AI Outbound Calling: What's Legal, What Works and What It Costs

AI outbound calling means an AI voice agent places the phone call itself — it dials, holds a real two-way conversation, and logs the outcome without a person on the line. For a small business it works well on calls people expect: callbacks to new inquiries, appointment confirmations, recalls, reactivation of past customers and payment reminders. It works badly, and is usually illegal, as a cold-calling machine pointed at a purchased list.

That last sentence is the part most of the vendor pages ranking for this term skip. Since February 2024 the Federal Communications Commission has treated AI-generated voices as "artificial" voices under the Telephone Consumer Protection Act, which means the consent rules written for robocalls apply to your AI agent. This post covers what an outbound program can do, which calls are on the safe side of the line, what it costs, and how to launch one without inviting a lawsuit. For the inbound side of the same technology, start with our complete guide to AI voice agents for business.

One note before the detail: this is an operator's explanation of the rules, written by people who build these systems, not legal advice. If your plan involves calling people who have not given you their number for that purpose, talk to a telecommunications attorney first.

AI outbound calling is a phone call your business initiates where the voice on the line is generated by software. The agent works from a contact list or a trigger in your systems (a new web form, an appointment tomorrow, an invoice 15 days late), places the call, speaks in natural sentences, listens to the answer, and adapts. When the call ends, it writes the result back: confirmed, rescheduled, interested, call back Thursday, opted out.

That makes it different from two older tools it often gets confused with:

• Prerecorded robocalls play the same message to everyone and can at best take a key press. An AI agent holds an actual conversation and can answer questions it was given the facts for.
• Power dialers dial fast for a human rep, who still does the talking. The AI in a power dialer is usually transcription and coaching, not the voice.

There is also a hybrid that matters for compliance. Some "AI sales agent" products never speak on the call at all: they research the prospect, write the opener, dial for a human and summarize the call afterward. Because a person is talking, those calls are not artificial-voice calls. The rules below are about the version where the AI is the one talking.

Mostly no, at least not the way it is usually pitched. On February 8, 2024, the FCC released a declaratory ruling (FCC 24-17) confirming that the TCPA's restrictions on "artificial or prerecorded voice" calls encompass current AI technologies that generate human voices. In the Commission's words, such calls "require the prior express consent of the called party" absent an emergency purpose or an exemption.

The FCC's own rules then set the consent level by call type. Under 47 CFR 64.1200:

• Any non-emergency artificial-voice call to a cell phone needs the called party's prior express consent.
• An artificial-voice call that advertises or telemarkets to a cell phone or a residential line needs prior express written consent: a signed agreement, which can be electronic, that clearly authorizes that business to call that number with those calls.
• Informational calls (a confirmation, a recall reminder, a delivery update) sit at the lower bar of prior express consent, which usually means the person gave you the number in connection with the thing you are calling about.

So the honest answer to "is AI cold calling legal" is that calling strangers on their cell phones with an AI voice to sell them something is exactly what these rules prohibit. Buying a list does not transfer consent. Neither does scraping a number from a website.

The business-to-business escape hatch is narrower than people assume. The cell phone rule applies to the number, not to who owns it, and most small-business owners, contractors and office managers answer their cell phones. If your target list is owners of small companies, a large share of those numbers are wireless. Treat B2B cold calling with an AI voice as a question for counsel, not a loophole.

The reason this matters to a small business is the private right of action. Under 47 U.S.C. 227(b)(3), a person who receives an unlawful call can sue for actual losses or $500 per violation, whichever is greater, and a court can triple that when the violation was willful or knowing. A 2,000-number AI campaign without valid consent is, on paper, a seven-figure exposure. The FCC also framed its ruling as giving state attorneys general a clearer tool against bad actors.

The calls that work, commercially and legally, share one trait: the person on the other end gave you their number and expects to hear from you. Five of them carry most small-business programs.

