# AI cold calling: what's legal and what works

Source: https://dialmilo.com/hub/ai-cold-calling
Updated: 2026-10-23

AI cold calling means two things, and only one is lawful for most consumer lists. The consent rules for AI voices, and where an AI caller really pays off.

"AI cold calling" covers two very different things. **AI-assisted cold calling** means a human rep makes the call, with AI handling research, scripts, dialling and notes. That's an ordinary live call under the normal cold-calling rules. **An AI voice cold-calling consumers** means a synthetic voice dialling people who never agreed to hear from you. In the US, UK and EU that is either outright unlawful or needs a consent that a cold list doesn't have. The FCC confirmed in February 2024 that AI voices count as artificial voices under the TCPA. An AI voice agent pays off on the warm lists most teams under-call: fresh leads, opted-in customers, quotes nobody followed up.

Search "AI cold calling" and most top results describe the same pitch. An AI dials hundreds of prospects at once, handles objections and books meetings while your team sleeps. Some of them also describe AI that only assists a human caller, and not all of them say which one they mean. Few explain that the first version, aimed at consumers, collides with the consent rules in nearly every market worth selling into.

This article separates the two, sets out the rules as of October 2026, and shows where an AI caller is worth deploying.

## Two things called "AI cold calling"

| | AI-assisted cold calling | AI voice cold calling |
| --- | --- | --- |
| Who speaks | A human sales rep | A synthetic voice |
| What the AI does | Researches the prospect, drafts the opener, dials, transcribes, logs to the CRM | Holds the whole conversation |
| Legal category | A live call | An artificial-voice call (US), likely an automated call (UK), automated calling (EU) |
| Consent needed for consumers | Do-not-call screening; consent rules vary by country | Prior consent in most markets; prior express *written* consent for US telemarketing |
| Where it works | Classic outbound sales | Lists that already gave permission |

[Salesforce's own explainer](https://www.salesforce.com/sales/artificial-intelligence/ai-cold-calling/) describes the first kind. It says the technology "doesn't actually dial calls and interact with prospects" and instead supports the rep. Most voice-AI vendors mean the second kind. Both are real products. Only one of them can lawfully call a stranger's mobile.

## The US: why an AI voice can't cold call consumers

The Telephone Consumer Protection Act bars calls to mobile numbers that use "an artificial or prerecorded voice" without the called party's prior express consent. It applies the same rule to [residential lines](https://www.law.cornell.edu/uscode/text/47/227). For years, vendors argued about whether a voice generated live by an AI counted as "artificial".

The FCC settled it. In a Declaratory Ruling adopted on 2 February and released on [8 February 2024](https://docs.fcc.gov/public/attachments/FCC-24-17A1.pdf), the Commission confirmed that the TCPA's restrictions on artificial or prerecorded voice "encompass current AI technologies that generate human voices." Callers using them "must obtain the prior express consent of the called party." The order adds that the TCPA allows no "carve out of technologies that purport to provide the equivalent of a live agent", which closes the door on the argument that a conversational AI is different from a robocall.

For marketing calls the bar is higher. The FCC's rules require prior express *written* consent for artificial-voice telemarketing to mobiles and homes. That means [a signed agreement](https://www.law.cornell.edu/cfr/text/47/64.1200) that clearly authorises the seller to make such calls and says consent is not a condition of purchase. A cold list, by definition, has none of that.

The consequences come in two forms. The statute gives consumers a private right of action of [$500 per call](https://www.law.cornell.edu/uscode/text/47/227), and a court can treble that for a wilful or knowing violation. The FCC ruling also confirms that state attorneys general can pursue these calls.

Two recent developments are worth knowing, though neither changes the bottom line:

- **The one-to-one consent rule never took effect.** The Eleventh Circuit vacated it on 24 January 2025 in *Insurance Marketing Coalition v. FCC*. Consent still has to be [clear and unmistakable](https://www.adamsandreese.com/news-knowledge/fcc-one-to-one-consent-rule-struck-down), but it no longer has to name a single seller.
- **Courts are no longer bound by FCC readings of the TCPA.** In *McLaughlin Chiropractic v. McKesson* (20 June 2025), the Supreme Court held that district courts must [interpret the statute themselves](https://www.faegredrinker.com/en/insights/publications/2025/6/supreme-court-decides-mclaughlin-chiropractic-associates-v-mckesson-corp), giving the agency's view appropriate respect. Some people read this as an opening. In practice the 2024 ruling rests on the plain word "artificial", and an AI-generated voice is hard to argue out of it.

