
AI Cold Calling (2026): Legal Rules, Use Cases & Costs
Is AI cold calling legal? What the FCC's AI voice ruling means, where AI calling works (and where it doesn't), real per-minute costs, and how to set it up.
AI cold calling means using an AI voice agent, or AI tools that support a human rep, to make outbound sales calls. In 2026 an AI agent can hold a natural phone conversation, qualify a lead and book a meeting. But there's a catch most vendor pages skip: in the US, calls made with an AI-generated voice count as "artificial voice" calls under the Telephone Consumer Protection Act (TCPA), so they need the person's prior express consent, and prior express written consent for telemarketing. Using an AI-generated voice to make unsolicited telemarketing calls to consumers without the required consent can violate the TCPA. The exact requirements depend on the number called, the purpose of the call and the federal and state rules that apply.
That doesn't make AI calling useless for sales. It changes where it works: calling leads who asked to be contacted, following up within minutes, reminding, re-engaging customers who opted in, and helping human reps call better. This guide explains the rules, the use cases that work, what it costs, and how to set it up.
This guide is published by Convocore, a platform for building AI voice and chat agents. It's general information, not legal advice. Calling rules depend on the country, state and type of call, so check with a lawyer before you launch an outbound campaign.
- AI voices are "artificial voices" under the TCPA. The FCC confirmed this in February 2024. It didn't ban AI calling: it applied the existing artificial-voice rules. Covered calls need prior express consent, and telemarketing calls need prior express written consent.
- What works: calling inbound leads within minutes, appointment reminders and confirmations, re-engaging customers who opted in, and AI help for human reps (research, notes, role-play practice).
- What doesn't: an AI voice dialing purchased lists of consumers without the required consent.
- Cost: published rates run from about $0.045 to $0.31 a minute, but they don't all cover the same things (some include telephony, some are hosting only).
- Set-up: consent capture first, then the agent, then a small test campaign.
AI calling platforms at a glance
| Platform | Type | Best for | Pricing model |
|---|---|---|---|
| Convocore | No-code voice and chat agent platform, with white label | Businesses and agencies that want phone, WhatsApp, Instagram and web chat agents in one place | Monthly plan plus per-minute usage from credits |
| Retell AI | Voice agent platform | Teams that want to pick their own voice, LLM and telephony stack | Per minute, varies by stack |
| Vapi | Developer voice API | Developers building custom voice apps | Hosting fee per minute plus your own providers |
| Bland AI | Voice agent platform | High-volume calling teams | Plan tiers with per-minute rates |
| Autocalls | Outbound calling tool | Small teams that want a simple campaign tool | Monthly plan with included minutes |
| Synthflow | Voice agent platform | Larger teams | Quote-based (enterprise) |
Prices and details are in the cost section below.
Is AI cold calling legal?
United States
On February 2, 2024, the Federal Communications Commission unanimously adopted a Declaratory Ruling (FCC 24-17, released February 8, 2024) stating that AI-generated voices that imitate humans are "artificial" voices under the TCPA, including voice cloning. The ruling took effect immediately because it clarified existing rules. It doesn't ban AI calling. It means AI voice calls must follow the same rules as any other artificial or prerecorded voice call.
What that means for a call made with an AI voice:
| Requirement | What the ruling says |
|---|---|
| Consent (non-marketing calls) | Informational calls, such as reminders or service follow-ups, need the called party's prior express consent, unless an emergency purpose or exemption applies |
| Telemarketing or advertising | Prior express written consent is required: a signed (including electronically signed) agreement to receive these calls at that number |
| Identification | The call must state clearly, at the start, who is responsible for the call |
| Opt-out | Covered telemarketing calls must provide an interactive opt-out mechanism that records the recipient's do-not-call request and ends the call. Voicemail messages must provide the required callback opt-out method |
| Enforcement | The FCC and state attorneys general can act; 26 state AGs supported the approach |
Which rules apply depends on the call. Two questions decide it:
- What number are you calling? Federal TCPA artificial-voice restrictions apply to wireless numbers and residential lines, with different requirements and exemptions. Business landlines are treated differently under federal law, but state laws may still apply. Many B2B contacts give a mobile number, and calls to mobiles are covered whether the phone is used for work or not.
- What is the call for? Calls to wireless numbers generally require prior express consent. Telemarketing calls to covered wireless or residential numbers generally require prior express written consent. A callback request alone doesn't necessarily meet the written-consent requirement.
On top of the TCPA, the National Do Not Call Registry, the FTC's Telemarketing Sales Rule and state laws can apply. Some states, such as Florida and Oklahoma, have their own telemarketing laws that are stricter than federal law.
So an AI voice agent dialing consumers' mobile numbers with a sales pitch, when they never agreed to be called, is likely to violate the TCPA. An AI agent calling back a lead who gave the right consent, recorded properly, is a normal and lawful use. But a demo request or enquiry doesn't automatically cover every AI sales call that follows, so check what your form actually says.
