The short answer on AI cold calls

AI-generated voice is not a loophole around robocall regulation. In February 2024, the FCC adopted a declaratory ruling that AI-generated voices fall within the TCPA’s restrictions on artificial or prerecorded voices. For many calls to wireless numbers, artificial or prerecorded voice calls require prior express consent; telemarketing calls that use covered technology generally require prior express written consent. The exact rule depends on the call and recipient, so classify the campaign before dialing.

A live AI conversation generated dynamically can still use an artificial voice. It is not necessarily treated like a human placing a manual call simply because the caller can respond in real time. The FCC’s ruling addressed AI-generated voices as artificial under the statute; it did not create a blanket exemption for conversational agents. Use the actual method of initiation, voice generation, number type, and purpose in legal review.

“Cold call” is a business phrase, not a complete legal category. A manually dialed call from a human salesperson, a prerecorded message, an AI voice agent, a call to an existing customer, a business-to-business call, and a nonprofit informational message can have different treatment. Do not apply one campaign’s consent conclusion to another without checking the facts.

The safest answer for a consumer-facing AI sales call is to treat consent as a prerequisite until counsel confirms the precise exemption or permission that applies. Keep proof tied to the specific seller, phone number, communication type, and purpose. A purchased lead list or prior relationship does not automatically satisfy every consent requirement.

Understand what the FCC AI voice ruling says

The FCC’s February 2024 Declaratory Ruling says calls made with AI-generated voices are “artificial” under the TCPA. The ruling applies the statute’s existing framework to a newer voice-generation technology. It does not mean every use of AI in a phone system is prohibited; it means the voice and dialing context must be analyzed under the rules that apply to artificial or prerecorded calls.

The distinction between inbound and outbound matters. An AI receptionist answering a call initiated by a customer is different from a business using an AI voice agent to place unsolicited sales calls. This guide focuses on outbound calls. An inbound service still needs privacy, disclosure, recording, and consumer-protection review, but outbound consent restrictions should not be casually imported or ignored.

Do not assume a text-to-speech voice is outside the ruling because it is not a recording of a specific person. The FCC’s interpretation covers AI-generated voice. Likewise, changing the model or making the conversation less scripted does not itself remove the artificial-voice issue. If the campaign places calls to consumers, have counsel assess the call technology against current FCC rules and court decisions.

The FCC’s rules and court decisions can change. The Commission was considering a political robocall waiver request on October 7, 2026, but a pending petition is not a general commercial exemption. Check the current FCC docket and the rule in force at launch rather than assuming a proposal has already changed the law.

Separate calls by purpose and recipient

Create a campaign inventory. For every call type, record the seller, purpose, recipient category, number source, dialing method, voice method, expected content, call frequency, time zone, consent basis, and opt-out path. Separate sales calls from transactional notices, appointment reminders, debt collection, political communications, surveys, and business-to-business outreach. Labels do not control legal treatment; the call’s actual purpose and content matter.

A call that begins as a survey but includes a pitch may be treated as telemarketing under FTC guidance. A customer-service reminder that adds an offer can change its purpose. A call to an employee’s mobile number may still be a call to a wireless number. Ask counsel to classify edge cases, especially mixed-purpose campaigns and calls generated from a lead marketplace.

Identify whether the recipient is a consumer, business contact, existing customer, or someone who has opted out. The FTC Do Not Call Registry provisions generally cover telemarketing to consumers, while the FTC FAQ notes that the Registry does not cover business-to-business calls; other FCC or state requirements may still apply. Do not interpret the B2B distinction as a universal exemption from consent, deception, or state laws.

Build separate scripts and suppression rules for each campaign. A sales team should not reuse a consent record gathered for appointment reminders unless that record clearly authorizes the new seller, channel, and purpose under applicable law. Keep consent evidence and campaign definitions together so an audit can reconstruct why a number was called.

Know when prior express written consent may be required

The TCPA and FCC rules restrict certain calls that use an automatic telephone dialing system or an artificial or prerecorded voice. Telemarketing calls to covered numbers, including wireless numbers, generally require prior express written consent. Federal regulations define detailed restrictions and exceptions. The caller needs to understand the actual technology, recipient type, and purpose rather than relying on a generic “we have consent” field in a CRM.

For calls delivering prerecorded telemarketing messages, the FTC’s Telemarketing Sales Rule includes a prior signed written agreement requirement and an interactive opt-out mechanism, subject to stated exceptions. A dynamically generated AI conversation can raise questions about how these provisions apply alongside TCPA rules. Do not assume the FTC’s prerecorded-message provisions are the only relevant rules or that conversational generation eliminates them. Ask counsel to analyze both frameworks.