The Telemarketing Sales Rule adds the FTC's layer on top. It covers the National Do Not Call Registry, calling hours of 8am to 9pm in the called person's time, and prompt disclosure of the seller and the sales purpose. It also has a rule that matters for AI pickups: a call counts as abandoned if no representative speaks within [two seconds of the person's greeting](https://www.law.cornell.edu/cfr/text/16/310.4). That rule was written for human agents, but an AI that takes three seconds to say hello is a bad idea on any reading.

### Business numbers in the US

B2B is where AI voice outreach has the most room, and less of it than the pitch decks suggest:

- **Mobiles are protected whoever owns them.** The TCPA's artificial-voice rule for mobile numbers doesn't care whether the person is a consumer. A founder's mobile on a scraped B2B list is covered.
- **Business landlines are outside the residential-line rule**, but every artificial-voice message still has to [state the caller's identity](https://www.law.cornell.edu/cfr/text/47/64.1200) at the start and give a callback number.
- **The TSR mostly exempts calls to businesses**, apart from a few provisions and retail sales of office or cleaning supplies.
- **State law sits on top.** Many states have their own telemarketing statutes, so check the states you dial.

## The UK: automated calls need specific consent

PECR splits marketing calls into two tracks. Live calls can go to anyone not registered with the TPS or Corporate TPS, unless they told you to stop. Automated calls are stricter. The ICO says they may only go to people who have specifically consented to automated calls from you, and that ["general consent for marketing, or even consent for live calls, is not enough."](https://ico.org.uk/for-organisations/direct-marketing-and-privacy-and-electronic-communications/guide-to-pecr/electronic-and-telephone-marketing/telephone-marketing/)

Is an AI voice an automated call? PECR regulation 19 defines an automated calling system by what it can do. It dials [a sequence of numbers](https://www.legislation.gov.uk/uksi/2003/2426/regulation/19) and transmits "sounds which are not live speech". A synthetic voice is not live speech. No UK court has ruled on a conversational AI yet, but the ICO's recent enforcement suggests which way it leans. In September 2025 it fined two energy firms [£550,000 in total](https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2025/09/warning-over-robo-calls-as-energy-firms-fined-half-a-million-pounds-for-unlawful-marketing-calls) for automated marketing calls. Those calls used avatar software that gave people the impression they were talking to "Jo, Helen or Ian", and were made without consent. If scripted voice clips triggered by an agent count, a fully synthetic voice is unlikely to get kinder treatment.

The regulation 19 consent rule also covers corporate subscribers. The B2B relaxation in the UK applies to live calls, not automated ones.

## The EU: consent, plus a disclosure duty

The ePrivacy Directive requires [prior consent for automated calling systems](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:02002L0058-20091219) used for direct marketing. It leaves live calls to each member state, and several have moved to opt-in. The Netherlands is the sharpest example: since 1 July 2026 even existing customers need to have [consented to sales calls](https://www.osborneclarke.com/insights/netherlands-tightens-telemarketing-rules-july).

On top of that, Article 50 of the AI Act has applied since 2 August 2026. AI systems that interact with people must be designed so the people concerned are [informed they are interacting with an AI](https://artificialintelligenceact.eu/article/50/), at the latest at the first interaction, unless it's obvious. An AI that dodges the question "am I talking to a robot?" is now a compliance problem in the EU, not just an ethical one.

The full jurisdiction-by-jurisdiction picture, with sources, is in [AI phone call laws](/hub/ai-phone-call-laws).

This isn't legal advice, and the rules above are a summary as of October 2026. If you plan to put an AI voice on any list you didn't collect yourself with call consent, take advice from a lawyer in each country you'll be dialling.

## Where an AI caller beats a cold list

Take away the cold list and an AI voice agent looks less like a cheaper SDR. It looks more like a way to finally work the permission-based lists your team never gets to. These lists are smaller, and they answer.

**New inbound leads.** Someone fills in a quote form at 9:40pm. A rep calls back the next afternoon, if at all. An AI can call within minutes, inside calling hours, qualify with your questions and book the slot. This is the case covered in [speed to lead](/hub/speed-to-lead). Put the call-consent wording next to the phone field and keep a record of it.

**Quotes and trials that went quiet.** The prospect asked for a price, got it and disappeared. A short call ("did the quote make sense, anything you'd change?") recovers deals that an email reminder doesn't.

**Lapsed customers who opted in to calls.** Win-back calls to customers who agreed to hear from you by phone are where AI calling has the clearest economics. Nobody's time is spent on the calls that go unanswered.

**Operational calls that aren't sales.** Appointment reminders, payment reminders and delivery confirmations. These still need consent for an artificial voice in the US (informational calls need prior express consent rather than written consent), but customers usually gave their number for exactly this.

## What a good AI outbound call does differently from a cold call

A cold caller's job is to survive the first ten seconds. A warm AI call has a different job: be recognisable, honest and quick.

