Software dashboard showing a live call transcript, a seller motivation score and the status of an AI voice agent
AIReal EstateVoice Agents

AI Cold Calling: What Is Legal and What Works

AI cold calling needs prior consent under the TCPA. What an AI cold caller can legally do, Vapi, Bland and Retell pricing checked October 2026, and setup.

JM

Jason Macht

Founder @ White Space

May 25, 2026 · Updated October 16, 2026
9 min read

AI cold calling means a voice agent places the sales call and holds the conversation itself, with no human on the line until a lead qualifies. The technology works well now. The legal problem is the word "cold." Since February 2024 the FCC has treated AI-generated voices as "artificial voices" under the TCPA, so an AI can't call someone who never gave you consent, and a sales call needs that consent in writing. That leaves three jobs an AI cold caller can do legally: answer inbound calls, call back people who opted in, and support the humans who do the actual cold dialing. Below is what the rules require, what the main platforms cost (checked on each vendor's own pricing page in October 2026), and how I'd set one up.

Some context on where I'm coming from. I build voice and chat agents for a living, and White Space's own assistant runs on Vapi, with its prompt and knowledge base kept in a git repo rather than edited in a dashboard. We don't sell cold outbound dialing, and the rules below are most of the reason why.

Is AI cold calling legal?

Not in the way most people picture it, which is an AI working through a skip-traced list of homeowners who have never heard of you. Here is the chain of rules that stops it.

The FCC's Declaratory Ruling FCC 24-17, adopted in February 2024, says the TCPA's restrictions on "artificial or prerecorded voice" calls cover AI technologies that generate human voices. A cloned voice counts, and so does an LLM agent speaking through a synthetic voice.

Once a call counts as an artificial voice call, the FCC's rules at 47 CFR 64.1200 apply (I checked the current text in October 2026):

  • A call to a cell phone using an artificial voice needs the called party's prior express consent.
  • If the call is telemarketing, which an offer to buy someone's house is, it needs prior express written consent. The same goes for residential landlines.
  • Every artificial voice message has to state, at the beginning, the name of the business responsible for the call, and give a phone number. Telemarketing messages also need an automated opt-out the person can use by voice or keypress, offered within two seconds of that identification.

Two more rules catch people out. A ringless voicemail is still a call: the FCC said so in its November 2022 ruling, so dropping an AI-voiced voicemail on a cold list carries the same consent requirement as ringing the phone. And the Do Not Call registry and the 8 a.m. to 9 p.m. calling window apply to sales calls whether a person or a model is talking.

Texting has its own version of this. Marketing texts need prior express written consent, and the A2P 10DLC registration every business texter has needed since February 1, 2025 is a carrier requirement, not consent. I covered that in the A2P 10DLC guide. For the rest of the cold calling rules, including what human callers can and can't do, see real estate cold calling costs and rules.

One more rule I hold every agent to, whatever the law in a given state says: it tells people it's an AI in its opening line, and it says yes if someone asks. An agent that pretends to be human usually gets found out, and a seller who feels tricked doesn't sell to you.

This is not legal advice. Talk to a TCPA attorney before you put any automated calling in front of the public.

Three jobs an AI cold caller can do legally

Answering inbound calls

When a seller calls the number on your postcard, bandit sign or PPC ad, they started the call. An AI agent can pick up at 9 p.m. on a Sunday, ask the qualifying questions and book an appointment, and no outbound consent rule is involved. For most investors this is the first agent worth building, and the AI receptionist guide for investors goes deeper on it.

Calling back people who opted in

A seller fills out your "get a cash offer" form at 11 p.m., and the form has a consent checkbox that names your company and says you may call and text them using automated technology and AI voices. Now an AI can call back within a minute, which a human team rarely manages overnight. This is the closest legal thing to AI cold calling, and the consent has to be real: unchecked by default, specific to your company, and stored with a timestamp.

That last part is a database decision as much as a legal one. On our own lead pipeline, every form submission is written to Postgres before anything else happens, and call consent and the two kinds of SMS consent live in their own columns that default to false. A missing checkbox reads as "no." Anything automated that reads from a table built this way should pull only rows where the right consent is true, so a bug upstream fails closed instead of calling someone who never agreed.

Supporting the human dialers

Your cold list still gets called by people on a regular dialer. The AI works after the call: transcripts, summaries, a score for every conversation, and the follow-up for anyone who agreed to hear from you. Our revenue intelligence system scores 100 percent of calls, against the roughly 5 percent a manager reviews by hand, with about a 23 percent average improvement in rep performance and measurable results inside 30 days. That's one client's result, not a typical one. If you're choosing the dialer itself, I compared the options in the AI dialer guide for real estate.

