The short version
- In February 2024 the FCC ruled that AI-generated voices count as "artificial" voices under the TCPA.
- Outbound AI calls therefore need the same prior express consent as prerecorded calls, and written consent for telemarketing.
- Opt-outs must be honored by any reasonable means. Several states add stricter rules of their own.
If you plan to use AI voice agents for outbound calls in the United States, the Telephone Consumer Protection Act is the law that matters most. This article explains how the FCC has applied the TCPA to AI voices and what that means in practice for consent, opt-outs and campaign design.
This is general information, not legal advice. TCPA litigation is expensive and the rules change, so have counsel review your specific lists and scripts.
The February 2024 ruling
On February 8, 2024, the FCC issued a Declaratory Ruling stating that calls made with AI-generated voices are "artificial" voices under the TCPA. That includes voice cloning and voices generated by text-to-speech models.
The ruling didn't create a new category of rules. It confirmed that existing TCPA restrictions on calls using an "artificial or prerecorded voice" apply to AI voices. The practical effect is that an AI agent making outbound calls is treated like a robocall for consent purposes, even though it holds a real two-way conversation.
The ruling took effect immediately, and the FCC noted it gives state attorneys general an additional tool against illegal robocalls.
What consent you need
The TCPA's consent rules depend on who you're calling and why. Simplified:
| Call type | Consent generally required for artificial or prerecorded voice |
|---|---|
| Informational call to a mobile phone (e.g. appointment reminder, delivery update) | Prior express consent |
| Telemarketing call to a mobile phone | Prior express written consent |
| Telemarketing call to a residential landline | Prior express written consent |
| Emergency calls | Exempt |
"Prior express written consent" has specific requirements: a signed agreement (electronic signatures count) that clearly authorizes calls using an artificial or prerecorded voice to a specific number, and that tells the person consent isn't a condition of purchase.
A few practical points:
- Consent belongs to the number and the caller. Consent someone gave to a different company doesn't automatically cover you. Be very careful with purchased leads.
- Keep records. If a call is challenged, you need to show when and how consent was given. Store the form version, timestamp, IP address and the consent language shown.
- Existing customers aren't automatic. A customer relationship helps with informational calls but doesn't replace written consent for marketing calls with an AI voice.
What happened to "one-to-one consent"
In late 2023 the FCC adopted a rule that would have required consent to be given to one seller at a time, and limited calls to topics "logically and topically related" to the original interaction. It was aimed at lead generation sites that collected consent on behalf of many companies at once.
On January 24, 2025, the Eleventh Circuit Court of Appeals vacated that rule in Insurance Marketing Coalition v. FCC, holding that it went beyond what the statute allows. The one-to-one requirement never took effect.
That doesn't make shared-consent leads safe. The underlying requirement for clear, voluntary consent still applies, and courts look closely at whether consumers understood who would call them.
Opt-outs and revoking consent
People can withdraw consent, and the FCC tightened how that works. Rules that took effect on April 11, 2025 say, in summary:
- Consumers can revoke consent by any reasonable means: saying "stop calling me" on a call, replying "stop" to a text, emailing, and so on. You can't require one specific method.
- Revocation requests must be honored within 10 business days.
- A one-time confirmation text is allowed if sent within five minutes and it contains no marketing.
A related provision would require a revocation made in response to one type of message to apply to all future robocalls and texts from that caller. That "revoke all" piece has been delayed more than once, most recently to January 31, 2027. Check its current status before designing your opt-out handling.
For an AI agent, this means the agent must recognize opt-out requests in natural language ("take me off your list," "don't call me again," "I'm not interested, stop calling") and record them reliably. It's one of the most important things to test before launch.
Other rules that apply to every campaign
Identification. Artificial or prerecorded voice messages must state the identity of the business at the beginning of the message and provide a phone number during or after it.
Opt-out mechanism for telemarketing. Prerecorded telemarketing messages must offer an automated way to opt out during the call.
Calling hours. Telephone solicitations are restricted to between 8 a.m. and 9 p.m. in the called party's local time. Some states set narrower windows.
Do Not Call. Scrub telemarketing lists against the National Do Not Call Registry and your own internal DNC list. Honor internal DNC requests promptly.