1. New-lead callbacks. Someone filled out your form or requested a quote, and the agent calls back within a minute or two, while they are still at their desk. Response speed is the single biggest lever in lead conversion — our breakdown of the real cost of slow lead response runs the numbers, and Talos Connect's guide to speed to lead covers the phone-side setup. Make sure your form's consent language actually covers a call about the request.
2. Appointment confirmations and reschedules. The day-before call that confirms, moves the slot in the same conversation, and frees the time for someone else when the answer is no. This is the core of our no-show reduction build.
3. Recalls. Dental hygiene, annual exams, vaccine boosters, HVAC maintenance visits: the customer has an ongoing relationship and the call is a service reminder, not an ad.
4. Reactivation of past customers who opted in. A lapsed patient or client who agreed to hear from you about offers. If the call promotes something, this is the category that needs written consent on file, so check before you load the list.
5. Past-due balance reminders. A courteous reminder with a way to pay. Keep it to the facts of the balance and route disputes to a person.

What these have in common is also why they convert: nobody is surprised by the call. Most owners who try outbound AI expect sales calls to be the payoff, and find that the boring calls — confirmations and callbacks — return the money first.

Consent is the first rule, not the only one. The same section of the FCC's rules sets operating requirements that apply on every call, and your agent should enforce them in software rather than relying on someone remembering.

• Calling hours. No telephone solicitation to a residential subscriber before 8 a.m. or after 9 p.m., measured in the called party's local time, not yours. An agent calling across time zones has to know where each number is.
• Identification up front. An artificial-voice message must state the business responsible for the call at the beginning and give a callback number during or after the message.
• An opt-out on the call. For telemarketing messages, the rules require an automated voice or key-press opt-out offered within two seconds of the identification. Build "stop calling me" recognition into every call type anyway.
• The National Do Not Call Registry. Telephone solicitations cannot go to numbers on the registry, and those registrations are honored indefinitely. Scrub marketing lists against it before each campaign.
• Your own do-not-call list. Opt-out requests must be recorded and honored within a reasonable time, which the rule caps at 10 business days, and the record kept for five years. The agent should write the opt-out to the contact the moment it hears it.

Beyond the letter of the rules, three design choices keep a program out of trouble. The agent should say it is an AI when asked and offer a person, which is how ours are built on the voice AI agents platform. It should stay on the reason for the call, so a confirmation call cannot drift into a sales pitch the customer never consented to. And every call should leave a record — transcript, outcome, consent source — because if a complaint arrives, the record is your defense.

Texting follow-ups bring their own rulebook (carrier registration and separate consent). Talk Is Cheap's guide to A2P 10DLC business texting rules covers that side.

Texas adds its own telephone solicitation law in Chapter 302 of the Business and Commerce Code, often called the state's "mini-TCPA." Sellers it covers must register with the Texas Secretary of State before soliciting. According to the Secretary of State's registration FAQ, registration carries a $200 filing fee and a $10,000 security deposit, the Texas Attorney General enforces it, and it is the seller's responsibility to decide whether the law applies.

Senate Bill 140, effective September 1, 2025, extended the state's telemarketing rules to text messages and widened private lawsuits under them. Following a 2025 settlement, the Secretary of State's FAQ now says a business that sends text messages with the consumer's prior consent does not have to complete the registration. That carve-out is written for consent-based texting; do not assume it covers a cold-calling campaign.

For a Houston, Dallas or San Antonio business, the practical upshot is the same as the federal one: a program built on consented, informational and service calls stays well clear of Chapter 302 questions, and a cold sales campaign needs a lawyer's read before the first dial. Our Texas AI automation page covers how we work with businesses across the state.

Pricing in this market comes in three shapes: a per-minute rate, a per-hour rate for the agent's talk time, or a flat monthly platform fee with a minute allowance. Per-minute pricing looks cheapest on a pricing page and gets expensive when calls run long or retries pile up; a flat fee is easier to budget for a steady reminder program.

Our own prices are published. On Talos Connect, the AI receptionist is $149 a month including 500 minutes, extra minutes are $75 per 500, and AI outbound campaigns — recalls, reactivation, reminders and past-due balances — are an optional $199 a month add-on with consent, opt-out and calling-hour guardrails built in. The full breakdown is on our voice AI agents page.

