The Two-Part Test That Decides Almost Every Autodialer Case

When the Supreme Court finally weighed in on what counts as an autodialer, the answer surprised a lot of people. In a unanimous 9-0 decision on April 1, 2021, the Court held that equipment must use a random or sequential number generator to qualify under the TCPA. Dialing numbers from a list you already have? That doesn't count.

The statutory definition comes from 47 U.S.C. § 227(a)(1), and it's been on the books unchanged since 1991. It defines an autodialer as equipment with the capacity to "store or produce telephone numbers to be called, using a random or sequential number generator" and to "dial such numbers." The fight was over what those words actually mean, and six federal circuits were divided on the question before the Court stepped in.

Justice Sotomayor's opinion applied the series-qualifier canon, holding that "using a random or sequential number generator" modifies both "store" and "produce." Congress, the Court reasoned, was targeting a specific problem: equipment that randomly dials emergency lines or ties up sequentially numbered lines at a single business. A broader reading would capture "virtually all modern cell phones," and the Court warned against taking "a chainsaw to these nuanced problems when Congress meant to use a scalpel."

That last point matters for anyone running outbound campaigns. Lower courts have since rejected repeated attempts to stretch the ruling. Plaintiffs have leaned on footnote 7 of the decision to argue that list-mode calling still qualifies as autodialer use, but three circuits, including the Third in Perrong v. Montgomery County Democratic Committee (April 2024), have said no. The practical rule from these cases: even if your system selects the next number to dial, it isn't an autodialer under the TCPA if the customer supplied the list of phone numbers being called.

So why hasn't Congress updated a 34-year-old definition? The statute was written when random number generation was the threat. Today, most calling platforms work from pre-established lists, which is why the Supreme Court's narrow reading "significantly limits the reach of the TCPA" and is expected to stem the tide of class actions against legitimate businesses. Nonpartisan analysis from the Congressional Research Service notes that a technology-specific definition may eventually require amendments if new tools can circumvent the law's restrictions, and the plaintiffs' bar is already lobbying for change.

A few things to keep in mind:

• The "capacity" question is still unsettled. Some circuits look at a device's capabilities at the time of the call; others focus on its functions.
• The FCC's 2015 broad definition was invalidated in 2018 (*ACA International v. FCC*) for creating an "eye-popping sweep" that threatened to bring every smartphone under the statute.
• Calling a large volume of numbers is "not probative" of autodialer status, per the FCC's 2020 ruling.
• A narrow autodialer definition is not a safe harbor. The TCPA separately prohibits "artificial or prerecorded voice" calls, and the FCC confirmed in February 2024 that AI-generated voices count as artificial, requiring prior express consent.

That last point is the one we take most seriously at My AI Call Center. Even when calling from approved, permissioned, or reviewed lists, AI voices still require prior express consent, disclosure, and immediate opt-out handling. The autodialer test decides a lot, but consent discipline decides more. If you're planning outbound calls and want the list and consent review done before anything launches, you can talk through a campaign with us first.

Why the Old Broad Definition Died, and What Courts Do Now

The FCC tried to stretch the definition of "autodialer" once, and a federal appeals court cut it back. In 2015, the agency issued a Declaratory Ruling that read "capacity" to include "potential functionalities" and "future possibility." That meant a device could count as an autodialer based on what it might someday do. The D.C. Circuit struck that reading down in ACA International v. FCC (2018), calling it an "eye-popping sweep" that threatened to bring every smartphone under the statute (Wiley).

The Supreme Court settled the core question in Facebook, Inc. v. Duguid on April 1, 2021. In a unanimous 9-0 decision, the Court held that "using a random or sequential number generator" modifies both "store" and "produce." Equipment that dials from a pre-existing list is not an autodialer. The Court applied the series-qualifier canon, reasoning that Congress targeted "a unique type of telemarketing equipment" and warning that a broad reading would "take a chainsaw to these nuanced problems when Congress meant to use a scalpel" (Crowell & Moring).

Plaintiffs then tried to expand the definition through footnote 7 of Duguid. Three circuits rejected that move:

• Borden v. eFinancial (9th Cir. 2022): an autodialer must randomly or sequentially generate telephone numbers.
• Beal v. Outfield Brew House (8th Cir. 2022): systems that randomly select from non-random phone numbers are not autodialers.
• Perrong v. Montgomery Cnty. Democratic Comm. (3d Cir. 2024): using a number generator to select and dial from a previously compiled list is not autodialer usage.

The practical rule from these decisions: if the customer supplied the set of numbers, the technology isn't an autodialer, even when it picks the next number to dial (Roth Jackson).