Consent should be specific and provable. Preserve the disclosure shown to the consumer, the seller identified, the phone number, timestamp, source page or form, affirmative action, and any later revocation. If a third party collected the lead, obtain evidence of the form and language the person actually saw. A spreadsheet with a “consented” flag is weak evidence if no one can reconstruct the permission.

Do not make consent to marketing calls a hidden condition of receiving a product or service where the law prohibits conditioning. Present the choice clearly and keep the sales seller identified. If a form names multiple affiliates or “marketing partners,” get legal review before treating it as consent for every entity. Consent that does not identify the seller or number may be insufficient for a particular campaign.

Apply Do Not Call and opt-out controls

The FTC’s National Do Not Call Registry FAQs state that businesses making covered telemarketing calls generally must access the Registry and remove listed numbers from their calling lists. They also describe entity-specific requests, calling-hour limits, prompt disclosures, and other obligations. The FTC’s compliance guide explains that telemarketers must maintain procedures and update lists using a Registry version no more than 31 days old for its safe-harbor process. Verify current requirements and exceptions with counsel.

Maintain an entity-specific suppression list for people who ask your company to stop calling. Make the AI agent recognize ordinary language such as “remove me,” “don’t call again,” “stop,” or “I’m not interested in calls.” Do not make the person repeat the request or continue the pitch after an opt-out. Confirm the preference, stop the call, and propagate suppression to all dialers, CRMs, vendors, and future uploads.

If a prerecorded telemarketing message is used, FTC guidance describes an automated interactive opt-out mechanism at the outset, with a way to add the number to the seller’s internal do-not-call list. A voicemail path also requires attention to the applicable callback and opt-out procedures. An AI voice agent should be tested for interruptions and natural-language opt-outs, not only keypad input.

Test suppression end to end. Enter a number into the internal list, sync to campaign tools, import a fresh lead file, and verify the number remains blocked. Then place a test opt-out during a call and confirm the event is logged. A compliance control that exists only in the voice vendor but not in the CRM and dialer can fail when the next campaign is uploaded.

Respect calling hours and required disclosures

FTC consumer guidance says telemarketers generally may not call before 8 a.m. or after 9 p.m. in the consumer’s local time. A campaign must use the recipient’s location, not simply the caller’s office time zone. State laws may impose narrower windows or additional restrictions. Build time-zone-aware scheduling and daylight-saving handling into the dialer, then test numbers near state borders and records with missing location data.

The FTC consumer FAQ says covered telemarketers must promptly identify the seller and that the call is a sales call or charitable solicitation, disclose material information, transmit caller ID where possible, and connect a live telemarketing call to a sales representative within two seconds after answer. Other rules apply depending on the call type. Review the exact prompt disclosures and abandonment rules for the campaign.

Make the AI identity clear and avoid impersonation. Do not have the voice claim to be a named employee, a government office, or a local business owner when it is not. A human-sounding voice does not justify obscuring the caller’s identity. Keep the company name, sales purpose, and a real path to a person clear at the opening of the interaction.

Review caller ID and number reputation. Use numbers the business is authorized to use, transmit accurate identification, and monitor carrier labeling. Do not rotate numbers to evade blocks, hide origin, or continue calling people who have opted out. A high answer rate achieved through deceptive caller ID is not a durable acquisition strategy.

Check state laws and recording rules before scaling nationally

Federal TCPA and FTC rules are not the entire picture. States may have mini-TCPA statutes, telemarketing registration or bonding requirements, broader definitions of automated calls, stricter calling hours, additional consent rules, private rights of action, and laws covering call recording or privacy. Local rules can matter too. A campaign that is designed for one state should not automatically be expanded nationwide.

Build a state matrix with the rules counsel has identified: registration, consent, time windows, frequency caps, disclosure language, recording notice, caller-ID requirements, exemptions, and record retention. Assign a source and last-reviewed date to each entry. If a consumer’s location is unknown, choose a conservative route or block the call until the record can be classified.

Recording a call to monitor quality is a separate legal issue. Federal and state laws differ, and participants may be in different jurisdictions. Have counsel review recording and transcription notices. Decide whether calls need to be recorded at all, whether transcription can be used without storing audio, and how recordings are secured and deleted. Consent to receive a sales call is not necessarily consent to every downstream use of the recording.

The regulatory status of proposals, court decisions, and FCC proceedings can change. Keep a launch checklist that triggers legal review after a material legal update, new state entry, vendor change, new call purpose, new AI model, or new lead source. Version the script and consent language so the business can prove which disclosure was in use when a number was called.

Design a consent and evidence record

A strong consent record should show who gave permission, which seller received it, the number authorized, what communication type and purpose were described, the exact disclosure, how the consumer affirmatively acted, when it happened, and any revocation. Preserve the page or script version, timestamp, source URL, and audit trail. If the lead came from a partner, preserve the partner’s collection evidence and the transfer terms.

