AI dialers and AI cold calling: what they do, and what the law allows
Two different products are sold as AI dialers: software that helps reps dial faster, and voice agents that do the talking. Under US law, calls in an AI voice need consent, so the second kind is for leads who asked to hear from you.
In this guide
Search for “AI dialer” and you’ll find two kinds of product under one name. The first helps human reps make more calls by dialing several lines at once, skipping voicemail and writing the notes. The second is a voice agent that places the call and holds the conversation itself. They solve different problems for different teams, and the law treats them differently.
Two kinds of AI dialer
| AI-assisted dialer | AI voice agent | |
|---|---|---|
| Who talks | Your rep | The AI agent, with a person available on transfer |
| What the AI does | Dials several numbers at once, spots voicemail and phone menus, picks the likeliest number, transcribes and logs the call | Places the call, holds the conversation, completes the task and logs the result |
| Capacity | One conversation per rep at a time | As many conversations as your rules allow |
| Typical buyer | B2B sales teams with SDRs | Contact centers and consumer brands following up with leads and customers |
| Consent rules | The rules for human-placed calls, plus any for autodialed calls | The TCPA’s artificial-voice rules, because the voice is AI-generated |
AI-assisted dialers
These descend from the predictive dialer. The well-known ones dial several numbers in parallel for each rep and connect whichever call a person picks up first. The AI separates people from voicemail and phone trees, guesses which of a prospect’s numbers will connect, and summarizes the call into the CRM. The rep still has every conversation, so the team can only talk to as many people as it has reps.
AI voice agents
Here the AI is the one on the call. It dials, works out who answered, talks with them, answers their questions and gets to an outcome: a booked appointment, a finished application, an order, a time to call back, or a warm transfer to a rep. Every answered call becomes a conversation, whether ten reps are on shift or none.
Is AI cold calling legal?
For calls to US consumers, mostly no. This isn’t legal advice, but the outline is clear. In February 2024 the FCC ruled unanimously (FCC 24-17) that AI-generated voices fall under the Telephone Consumer Protection Act’s rules for artificial and prerecorded voices. An artificial-voice call to a mobile phone or a home line needs the called person’s prior express consent unless an exemption applies, and a telemarketing call needs prior express written consent.
A cold call, by definition, goes to someone who never gave you permission. An AI agent cold calling consumers is therefore making the kind of call the TCPA restricts. A voicemail left in an AI voice is an artificial-voice message too, so the same consent applies to it.
Calls to business landlines are treated differently under some of these rules, which is why most AI cold calling tools are sold to B2B teams. Many business contacts answer on their mobile phones, though, and the mobile rules apply to them. Do Not Call registries, calling hours and state telemarketing laws sit on top. If you plan AI calls to anyone who hasn’t asked to hear from you, take the plan to counsel first.
An AI-assisted dialer with a person on every call falls under different rules, since the voice is human. Parallel dialing still creates the “hello? … hello?” pause while the system finds a free rep, and that pause is what gets numbers marked as spam.
Why consumer brands should call warm leads instead
The calls a consumer brand most needs to make are warm ones anyway. Someone asked for a quote, filled in a form, left a cart, started an application, booked a visit or bought from you before. With consent captured on the form, an AI agent can call every one of them.
Those calls start ahead. The person expects to hear from you, so “you asked about…” keeps them on the line where a cold opener has a few seconds to earn attention. The lead record already says which product, which ZIP code and when they asked, so the agent can skip discovery and start where the form left off.
Warm leads also cool quickly. In the best-known study, the odds of qualifying a web lead fell sharply between a call at five minutes and one at thirty, and our speed-to-lead post walks through it. An agent never has to finish another call before it dials. Old leads count too. Someone who asked for a quote last spring and never booked is cheaper to win than a stranger, which is the case for lead reactivation.
AI makes cold calling cheap, and cheap unwanted calls are the reason people stopped answering unknown numbers. Spend the capacity on people who asked.
What to check before you buy
For warm outbound to consumers, an AI dialer should handle all of the following without custom work:
- Opt-outs per channel that take effect at once, whether the person says “don’t call me” on a call or texts STOP.
- Calling hours in each contact’s own time zone, worked out from the best data you have, with a safe fallback when the zone is uncertain.
- Contact limits per day and per month that count every campaign, so two overlapping programs can’t call one person five times.
- A goal that ends outreach the moment someone books or buys.
- A local caller ID, and the same number every time so callbacks land in the right place.
- An approved voicemail, and a text when the call doesn’t connect.
- Callbacks at the time the person asked for.
- A warm handoff into your contact center, with the conversation on the rep’s screen.
- Every outcome written to your CRM, and every call recorded, transcribed and scored.
Which kind Simple is
Simple is the second kind, where the agent holds the conversation. With Book, it calls new leads back within minutes and follows up over days by call and text through Journeys. Journeys checks calling hours, consent and contact limits before every call and text, dials from local numbers, and stops when the goal is met. In Guided mode it picks the next allowed step for each contact, for example a text for someone who replies to texts but misses calls.
It can place calls itself or take the connected calls from the dialer you already run. When a person should take over, it hands off warm to your team in Five9, NICE, Genesys, Talkdesk or Amazon Connect.