AI cold calling: what is allowed and what works

AI cold calling sounds like a shortcut: software dials a list, talks like a person and books meetings while your team sleeps. The legal reality is less forgiving. In the United States, the Federal Communications Commission has confirmed that calls made with AI-generated voices are "artificial" under the Telephone Consumer Protection Act, which pulls them into the same consent rules that govern prerecorded robocalls.

That one ruling explains why so many AI calling projects stall at legal review. This guide sets out what the main markets allow, where AI genuinely helps a sales team without breaking those rules, and how to decide whether an AI caller belongs in your outbound process at all. It is general information, not legal advice, so check your own position with a qualified adviser before you dial.

Quick answer

AI cold calling is lawful only where the rules for the call type are met. In the US, an AI voice counts as an artificial voice and needs prior express consent. In the UK, Do Not Call registers still apply, and the EU requires people to be told they are speaking to an AI. Most firms get safer, better results using AI to support human callers.

What counts as AI cold calling

The phrase covers several different things, and the law treats them differently. A fully automated voice agent that dials a prospect and holds a conversation is one end of the spectrum. A human caller who uses AI to research the prospect, draft a script and log the outcome is the other. Most legal risk sits at the first end, and most practical value sits at the second.

It helps to separate four activities:

The rest of this guide focuses on the first category, because that is where the regulators have been most active. If your real goal is to answer inbound enquiries quickly, our guide to an AI voice agent covers that case in depth.

The United States: TCPA, FCC and FTC rules

The Telephone Consumer Protection Act restricts calls that use an artificial or prerecorded voice. A Cooley briefing on the FCC ruling quotes the core prohibition as using an artificial or prerecorded voice to deliver a message without the prior express consent of the called party. The FCC's position is that AI-generated voices fall inside that wording, so the ruling clarifies existing law rather than creating a new rule, and it applies immediately.

In practice this means a US business cannot treat an AI voice agent as a clever workaround for the consent rules. If a prospect has not agreed to receive that kind of call, an AI voice making it carries the same exposure as a recorded message would.

The Federal Trade Commission regulates telemarketing separately through the Telemarketing Sales Rule. When it updated that rule in 2024, the Federal Register notice addressed technologies that mimic or clone a specific person's voice, and said calls using them would be subject to the relevant provision to the extent the cloning creates a prerecorded message used in telemarketing. The FTC also maintains the National Do Not Call Registry, and its compliance guidance for the Telemarketing Sales Rule sets out the obligations for sellers and telemarketers.

The practical US checklist is short. Establish consent before any automated or artificial voice call, screen against the National Do Not Call Registry, respect calling hours, keep records of how consent was obtained, and never clone a real person's voice without their agreement. State laws add further requirements, so a national campaign needs a state-by-state review.

The United Kingdom: PECR, TPS and the ICO

In the UK, marketing calls are governed by the Privacy and Electronic Communications Regulations, enforced by the Information Commissioner's Office, with Ofcom handling related nuisance call issues. The ICO guidance on live marketing calls says you must check numbers against the Telephone Preference Service and the Corporate Telephone Preference Service and cannot call any number that appears on them, unless the person has specifically said they want your calls.

Two details catch businesses out. First, registration takes time to become active: the ICO says it can take 28 days, and Ofcom's guidance on live telesales calls says the same, so a list screened a month ago may already be out of date. Second, business numbers are not a free-for-all. The ICO states that both individual and corporate subscribers can object to live marketing calls, and the Corporate TPS exists for exactly that reason.

Enforcement is real and recent. In May 2026 the ICO announced a £160,000 fine against a Glasgow-based energy company for making over 700,000 unsolicited marketing calls. The case involved live calls, but it shows how regulators think about volume. An AI caller can reach far more people than a human team, which multiplies the exposure if the list is not screened properly.

Ofcom notes that the law treats live calls and automated calls that play a recorded message differently. An AI voice that talks in real time sits between the two, so take advice rather than assuming it counts as a live call. If you operate across the UK and EU, our note on GDPR and custom AI agents explains how data protection rules sit alongside the call rules.

The EU: AI Act transparency and local call rules

The European Union adds a layer that the US and UK rules do not yet match: a duty to tell people they are talking to a machine. Article 50 of the AI Act requires providers to design AI systems that interact directly with people so that those people are informed they are interacting with an AI system. The European Commission's AI Act Service Desk publishes the consolidated text, including the exception where it is obvious to a reasonably well-informed person that they are dealing with AI.

For a cold call, that exception is unlikely to help. A prospect who picks up an unexpected call and hears a natural voice has no reason to assume it is synthetic, so the safe reading is that disclosure is needed. The same article says the information must be given at the latest at the time of the first interaction. A law firm briefing from Addleshaw Goddard on the Commission's transparency guidelines reports that these transparency obligations became applicable on 2 August 2026.

The AI Act does not replace national telemarketing and consent law. Each member state has its own telemarketing and consent rules, so check the country you are calling. Treat the AI Act disclosure as an additional requirement on top of local consent rules, not a substitute.

Australia and Canada: registers, hours and consent

Outside the US, UK and EU, the pattern is similar: registers, calling hours and consent. In Australia, the Do Not Call Register keeps a registered number on the list indefinitely unless the owner removes it, and telemarketers have 30 days to recognise a new registration. The industry guide sets telemarketing hours of 9 am to 8 pm on weekdays and 9 am to 5 pm on Saturdays, with no calls on Sundays or national public holidays. That guide dates from 2018, so check the current standard before relying on the hours.