Link consent to the number and seller at the point of dialing. If the number changes, consent may not follow automatically. If a number is reassigned, a previous person’s consent should not be treated as the new subscriber’s permission. Use number-validation and reassigned-number controls where appropriate, and follow counsel’s procedures for uncertain records.

Store consent in a system that the dialer checks before every call. Define expiration or review rules if the business uses them, and respect revocation immediately. Restrict editing rights and retain change history. Do not let a sales representative override a block without an authorized compliance review.

Review lead sources for provenance. Purchased data may be stale, recycled, duplicated, or collected under a disclosure that does not cover your company. Require suppliers to document how numbers were obtained and how consent was captured. Monitor complaints by source and suspend a supplier when evidence is incomplete or opt-out rates spike.

Set AI voice guardrails and quality controls

An AI agent can make a conversation more responsive while also creating new failure modes. It may misunderstand a refusal, continue speaking over a person, misstate an offer, invent an answer, or fail to hand off. Restrict the agent to an approved knowledge base and a limited set of actions. Keep a human supervisor available for complaints, complex requests, and people who want to speak with someone.

The script should identify the seller and the sales purpose, explain the offer accurately, answer only approved questions, and provide a clear opt-out. Do not allow the model to invent discounts, guarantee savings, make unsupported product claims, or imply a prior relationship. If a caller asks who is calling, asks whether the voice is AI, or requests a person, answer plainly.

Set conversation-level stop rules: opt-out, wrong number, distress, dispute, legal threat, vulnerable-person concern, or low confidence. Record the result and suppress where needed. Create an escalation queue that is monitored and has an owner. Test interruption behavior, accents, background noise, silence, and nonstandard phrasing. Do not use a successful scripted demo as evidence that real-world calls are compliant.

Monitor quality with a statistically useful sample, and review every complaint and opt-out failure. Track disclosures, answer accuracy, opt-out detection, call duration, abandoned calls, transfers, and caller complaints. A low complaint count may reflect low call volume or poor reporting, so compare rates by campaign and data source. Pause a campaign when control performance falls below the threshold counsel and compliance set.

Use a pre-launch compliance checklist

Before launch, confirm the call purpose, seller, recipient class, number type, dialer behavior, voice generation method, consent basis, state coverage, calling hours, caller ID, script, recording approach, do-not-call scrubbing, opt-out processing, vendor contracts, and retention. Have counsel sign off on the classification and any relied-upon exception. Keep the approval with the campaign configuration.

Run test cases for a person who has consented, a number on the National Registry, an internal opt-out, a wrong number, a revoked or missing consent record, a number in a different time zone, a person who asks for a human, and a voicemail answer. Confirm no call launches when required data is missing. Test the scheduled-campaign queue after a suppression update.

Review vendors and subcontractors. Identify who initiates the call, who controls the phone number, who stores the recording, who handles opt-outs, and who responds to complaints. Contracts should allocate responsibilities clearly and require prompt data sharing. The seller remains accountable for its campaign even when a vendor operates the dialer or AI voice.

Define the stop switch. A compliance owner should be able to pause the campaign and downstream retries immediately. Keep a rollback contact list, preserve evidence, and document how to notify vendors. Do not depend on an engineer who is unavailable after hours to stop a call sequence.

Common AI cold-call compliance mistakes

One frequent mistake is treating a lead’s presence in a CRM as proof of consent. The record may have been imported without its source form, seller identity, date, or disclosure. Before dialing, require an evidence record that can be retrieved and tied to the number. If the consent language is missing or unclear, place the record in a review queue rather than guessing.

Another mistake is assuming that a human-like conversation is a manual call. The FCC has classified AI-generated voices as artificial under the TCPA. A conversational model can still be an artificial voice, and the dialer’s method of initiation can trigger separate rules. Document the technical call path with the vendor and have counsel classify it.

Businesses sometimes apply a prior customer exception too broadly. An established relationship may matter to a particular Do Not Call analysis, but it does not automatically authorize every prerecorded or AI-generated sales call, every seller, or every number. Check whether the person has opted out and whether the exact call technology and purpose are covered.

A further failure is collecting an opt-out but not syncing it everywhere. If the voice platform stops the current call but the CRM, dialer, lead vendor, and retry queue do not receive the suppression, the person may be called again. Test every system and keep a central suppression record that controls new imports.

How to handle consent disputes and complaints

When a recipient says they did not consent, stop the call and suppress the number according to the company’s policy. Do not argue about a lead provider’s records during the call. Preserve the consent evidence, call metadata, script version, voice recording or transcript if lawfully retained, and the path by which the number entered the campaign. Notify the compliance owner promptly.