What "consent" looks like in practice: for telemarketing, prior express written consent means a signed agreement that clearly authorizes the seller to deliver telemarketing calls using an artificial or prerecorded voice, names the seller, and includes the phone number the person agrees to be called at. Electronic signatures, such as a checkbox and submit on a web form that meets the E-SIGN Act, can qualify. Keep a record of the exact wording, the date and the source.
United Kingdom and EU
Two different sets of rules apply, and you need to meet both:
- Marketing-call rules (PECR in the UK, the ePrivacy Directive and national laws in the EU). These decide whether you can make the call at all. In the UK, the Privacy and Electronic Communications Regulations require specific prior consent for marketing calls made with automated calling systems, which is stricter than the rules for live human marketing calls. The Information Commissioner's Office has fined companies for unlawful automated marketing calls. The EU's ePrivacy Directive sets the same consent requirement for automated marketing calls, and each country adds its own rules.
- Data protection (UK GDPR and EU GDPR). These cover how you collect, store and use the person's data: a lawful basis for processing, telling people how their data is used, keeping records, and honoring objections and deletion requests. They apply to every call, marketing or not.
Meeting GDPR doesn't make an automated marketing call lawful under PECR, and the reverse is also true. If you call outside your own country, check that country's rules first.
The practical rule: use AI voice agents to call people who asked to hear from you, and use AI to help humans with genuinely cold outreach.
Where AI calling works in sales
| Use case | Why it works | Consent position |
|---|---|---|
| Speed-to-lead callbacks | The agent calls a new web or ad lead within a minute, while they're still interested, qualifies them and books a meeting | The lead asked to be contacted; for sales calls, capture prior express written consent on the form |
| Appointment reminders and confirmations | Fewer no-shows; the agent can reschedule on the call | Existing customers who gave their number for this purpose |
| Re-engaging past customers or old leads | Wakes up a list you already paid to build | Only contacts who opted in to calls |
| Event and webinar follow-up | Calls registrants to confirm, follow up or book a demo | Registrants who agreed to calls |
| Renewals and upgrades | Reminds existing customers before a contract ends | Existing customers with consent |
| Real estate leads | Calls portal and ad enquiries instantly, asks about budget and timing, books viewings | Enquiries that included call consent |
Example: a real estate callback.
- Someone submits an enquiry about a property on the agency's website.
- The form captures the right consent for AI calls, with the agency named and the number confirmed.
- Within a minute, the agent calls, introduces itself and names the agency.
- It asks about budget, preferred area and when they can view.
- It books a viewing in the agent's calendar, or transfers the call to a human agent if the caller is ready to make an offer.
Speed matters most. A lead that fills in a form at 9 p.m. gets a call at 9:01 p.m., not the next morning, and the agent can answer questions, qualify and book straight into your calendar.
AI voice agent vs AI-assisted human calling
There are two kinds of "AI cold calling" products, and they solve different problems.
| AI voice agent | AI-assisted human calling | |
|---|---|---|
| Who talks | The AI | Your sales rep |
| Best for | Inbound lead follow-up, reminders, re-engagement, qualification at volume | Genuinely cold B2B prospecting |
| AI's job | The whole conversation, then hand off or book | Research, call scripts, live prompts, notes and CRM updates |
| Consent rules | Artificial-voice rules apply | Normal telemarketing rules for live calls |
| Examples | Convocore, Retell, Vapi, Bland, Synthflow, Autocalls | Dialers and sales-engagement tools with AI features |
Many teams use both: an AI agent follows up every inbound lead instantly, and human reps, with AI help, handle cold prospecting.
What AI calling costs
Voice AI is priced per minute. These are the published rates for six platforms, checked on their official pricing pages on November 10, 2026:
| Platform | Published price | What it includes |
|---|---|---|
| Convocore | Business plan $99/month (25,000 credits); about $0.045–$0.105/minute by voice model | Voice model + $0.02 platform fee + Twilio telephony estimate, billed from credits |
| Retell AI | $0.07–$0.31/minute | Depends on the voice, LLM and telephony chosen |
| Vapi | $0.05/minute hosting | Plus your own speech, LLM and telephony providers |
| Bland AI | Start plan: $0.14/minute. Build plan: $299/month subscription, plus $0.12/minute | Per-minute rate on top of the plan fee; telephony billed separately |
| Autocalls | From $34/month, including 120 minutes; $0.20/minute overage | Monthly plan with a minute allowance; higher plans include more minutes and lower overage rates |
| Synthflow | Quote-based (enterprise) | Not published, so not directly comparable |
These rates aren't directly comparable. Some include telephony and voice infrastructure, while others show only the platform or hosting fee. Convocore's range includes the voice model, platform fee and estimated Twilio telephony. Vapi's $0.05 is hosting only, before speech, LLM and telephony costs. Retell's range depends on the providers and configuration you pick. Your actual cost depends on the providers and configuration you choose, so compare a full cost per minute for your own setup.
Example: 1,000 minutes of calls a month on Convocore. On the Gemini Live voice model, 1,000 minutes × $0.045 = $45 of usage. The Business plan's 25,000 credits cover $25, so you'd pay $99 + $20 = about $119 a month. On Grok Live or Ultravox ($0.105/minute all in), usage is $105, so about $179 a month. Phone numbers are included on Business (2 Twilio numbers).