Caller ID. Display a number that can be called back and that's associated with your business. Spoofed or misleading caller ID creates separate legal risk. Our article on STIR/SHAKEN and caller ID covers how carriers treat your calls.
State laws can be stricter
Several states have their own telemarketing laws, often called "mini-TCPAs." Florida and Oklahoma, for example, have laws that restrict automated calls and set their own consent and calling-time rules, and other states have followed. If you call nationally, your rules need to satisfy the strictest state you call into, or you need state-specific logic in your dialer.
Call recording adds another layer: some states require every party's consent to record. See call recording consent laws.
A compliance checklist for AI outbound calls
- Confirm the purpose of the campaign: informational or telemarketing.
- Confirm you hold the right consent for every number, with records.
- Scrub against national, state and internal DNC lists before each campaign.
- Set calling windows by the recipient's time zone and state.
- Have the agent identify the business at the start and offer a callback number.
- Make the agent disclose that it's an automated assistant.
- Teach the agent to recognize opt-outs in plain speech, confirm them and log them.
- Sync opt-outs to every system that can trigger a call or text within the required time.
- Keep transcripts or recordings in line with your recording consent and retention policy.
- Review a sample of calls every week.
How this plays out in real campaigns
Some common scenarios, and how the rules generally apply. As always, get advice on your specific facts.
Appointment reminders for existing patients or customers. These are usually informational calls. If they go to mobile phones using an AI voice, you generally need prior express consent, which is often obtained when the customer provides their number for that purpose. Keep the content strictly informational. Adding a promotion can turn the call into telemarketing.
Calling web leads who just filled in a quote form. These are typically telemarketing calls, so an AI voice requires prior express written consent. That consent has to be captured on the form, in clear language, naming your business and covering automated or artificial voice calls. Check the wording before launching.
Renewal reminders for current policyholders. Often a mix: reminding someone their policy renews is informational, while offering an upgrade or a new product is marketing. Decide which the call is, script it accordingly and make sure your consent matches.
Reactivating old customers. Usually telemarketing. Unless you have written consent that covers AI-voice calls, use human callers for this, and still scrub against Do Not Call lists.
Purchased lead lists. The riskiest category. Consent gathered by a third party may not clearly name your business or cover artificial voice calls. Many companies decline to use AI voice calls for purchased leads at all.
What good consent language looks like
The details vary and should come from counsel, but strong consent language for AI voice calls generally:
- Names the company that will call (not "our partners")
- Says calls may use an automated system or an artificial or AI-generated voice
- Covers the specific number the person entered
- States that consent isn't required to make a purchase
- Is presented clearly next to the submit button, not hidden behind a link
- Is captured with a record of the exact wording, the time and the source
Keep versions of your consent language over time. If your form changes, you need to know which version a given lead saw.
Building opt-out handling into an AI agent
Opt-outs are where AI agents can either shine or create serious risk. A well-built agent:
- Recognizes opt-out requests in many phrasings, including indirect ones like "why do you keep calling me?"
- Confirms the request plainly: "Understood. I'll make sure you don't receive more calls from us."
- Writes the opt-out to your systems immediately, not at the end of a batch
- Ends the call politely without trying to change the person's mind
- Logs the exact words used, for your records
Test this before every campaign with a set of real opt-out phrasings, and spot-check transcripts weekly.
Keeping up with changes
TCPA rules have changed several times in recent years, through FCC orders and court decisions, and more changes are likely. Assign someone to watch for updates, review your scripts and consent language at least twice a year, and treat any FCC announcement about robocalls or consent as a prompt to check your processes.
Frequently asked questions
Does the TCPA apply to inbound calls answered by AI?
The consent rules discussed here are about calls you place. An AI agent answering calls that customers make to you isn't a robocall. Recording and privacy rules can still apply.
Is an AI voice agent legal for cold calling?
Calling people who haven't given prior express written consent with an artificial voice for telemarketing violates the TCPA. In practice, AI outbound calling is suited to consented lists: existing customers who agreed to calls, and leads who asked to be contacted with proper consent language.
What are the penalties?
The TCPA allows statutory damages per call, which can be increased for willful violations, and class actions are common. That's why consent records and opt-out handling deserve real investment.
Written by the Voxvencer editorial team. We build and run AI voice agents for call centers and small businesses, and we write about what we see on real phone lines. Questions or corrections: info@voxvencer.com.
