The short answer
Usually, yes, and for marketing, the strict kind. In February 2024 the FCC ruled that AI-generated and voice-cloned calls are an “artificial or prerecorded voice” under the TCPA. That means an AI agent placing marketing calls to a mobile or residential line generally needs **prior express *written* consent first. Purely informational or transactional calls (e.g., an appointment reminder the person asked for) face a lower bar but still have rules. Get it wrong and statutory damages are $500-$1,500 per call**.
Let's make this concrete with a story. Imagine you build a friendly AI agent named “Ada” to call people about a sale. Ada is fast, polite, and tireless. The law's question is simple: did the person on the other end agree, in advance, to get this kind of call from you? If Ada is selling something, the answer needs to be a documented “yes” *before* she dials. That documented yes is what the law calls consent.
What “prior express written consent” means in plain English
It sounds heavier than it is. Broken down, it's four things:
- Prior: you got it before the call, not during or after.
- Express: a clear, affirmative agreement (a pre-checked box doesn't count).
- Written: a signed form, a checkbox they actively ticked, a documented online opt-in, or a text “START”. Digital signatures count.
- Consent to be called this way: the agreement specifically covers autodialed / artificial-voice marketing calls to that number.
For informational, non-marketing calls (a flight delay, a fraud alert, a reminder for an appointment the person booked), the standard is the lower “prior express consent,” often satisfied by the person giving you their number for that purpose. The line between “marketing” and “informational” is exactly where companies get burned, so when in doubt, treat it as marketing.
Who this hits hardest
| If you run… | Consent risk | Why |
|---|---|---|
| Outbound sales / SDR agents | High | Almost always 'marketing' → written consent needed |
| Collections agents | High | TCPA + FDCPA stack; revoked consent must be honored |
| Appointment reminders | Medium | Informational, but scope & opt-out still apply |
| Inbound support (caller dials you) | Lower | The caller initiated, but recording rules still apply |
The other half people forget: honoring opt-outs
Consent isn't a one-time checkbox; it's a relationship you have to maintain. If someone says “stop calling me,” that revocation has to actually propagate to your AI agent, and you need to be able to prove you honored it. An agent that keeps dialing a number after an opt-out turns one mistake into a willful violation (the $1,500 tier). This is precisely the kind of thing that's invisible until you check every call.
A rule that recently changed
The FCC's 2023 'one-to-one consent' rule (tightening how lead-gen consent works) was vacated by a federal appeals court in early 2025. Compliance rules in this area move fast and get challenged, so don't rely on a single blog's snapshot, including this one. Confirm the current state of play with counsel before you design a consent flow.
How to prove it, not just do it
When a TCPA claim lands, the fight is about evidence: can you show consent existed and that you honored opt-outs? That's where per-call monitoring earns its keep. Trace scans every call and flags missing disclosures, missing consent capture, and calls placed to numbers that had opted out, with a finding that cites the rule and a tamper-evident record. You move from “we're pretty sure we complied” to “here's the proof for call #4,822.”
- TCPACalling-time windowClear
Dialed 2:14pm local, inside the permitted window.
- TCPAIdentificationClear
Agent identified caller and company in the first 15s.
- PIICard dataClear
1 card number captured and auto-redacted from the transcript.
- Brand voiceProhibited claimsClear
No guarantee or earnings claims detected.
9f2c…a71eReady the moment the call endsIllustrative scorecard. Findings cite the rule pack and regulation; verdicts, redactions, and the tamper-evident hash are generated per call.
Primary sources
We cite the statute or ruling so you can verify every claim. Status changes, confirm currency with counsel.
Frequently asked questions
Do AI calling agents need consent under the TCPA?
Yes. The FCC's February 2024 ruling treats AI-generated and voice-cloned calls as an 'artificial or prerecorded voice' under the TCPA, which generally requires prior express written consent for marketing calls to mobile and residential numbers. (Educational, not legal advice.)
What is prior express written consent?
A clear, signed-or-actively-opted-in agreement, obtained before the call, in which the person specifically agrees to receive autodialed or artificial-voice marketing calls at that number. A pre-checked box does not qualify.
Do informational AI calls need consent?
They face a lower 'prior express consent' standard, often met when the person gives you their number for that purpose, but scope limits and opt-out rules still apply. The hard part is correctly classifying a call as informational vs. marketing.
What happens if my AI agent calls someone who opted out?
Continuing to call after an opt-out can convert an ordinary violation into a willful one, exposing you to the higher $1,500-per-call tier. Revocations must propagate to the agent, and you should be able to prove you honored them.