What the AI cold calling platforms cost

These are the three platforms I'd look at for building a seller-facing voice agent. Every number comes from the vendor's own pricing page, checked October 2026.

PlatformPer-minute priceWhat the minute includesLimits to know
Vapi$0.05 hosting, plus models at costHosting only. Transcription, LLM and voice are billed at provider costFree tier: 4 concurrent calls. Core plan $29/month: 10 concurrent. Extra lines $10 each/month
Bland$0.14 (Start) or $0.12 (Build)LLM, speech-to-text and text-to-speech. Telephony billed separatelyStart: 10 concurrent, 100 calls/day. Build: $299/month platform fee, 50 concurrent, 2,000 calls/day
Retell$0.07 to $0.31Depends on the LLM and voice you pick. US Twilio telephony $0.015/min20 concurrent calls free, then $8 each/month. Batch calls +$0.005 per dial

What those mean in practice:

Vapi's own estimator puts 1,000 minutes at $82 to $129 a month using Deepgram, an OpenAI model and ElevenLabs, which works out to roughly $0.08 to $0.13 a minute. Calls over Vapi's own telephony carry no transport fee; Twilio outbound is listed at $0.014 a minute. You get the most control here, and you also own more of the engineering.

Bland bundles the AI into one rate, which makes billing predictable. Watch the call caps: 100 calls a day on the Start plan is fine for a pilot and too low for a callback queue after a big mail drop. In its plan comparison table, warm and live transfers are marked for Enterprise, while the self-serve plans bill transfer minutes, so confirm which kind of transfer your plan gets before you design the handoff around it.

Retell's calculator default is $0.15 a minute: $0.08 for the LLM, $0.055 for its voice infrastructure and $0.015 for text-to-speech. Billing is by the second, and on a voicemail you pay only while the agent is active.

A hypothetical to put that in scale: an investor who calls back 300 opted-in web leads a month, with two attempts each and about three minutes per connected call, is looking at well under 2,000 minutes. That's a few hundred dollars a month on any of the three. The minutes are rarely the expensive part. The build, the testing and the CRM work are.

Air.ai used to show up in most of these comparisons. I left it out. The FTC sued Air AI in August 2025 over allegedly deceptive earnings and refund claims, and in March 2026 the FTC announced a settlement that bans Air AI from marketing business opportunities.

For a deeper look at the three platforms on latency, SDKs and developer experience, see Bland vs Vapi vs Retell. If you are moving off CallPorter, I also compared five CallPorter alternatives.

How I'd set up a consented callback agent

This is the shape of the agent I'd build for an investor, in the order the pieces matter.

  1. Fix the form first. Add the consent language, name your company in it, leave the box unchecked, and store the consent with a timestamp and the page it was captured on. Without this, nothing below is legal to switch on.
  2. Store the lead before you call. Write the submission to your database, then notify the agent and your CRM. If the notification fails, you still have the lead and can retry it.
  3. Call fast, and open honestly. Something like this hypothetical opener:

Agent: Hi, this is an AI assistant calling for [Company]. You filled out our form about the house on Elm Street last night. Do you have two minutes?

  1. Qualify without negotiating. Ask about ownership, condition, occupancy and timeline. The agent never quotes a number. An emotional seller pushing for a ballpark can talk a model into being "helpful," so the prompt forbids it and the tests check it.
  2. Hand off to a person. Book the call with your acquisitions rep or transfer live, and write a structured summary back to the CRM so the rep doesn't re-ask the same questions.
  3. Test it like software. Our assistant has scripted test conversations that run against it: a greeting, a pricing question, an "are you a real person?" check, an off-topic caller and a few others. If a prompt change breaks one, we see it before a seller does.

Opt-outs need the same care. When someone says "stop calling me," the agent should confirm, end the call and put the number on your internal do-not-call list automatically, not in a note someone might read next week.

When you demo any platform, call it from your own cell phone on a weak signal, talk over the agent mid-sentence, ask it whether it's a robot, and ask it what your house is worth. How it handles those four things tells you more than the sales deck. If you'd rather have this built for you, our AI voice agents for real estate cover inbound and consented follow-up.

JM

Jason Macht

Founder & CEO, White Space Solutions

Jason builds AI automation systems for real estate investors and business owners. With experience spanning data analytics, direct mail automation, AI voice agents, and revenue intelligence, he helps companies replace manual workflows with intelligent systems that drive measurable results.

Want to get more out of your business with automation and AI?

Let's talk about how we can streamline your operations and save you time.