The reviewer should verify the seller named in the consent, number, call purpose, date, disclosure, affirmative action, and revocation history. Check whether the campaign used the approved dialer and whether any vendor changed the call path. If evidence is incomplete, pause further calls from the same source while it is investigated. Avoid deleting records needed for the review; follow the retention and legal-hold policy.

Track complaint rates by source, campaign, vendor, state, and script version. A small cluster may expose a systematic problem such as a misleading landing page, duplicate data, stale consent, or an opt-out bug. Correct the underlying process and revalidate the affected records. Do not treat each complaint as an isolated customer-service event if the same pattern recurs.

Prepare a response process with named owners in legal, compliance, sales operations, and vendor management. Decide who can stop campaigns, who answers regulators or carriers, and how evidence is exported. Run a tabletop exercise before scaling a large campaign. A fast and documented response can limit continued exposure.

Calculate campaign economics without rewarding risky volume

Outbound automation can make call volume inexpensive to scale, but cost per connected conversation is not the same as cost per compliant opportunity. Include data acquisition, consent verification, registry access, phone numbers, AI minutes, integration, monitoring, legal review, complaint handling, and opt-out administration. Estimate the cost of quality assurance and customer support before evaluating a vendor’s per-minute rate.

Measure qualified conversations, requested callbacks, booked meetings, complaints, opt-outs, wrong-party contacts, and conversion by consent source. If one source creates more complaints or missing consent records, suppress it even if its immediate conversion rate looks attractive. A long-term campaign depends on trust, deliverability, and reliable evidence, not only on short-term lead count.

Use a small controlled pilot with a low daily cap and a human review queue. Compare it with a lawful baseline channel and define stop conditions before the first call. Avoid changing the script, model, caller ID, target list, and time window simultaneously; otherwise, the business will not know which change caused a quality shift.

A campaign should scale only after it demonstrates that consent is valid, opt-outs propagate, disclosures are accurate, state controls work, and staff can handle complaints. If the only way a campaign meets its return target is by ignoring consent uncertainty or calling people who do not want contact, the unit economics are not healthy.

Inbound AI reception and outbound AI cold calls are different

An AI receptionist that answers an inbound call after a person dials the business is not the same workflow as an AI voice agent that initiates a sales call. Outbound campaigns raise consent, dialing, do-not-call, time-of-day, caller-identification, and opt-out issues that may not apply in the same way to an inbound conversation. Do not use an inbound product’s compliance page as proof that outbound use is permitted.

A business may use AI internally to summarize a call or suggest a response without generating an outbound voice. That can still involve privacy, recording, and data-use questions, but it is analytically different from placing an AI-generated sales call to a consumer. Specify which functions the product will perform and disable features outside the approved scope.

A customer-requested callback may be different from a cold sales call, but the purpose and consent language still matter. Record the original request, who made it, which number they supplied, what follow-up they authorized, and whether the request was for a single callback or an ongoing campaign. Do not expand a narrow callback permission into repeated promotional calls.

Keep campaign approvals attached to the technical configuration. If the team changes from human callers to AI voice, changes from appointment reminders to sales, or expands from one state to fifty, return to legal review. A small label change in the dialer can materially change the regulatory analysis.

Frequently asked questions

Are AI cold calls illegal in the US?

There is no single yes-or-no answer. The FCC treats AI-generated voices as artificial voices under the TCPA. Consent requirements depend on call purpose, number type, dialing method, and applicable federal and state rules.

Does an AI voice need prior written consent to make a sales call?

Many covered telemarketing calls using an artificial or prerecorded voice require prior express written consent, particularly to wireless numbers. Have counsel analyze the campaign and preserve the exact consent evidence.

Can an AI call a business number without consent?

Some rules treat business-to-business calls differently, but that is not a blanket exemption from all federal, state, deception, caller-ID, or do-not-call requirements. Verify the recipient and campaign facts with counsel.

Can a previous customer relationship authorize an AI sales call?

An established business relationship may affect some do-not-call analysis, but it does not automatically satisfy every artificial-voice or prerecorded telemarketing consent requirement. Check the exact rule and any consumer opt-out.

What should an AI telemarketing call say at the start?

The seller and sales purpose should be clear, with any disclosures and opt-out mechanisms required for the call type. Counsel should approve the script for each campaign and state footprint.

Related Sysevo guides

• AI receptionist client intake for law firms
• AI receptionist lead response for real estate agents

Sources and further reading

• FCC announcement on AI-generated voices under the TCPA
• FCC Declaratory Ruling 24-84
• FTC guide to complying with the Telemarketing Sales Rule
• FTC National Do Not Call Registry FAQs
• 47 CFR § 64.1200, delivery restrictions

See Sysevo voice workflows for inbound and outbound call operations with connected follow-up.