The way to judge any quote is cost per completed outcome, not cost per minute. As illustrative arithmetic, not a promise: the receptionist plan plus the outbound add-on is $348 a month, so a practice running 600 confirmation calls a month pays about 58 cents a call before any extra minutes. If those calls save even a handful of empty slots, the program has paid for itself. Run the same math on your own volumes before you sign anything.

Be blunt with yourself about the use case. AI outbound is a poor fit when:

• You do not have consent records. If you cannot show where each number came from and what the person agreed to, fix the intake form first. The agent will faithfully call everyone on a bad list.
• The call needs judgment. Delivering bad news, negotiating a dispute, or anything that touches medical, legal or financial advice belongs with a person. We cover where the line sits in the calls owners are most afraid to automate.
• The list is small and high-value. Twenty enterprise prospects deserve a human who has read their website. An AI research assistant helping that human is a better buy than an AI voice.
• Your data is a mess. Reactivation calls that greet a customer by the wrong name, or remind them of an appointment they cancelled, cost more goodwill than they recover. Clean the CRM first; our CRM automation work usually starts there.

The broader pattern of using AI agents across the sales funnel, beyond the phone, is in our post on how AI agents improve lead generation and sales conversion.

Start narrow, measure, then widen. The sequence we use:

1. Pick one informational call type. Confirmations or new-lead callbacks. Both have clear consent, short scripts and an outcome you can count.
2. Audit consent at the source. Read your web forms, intake paperwork and booking flow. Confirm each one tells people you may call them about their request, and log where every number came from.
3. Write the script around outcomes. Identify the business, state the reason, give two or three paths (confirm, reschedule, talk to someone), and offer the opt-out. Keep the agent from improvising offers.
4. Wire the results back. Each outcome should land on the customer record automatically, the way our lead follow-up automation writes every touch to the CRM, so staff never re-enter what the agent learned.
5. Run two weeks, then review transcripts. Read 30 calls. Fix where the agent misunderstood, where people asked for something it could not do, and where the opt-out was missed.
6. Only then add a second call type. Recalls or reactivation next; promotional calls last, and only to numbers with written consent on file.

Most single-system voice installs go live in two to three weeks, and the review in step 5 is what turns a working agent into one your customers do not mind hearing from.

Yes, technically, but legally the bar is high. The FCC treats AI-generated voices as artificial voices under the TCPA, so a sales call to a cell phone or residential line needs the person's prior express written consent. Calls to people who opted in, such as past customers who agreed to offers, can work well. Cold sales calls to purchased lists are the use case the rules were written to stop.

Federal rules require the call to identify your business at the start and give a callback number, and telemarketing calls must offer an opt-out. Separately, the safest and most honest design is for the agent to confirm it is an AI whenever asked and offer a person. Several states have added or proposed their own bot-disclosure rules, so assume disclosure is expected.

An AI SDR is a broader label for software that handles sales development work: researching prospects, writing emails, booking meetings, and sometimes calling. AI outbound calling is specifically the phone part, where an AI voice holds the conversation. Many AI SDR tools never speak on a call and only assist human reps, which keeps those calls outside the artificial-voice rules.

Far more than a human team, since each conversation runs independently and the agent does not get tired at 4 p.m. In practice the limits are your list, your consent records, the 8 a.m. to 9 p.m. calling window in each recipient's time zone, and carrier reputation: dialing too many numbers too fast can get your caller ID flagged as spam.

They can, the same as any high-volume calling. Calls to people who expect to hear from you, from a consistent local number, at reasonable volume, with low hang-up and complaint rates, protect caller reputation. Bursts of cold calls to strangers from fresh numbers are what carrier analytics are built to label. A clean consented list is a deliverability strategy as much as a legal one.

Thinking about outbound calls for your business? Book a free consultation — bring the calls your team makes today and where your contact numbers come from, and we will tell you which ones an AI agent can take over, what it would cost at your volume, and honestly which ones should stay with a person or not be made at all.