> **Milo:** Hi, this is Milo, an AI assistant calling from Harbour Kitchens. Is that Priya?
>
> **Prospect:** Yes?
>
> **Milo:** You asked for a quote on a kitchen refit on Tuesday. I'm calling to see if you'd like a free measure-up visit. We have Thursday morning or Saturday.
>
> **Prospect:** Oh, right. Saturday's better.
>
> **Milo:** Saturday at ten works. I've booked it, and you'll get a confirmation by text.

That call, an illustrative one, has everything a cold call lacks. It discloses an AI in the first sentence, gives a reason the person recognises, and asks for something small. It's also the call a predictive dialler can't make, because there's no rep free to take it. The comparison is in [predictive dialler vs AI voice agent](/hub/predictive-dialer-vs-ai-voice-agent).

Dial Milo is built for this kind of list. Milo says it's an AI at the start of every call (on by default). It calls each person in their own local time within the calling hours you set. It records "stop calling" as an opt-out that no campaign can override, and it keeps anyone from being called twice in a set number of days across campaigns. Those controls are described in [calling hours and contact limits](/help/calling-hours-and-contact-limits). None of them makes a cold list lawful. They make a consented list behave well.

## The cost question

Vendors sell AI cold calling partly on cost per dial. That number is real. Dial Milo charges the phone line for the exact number dialled plus 9.57¢ per minute of AI talk time, which comes to roughly 14–20¢ a minute all-in for US, UK and Dutch numbers. Calls that don't connect have no minute charge. You can price any destination on the [rates page](/rates).

But cost per dial is the wrong number to optimise on a cold list, because the expensive part comes afterwards. One TCPA claim at $500 per call wipes out the savings on tens of thousands of dials. Optimise cost per useful conversation on a list that's allowed to hear from you. Then the AI's price advantage is pure gain.

## A checklist before any AI voice dials a list

  
    For each list, write down how the number was collected and what the person agreed to. If the answer is "bought" or "scraped", an AI voice shouldn't call it.
  
  
    US marketing to mobiles and homes needs prior express written consent. Informational calls need prior express consent. UK automated calls need specific consent to automated calls from you. Dutch sales calls now need consent even from existing customers.
  
  
    Say the name of the business and that the caller is an AI before anything else. It satisfies the US identification rule and the EU AI Act, and it's what people expect.
  
  
    In the US, revocation of consent by any reasonable means has to be honoured within ten business days. Make it instant, and make it apply across every campaign.
  
  
    Aim for under a second between the person's "hello" and the AI's first words. Long silences read as a robocall and, under the TSR, can count as an abandoned call.
  

Do that and "AI cold calling" stops being a legal risk wrapped in a growth hack. It becomes something more useful: a caller that never forgets a lead, working the people who asked to hear from you. For the wider playbook, start with the [AI outbound calling guide](/hub/ai-outbound-calling).

## Frequently asked

**Is AI cold calling legal?**

It depends on who is speaking and who you call. A human rep using AI tools for research, scripts and notes is making an ordinary live call, under the usual cold-calling rules. An AI voice that speaks to the person is a different matter. In the US, the FCC confirmed in February 2024 that AI voices are artificial voices under the TCPA, so calls to mobiles and homes need prior consent, and marketing calls need prior express written consent. A call to someone who never gave consent is, by definition, not allowed.

**Can an AI call business phone numbers without consent?**

Sometimes, and less often than people assume. In the US the TCPA's artificial-voice consent rules cover mobile numbers whoever owns them, so a sales director's mobile is protected. Business landlines are treated differently, but the caller still has to identify itself and give a callback number. In the UK, consent for automated marketing calls applies to corporate subscribers too. Check each list with counsel before an AI voice dials it.

**Do I have to tell people they are talking to an AI?**

In the EU, yes. Since 2 August 2026 the AI Act requires AI systems that interact with people to make that clear at the first interaction, unless it is obvious. US federal rules require the caller to identify the business at the start of any artificial-voice call, and some state laws go further. Saying it is an AI in the first sentence is the simplest way to meet all of them.

**What is the difference between AI cold calling and a robocall?**

To a regulator, often nothing. A robocall plays a recording, while an AI voice agent listens and answers, but both use an artificial voice. The FCC treats AI voices as artificial voices under the TCPA, so the consent rules that apply to robocalls also apply to AI voice calls.

**What should I use an AI caller for instead of cold lists?**

Use it on the contacts you already have permission to call and are not calling fast enough: new web leads, people who requested a quote, lapsed customers who opted in to calls, unpaid invoices and appointment reminders. These lists are smaller than a bought list, but they answer, convert, and do not put you in front of a regulator.
