AI cold calling: what it does, what the law requires

What is AI cold calling?

AI cold calling is an AI voice agent placing outbound sales calls. The agent speaks with a synthetic voice, listens, and responds in real time. On a typical call it opens with a reason for calling, asks a few qualifying questions, handles the objections it was prepared for, and books a meeting if the prospect is interested. When the call ends, it writes the transcript, a summary and the outcome to the CRM.

It is not the same as a power dialer. A dialer helps a rep place more calls and skips voicemail and dead lines; the rep still talks. In AI cold calling, the agent talks.

When this post was first published in June 2024, voice AI in sales was early: mostly agents answering inbound calls in place of phone menus. Two things have changed since. The tools now run structured outbound calls, and US regulators have said how existing calling law applies to them.

Where AI voice agents help on a sales call

The case for AI calling is the math of cold calling. In Gong's analysis of more than 300 million cold calls (July 2024), the average connect rate was 5.4 percent of dials, and the average rep booked meetings on 4.6 percent of conversations (Gong's set rate, which it measures per conversation, not per dial). At those rates, most of a calling day goes to calls that never reach a person. The median SDR in the Bridge Group's 2025 study made 44 phone touches a day. Volume is the part of the job a machine does best.

The pattern is the same one that runs through sales automation generally: anything with a clear standard and a clear rule can be run by the agent; anything that depends on judgment stays with a person. For calling, that usually means the agent takes the first touch and the qualification, and a rep takes the prospect who wants to talk.

What AI cold calling cannot do yet

• Long, multi-person conversations. Forrester's 2024 buying study counted 13 people in an average purchase decision. An agent can book the first meeting. It cannot run that committee.
• Anything far off script. Agents handle the objections they were prepared for. A prospect with an unusual situation, a complaint or a sharp question can push the call past what the playbook covers.
• Trust on its own. Some prospects hang up on an AI caller once they realize it is one. Disclosure is still the right default (below), so plan for that drop.
• Judgment about the account. Whether this account is worth a custom approach, or whether a "not now" is really a "no", is still a rep's call.

The law: consent and disclosure

This section is general information, not legal advice. Have counsel review any AI calling program for the markets you call into.

• TCPA (US). On February 8, 2024, the FCC ruled that AI-generated voices are "artificial" under the Telephone Consumer Protection Act. The Act's rules for calls that use an artificial or prerecorded voice, including consent, now apply to calls made with AI voices. That matters most for calls to mobile numbers, which many B2B contacts give as their main line.
• The Telemarketing Sales Rule (US). Business-to-business calls are largely exempt from the FTC's rule, but since May 16, 2024, its bans on material misrepresentations and false or misleading statements also cover B2B telemarketing calls. An agent that misstates who it is or what it offers is a compliance problem, not just a brand problem.
• States and other countries. Several US states and many countries set their own calling and consent rules. Check each market on your list.
• Disclosure. Tell the person they are speaking with an AI agent at the start of the call. It builds trust, it sets expectations, and it keeps the call clear of any claim that it pretended to be human.

How to roll out AI cold calling

1. Pick one narrow use case. Re-engaging leads who asked for information and went quiet, confirming meetings, or first touches on a single segment. Not your top accounts.
2. Clear consent and lists first. Decide which numbers the agent may call and on what basis, and suppress anyone who has opted out on any channel.
3. Write the playbook. The opener, two to four qualifying questions, the pitch, the objections you expect and how to answer them, and when to book or hand off. The agent is only as good as this document.
4. Test before it calls a prospect. Run test calls to your own numbers and read the transcripts.
5. Route the hand-off. Interested and complex prospects go to a rep with the transcript, at once.
6. Review flagged calls every week. Low qualification scores, unresolved objections and frustrated callers are where the playbook needs work.
7. Judge it on meetings held. Calls placed is the one number an AI agent can raise without limit. Meetings held, and the pipeline they create, show whether it works.

Where graph8 fits

On graph8, Voice AI agents place outbound calls linked to a campaign and a sequence step for attribution. An outbound playbook sets the opening, two to four qualifying questions, the pitch, objection handling and the close, and with a calendar linked, the agent books the meeting on the call. After every call, the disposition, summary, key points, next steps, sentiment, a 0 to 100 qualification score and buying signals go to the connected CRM. Calls with a low score, an unresolved objection, a transfer triggered by frustration or long silences are flagged for review, and you can run a test call from the agent page before the agent calls a real contact. Reps who dial themselves can call up to four contacts at once, connected to the first person who answers.