Canada's rules are explicit about automated dialling. The CRTC's key unsolicited telecommunications rules say telemarketers may make solicitation calls using an automatic dialling-announcing device only if the consumer has given express consent beforehand. Telemarketers must also subscribe to the National Do Not Call List, and calls are limited to 9:00 a.m. to 9:30 p.m. on weekdays and 10:00 a.m. to 6:00 p.m. on weekends.

Across all five jurisdictions, three themes repeat. Registers must be checked before dialling. Prior consent matters far more for automated and artificial voices than for human callers. And regulators in several markets are moving towards telling people when they are speaking to AI.

Where AI helps cold outreach without the legal risk

The strongest use of AI in outbound is rarely the voice itself. It is everything around the call, where a human still decides who to contact and speaks to them. That keeps you on the safer side of the rules while removing the slow, repetitive work that stops salespeople spending time on conversations.

Research and list quality

AI can build a profile of each account from public information, flag likely triggers such as a hiring spike or a funding round, and suggest why now might be a good time to call. It can also help clean a list by matching it against your suppression records, which is where many compliance failures begin.

Call preparation and follow-up

Before a call, an AI assistant can summarise past emails, open tickets and previous conversations in a single page. After the call, it can write the notes, draft the follow-up email and update the CRM. This is where a human caller often loses the most time, and it carries none of the consent risk that an automated voice does.

Warm callbacks and booking

Once a prospect has replied to an email, filled in a form or asked to be called, the legal position changes because there is a relationship and a request. An AI agent can then confirm a time, offer slots and send reminders. Our guide to an AI appointment booking agent covers how that works in practice.

Qualification before a human takes over

Some teams use AI to handle the first qualifying exchange on inbound enquiries, then pass a summary to a salesperson. Because the person initiated contact, consent is far easier to evidence.

How to decide whether to automate calls at all

Before building anything, work through a short set of questions. They are the same questions a regulator or a complaint would raise, so answering them early saves rework later.

A useful test is to imagine the call being played back to a regulator. If the person did not expect it, was not told it was an AI, and could not easily stop it, the design has a problem however good the technology is.

There is also a commercial argument. A synthetic voice that gets caught out can damage a brand more than a slower human process would, and trust is hard to rebuild once a prospect decides your company is a source of spam.

How Klevere approaches AI in outbound sales

Klevere builds custom AI agents for businesses, with 500+ AI agents deployed across 50+ projects and 12 industries. When a client asks for AI cold calling, we start by asking what outcome they want, because the answer is usually more meetings, faster response or a cleaner pipeline rather than the call itself.

Our default is to design the agent around a human salesperson. The agent researches accounts, prepares the call, handles follow-up and keeps the CRM current, while a person makes the decision to call and holds the conversation. Where a client wants a voice agent, we focus on inbound and warm callback scenarios, with a clear spoken disclosure and a simple route to a human.

We also build the compliance controls into the workflow instead of leaving them to memory: suppression list checks before any contact, consent fields that travel with the record, calling-hour rules by country and a log of every interaction. Our AI sales agent software is built on that principle. We do not give legal advice, so we recommend that clients confirm their consent basis with their own counsel.

Common mistakes to avoid

The same errors appear in most failed projects. Buying a list and assuming it is cleared for automated calls is the most frequent, followed by testing an AI voice on live prospects before the consent question has been answered. Others include ignoring time zones, forgetting that registers update, and cloning a real employee's voice without a clear written agreement.

A quieter mistake is measuring the wrong thing. Counting calls made rewards volume, which is exactly what regulators penalise. Counting meetings held, complaints received and opt-outs honoured gives a truer view of whether the programme is working.

Frequently asked questions

Is AI cold calling legal?

It depends on the country and the type of call. In the US, the FCC treats AI-generated voices as artificial voices, so prior express consent is needed. Elsewhere, registers and consent rules still apply. A human caller supported by AI is generally lower risk than a fully automated voice calling people who never asked to hear from you.

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

In the EU, Article 50 of the AI Act requires people to be informed that they are interacting with an AI system, unless it is obvious. For a cold call it is rarely obvious. Even where no law requires it yet, disclosure is sensible practice and reduces complaints, so we recommend it in every market.

Can I use AI to call business numbers?

Business numbers are not exempt everywhere. In the UK, the ICO says corporate subscribers can object to live marketing calls and the Corporate TPS exists for them. In Canada, the National Do Not Call List rules apply to telemarketing generally. Screen business lists against the relevant registers and keep records of consent.

What is the difference between AI cold calling and an AI voice agent?

Cold calling means contacting people who have not asked to hear from you. An AI voice agent is the technology, and it can be used for inbound calls, callbacks to people who made an enquiry, or outbound outreach. The legal risk comes mainly from the first use, not from the technology itself.

What are safer alternatives to AI cold calling?

Use AI to research accounts, personalise email, prepare human callers and follow up on warm leads. Respond quickly to inbound enquiries with a voice or messaging agent. These approaches keep a person in charge of outreach and rely on consent that you can evidence if challenged.

Who is responsible if an AI caller breaks the rules?

The business that makes the calls is normally the one regulators pursue, not the software supplier. The ICO's fine described above was issued to the company that made the calls. Contracts with vendors can allocate costs, but you should assume your own compliance duties stay with you, and confirm the consent basis before launch.

If you are weighing up an AI caller and want a clear view of what is safe, useful and worth building, book a free AI audit. We will map your outbound process, point out the compliance questions to settle first and show where AI saves your team the most time.