short enough that the listener reaches the end before deciding to delete it, which in practice means roughly fifteen to twenty seconds. The constraint is not a rule anywhere. It comes from the structure of the message: a reason, a claim, a number and a next step is about forty-five spoken words, and anything past that is usually the caller reassuring themselves rather than informing the listener.
Does federal regulation dictate anything you must say in a recorded message?
Yes, in one specific case. The Federal Trade Commission's Telemarketing Sales Rule safe harbour for abandoned calls, at 16 CFR 310.4, requires that when a sales representative is not available within two seconds of the person's completed greeting, the seller or telemarketer promptly plays a recorded message stating the name and telephone number of the seller on whose behalf the call was placed. That is a script written into federal regulation rather than a best practice.
Should you leave a voicemail on the first attempt or wait?
Leave one on the first attempt. A voicemail is the only artefact of the call the person will ever see, so an attempt without one is an unrecognised missed call from an unknown number and buys nothing except a data point. The sequencing decision that actually matters is what follows it: an email sent the same day that references the specific thing the voicemail said, so the two touches read as one person rather than two systems.
What is answering machine detection and should it be switched on?
Answering machine detection is a carrier-side feature that classifies whether a human or a machine answered, so the dialler can branch. On Telnyx it is requested with an answering_machine_detection parameter and reported on a call.machine.detection.ended event with a result of human, machine, fax or not_sure. It is worth enabling because it stops answering machines being counted as answered conversations, but it is not reliable enough to auto-disposition an ambiguous verdict, which is why an unclear result should leave the decision to the rep.
How many times should you call the same person before stopping?
That is a policy decision you should write down rather than a number to copy, and it should be enforced by the system rather than by memory. Autocloz's defaults are a 24-hour cooldown window allowing one attempt per number per window, a per-campaign maximum of three attempts, and automatic addition to the do-not-contact list after seven consecutive failed attempts. Adjust those to your market, but pick numbers and enforce them.