Remember the calls that don't connect. Outbound campaigns include unanswered calls, voicemails and short hang-ups. Budget for them when you estimate minutes, and check whether a platform charges a minimum per call attempt.
How to set up an AI outbound calling agent
- Fix consent first. Add clear call-consent wording to every form that feeds the campaign, and store the consent record with the lead.
- Write the agent's job in one sentence. For example: "Call new demo requests, confirm their company size and timeline, and book a 20-minute call."
- Build the agent. Give it your offer, pricing rules, objections and FAQs, and the exact opening line: who you are, why you're calling, and that the person asked to be contacted.
- Connect your calendar and CRM so it can book meetings and log results.
- Connect a phone number. You'll need a Twilio or SIP number for the agent to call from.
- Add a handoff. Transfer to a human when someone is ready to buy, upset, or asks for a person.
- Respect opt-outs on the call. If someone says "don't call me again", the agent should confirm and the number should go on your internal do-not-call list.
- Test on 20–50 calls, listen to the recordings, fix the script, then scale.
- Track the right numbers: connect rate, conversation length, meetings booked and opt-out rate.
In Convocore, outbound calls run from the agent's Leads → Campaigns tab. You import your consented leads, assign a Twilio or SIP phone number to the agent, then add a voice campaign and set options such as how many calls run at once. The campaign table then tracks total leads, total calls, successful, ongoing and failed calls in real time.
Why one platform for voice and chat matters. Many leads don't answer the first call. Convocore builds voice agents and chat agents in the same platform, using the same knowledge and instructions, so the same lead can be followed up by phone and then handled on WhatsApp, Instagram or web chat when they reply there. The Campaigns tab has a WhatsApp option next to Voice Calls for the same reason.
What happens when the AI calls someone
Buyers often ask what the call actually sounds like. Here's what a well-built outbound agent should do at each point:
| Moment | What the agent should do |
|---|---|
| The opening | Say who it is, which business it's calling for, and why: "Hi, this is Ava calling from Bright Smile Dental about the appointment request you sent this morning." The FCC rules require the caller to be identified at the start. |
| "Am I talking to a robot?" | Answer honestly that it's an AI assistant, and offer a person if they'd prefer. Don't let an agent claim to be human. Some state laws add AI disclosure requirements. For example, Utah requires certain businesses using generative AI in consumer transactions to disclose that the customer is interacting with AI when directly asked. |
| "Can I speak to a person?" | Transfer the call, or book a callback if no one is free. A warm transfer, where the agent tells the caller it's connecting them and passes on the context, is better than a cold one. |
| Objections and interruptions | Stop talking when the person speaks, answer the objection from approved talking points, and avoid arguing. A "not interested" ends the pitch politely. |
| "Don't call me again" | Confirm, end the call, and flag the number for your internal do-not-call list. |
| After the call | Log the outcome (booked, callback, not interested, wrong number, voicemail), the transcript and a summary in your CRM, and trigger the next step. |
On Convocore, voice agents can transfer calls to a person when the caller asks or your rules say so, book on Google Calendar or Calendly, and keep the transcript, recording and summary for your team and CRM. Test these on your own phone before you call real leads.
Practice cold calls with an AI role-play agent
One lower-risk AI calling use is training. Build an agent that plays a tough prospect, such as a busy office manager or a skeptical homeowner, and let reps practice their openers and objection handling. Because the rep initiates the call to the role-play agent, this doesn't raise the same outbound AI-calling consent issue. Recording and data privacy rules may still apply.
AI calling for agencies
Speed-to-lead calling is one of the easiest AI services to sell to local businesses: real estate offices, clinics, home services and car dealers all lose leads they don't call back fast enough. On Convocore's white label plans ($199/month for 20 clients, $349/month for unlimited clients), you can build a callback agent once, deploy it for each client in their own sub-account under your brand, and bill them monthly.
Frequently asked questions
Try it on your own leads
Build a callback agent with Convocore's voice agent, test it on your own phone, then connect it to your lead forms. Voice is available on the Business plan ($99/month, see pricing), or you can resell it to clients under your own brand.
Related reading: Bland AI alternatives · My AI Front Desk vs Convocore · How to start an AI agency
Sources
- FCC Declaratory Ruling FCC 24-17 (CG Docket No. 23-362), adopted February 2, 2024, released February 8, 2024
- FCC: FCC Makes AI-Generated Voices in Robocalls Illegal
- Cooley: FCC – AI-Generated Robocalls Illegal Under the TCPA (February 15, 2024)
- eCFR: 47 CFR 64.1200, delivery restrictions (TCPA rules)
- Convocore: AI voice agent with human handoff
- ICO: Electronic and telephone marketing (PECR)
- Convocore pricing (checked November 10, 2026)
- Retell AI pricing (checked November 10, 2026)
- Vapi pricing (checked November 10, 2026)
- Bland AI pricing (checked November 10, 2026)
- Autocalls pricing (checked November 10, 2026)
- Synthflow pricing (checked November 10, 2026)