Two loose ends remain. Duguid never defined "capacity," and a circuit split persists: the Second, Third, and D.C. Circuits look at a device's capabilities at the time of the call, while the Seventh, Ninth, and Eleventh Circuits focus on the device's functions (CRS). Separately, the FCC's June 25, 2020 Declaratory Ruling clarified that calling a large volume of numbers is "not probative" of autodialer status; the test is whether equipment can dial random or sequential numbers without human intervention (Hinshaw).

That narrow definition isn't a safe harbor. The TCPA separately bans "artificial or prerecorded voice" calls, and in February 2024 the FCC confirmed AI-generated voices qualify as artificial, with no carve-out for technology that mimics a live agent (Wilson Sonsini). For anyone running AI calling campaigns, consent discipline is the real compliance lever, which is why My AI Call Center reviews list source and consent records before any campaign launches.

The Trap Most Callers Miss: AI Voices Are Regulated Separately

A narrow reading of the autodialer definition feels like a safe harbor for AI calling programs. It isn't. The FCC closed that door on February 8, 2024, when it ruled in FCC 24-17 that AI-generated voices, including cloned voices, count as "artificial or prerecorded voice" under the TCPA. Critically, the ruling allows no carve-out for technologies that "purport to provide the equivalent of a live agent," as Wilson Sonsini's analysis of the ruling explains.

That matters because the artificial voice provision is a separate trigger from the autodialer definition. In Facebook v. Duguid, the unanimous Supreme Court noted the TCPA separately prohibits artificial or prerecorded voice calls, so winning the autodialer argument does not end the exposure, as Crowell & Moring's client alert points out. The autodialer test itself is narrow: equipment must store or produce numbers using a random or sequential generator. Dialing a pre-existing list is not autodialer use, a rule three circuits have now confirmed.

For an AI calling program, the practical consequences of FCC 24-17 are concrete:

• Prior express consent is required before an AI voice call, regardless of how the numbers were dialed.
• The caller must disclose its identity at the start of the call and state its telephone number.
• Telemarketing calls need an opt-out mechanism available within two seconds of the required disclosures.
• Statutory damages run $500 to $1,500 per call, assessed call by call.

That per-call math is why the autodialer definition is the wrong thing to build a program around. A campaign that never touches a random number generator can still generate four-figure exposure on every connection if consent records are missing.

This is the discipline My AI Call Center applies to the campaigns it runs: only approved, permissioned, or reviewed lists, with list source and consent records checked before launch. AI disclosure runs on every call, keyword opt-outs like STOP and REVOKE are honored immediately, and DNC requests carry across campaigns. The law here keeps moving, and the "capacity" question remains a live circuit split, so clients should get their own legal guidance before launch.

Run more useful calls without building a bigger call center. Campaigns start at 9¢ per connected minute on approved, permissioned lists. Plan my campaign and get the full number quoted before you approve anything.

What This Means for Your Campaign: Consent, Windows, and Records

The narrow autodialer definition is good news, but it's not a green light to dial anything. The FCC's February 2024 ruling that AI-generated voices count as "artificial" voices under the TCPA means consent discipline, not dialing technology, is what actually protects your campaign. Here's how that translates into operational steps.

Consent comes first. Non-emergency autodialed calls and texts to cell phones require prior express written consent for telemarketing and prior express consent for non-telemarketing calls. Even after the Supreme Court narrowed the autodialer definition, the FCC noted that liability can be avoided by obtaining the recipient's prior express consent. Statutory damages run $500 to $1,500 per call, so a bad list gets expensive fast. That's why list source and consent records are checked before any My AI Call Center campaign launches, and bought lists without clear permission records are flagged, and in most cases declined.

Disclose the AI voice. There's no carve-out for technologies that "purport to provide the equivalent of a live agent," so every call should disclose identity at the start, state the caller's telephone number, and let recipients ask whether the call is AI-assisted or request a human.

Give a fast opt-out. For telemarketing calls using AI voices, an opt-out mechanism must be available within two seconds of the required disclosures. Keyword opt-outs like STOP and REVOKE should be honored immediately and logged.

Respect quiet hours and state rules. State-specific quiet hours, day restrictions, and registration requirements vary by location, and calls should run only inside approved windows.

Keep DNC records current. Opt-outs should carry across all campaigns and into your DNC records, not just the campaign where the request came in.

• Confirm the list source and consent records before launch
• Disclose the AI voice and provide a two-second opt-out path
• Call only inside approved windows for the recipient's state
• Log and honor every opt-out across all campaigns

One caution: the law is still moving. Courts remain split on what "capacity" means for autodialer status, and lawmakers have discussed amending the statute. Campaign requirements also vary by location, industry, contact type, and consent status, so it's worth getting appropriate legal guidance before launch. When your list, consent records, and calling windows are solid, the rest of the compliance stack becomes manageable.