What the TCPA Actually Says About Autodialers

The word that stops most business texting projects before they start is autodialer. Someone read that the Telephone Consumer Protection Act has steep per-message penalties for using one, decided that any software that sends a lot of texts must be an autodialer, and concluded the whole idea is radioactive. That conclusion is wrong, but the instinct behind it is reasonable: the term has a specific legal meaning, the penalties are real, and almost nobody explains the definition in language a business owner can use.

So here is the definition, in plain terms, with the one Supreme Court case that settled what it means and the appeals-court decisions that followed. The short version is that an autodialer under the TCPA is a narrow, technical category — much narrower than “software that sends texts” — and whether a given system falls inside it turns on one clause of one sentence of the statute.

This article is only about that definition: what an autodialer is, what it is not, and how to work out which side of the line a tool sits on. It deliberately does not cover the rest of the law — the Do Not Call rules, opt-outs, send windows, state statutes, carrier registration — because those are separate questions that the autodialer definition does not answer. For the whole operating picture, the TCPA compliance playbook walks through all of it.

What the statute actually says

The whole question starts with one definition. 47 U.S.C. § 227(a)(1) defines an automatic telephone dialing system — the statute’s term, usually shortened to ATDS and loosely called an autodialer — as equipment which has the capacity:

• (A) to store or produce telephone numbers to be called, using a random or sequential number generator; and
• (B) to dial such numbers.

Read that twice, because the whole fight lives in the last six words of clause (A). The equipment has to do something with telephone numbers — store them or produce them — and it has to do that thing using a random or sequential number generator. A random number generator spits out numbers with no pattern. A sequential generator counts up: it takes a block and produces every number in it, in order. The statute is describing the machines that plagued people in the early 1990s, dialing entire exchanges digit by digit whether or not anyone there had ever been contacted.

That is a specific piece of functionality, and most business texting tools do not have it. A tool that sends to a list you uploaded is not generating anything — the numbers already existed and came from somewhere you can point to. Whether that distinction is enough to put a tool outside the definition is exactly what went to the Supreme Court.

Why the wording was worth a Supreme Court case

For years, courts could not agree on what clause (A) covered, and the disagreement was grammatical. Does “using a random or sequential number generator” attach only to the nearer verb, produce, or to both verbs, store and produce?

It sounds like a question for an English teacher, and the stakes were enormous. If the phrase modified only “produce,” then any equipment that could merely store numbers and dial them automatically would be an autodialer — and that reading sweeps in essentially every modern phone and every CRM, because they all store numbers and can dial them. Under the broad reading, a mass-texting platform that dialed from a saved list was an autodialer. Under the narrow reading, it was not, because it had no generator producing the numbers.

Different federal appeals courts had landed in different places, so the same texting setup could be lawful conduct in one part of the country and the basis for a class action in another. The Supreme Court took the case to resolve that split.

What Facebook v. Duguid decided

In Facebook, Inc. v. Duguid, 592 U.S. 395 (2021), the Supreme Court chose the narrow reading. It held that the clause “using a random or sequential number generator” modifies both “store” and “produce.” To be an autodialer, equipment must use a random or sequential number generator either to store the numbers it will dial or to produce them.

The Court’s own reasoning is the part worth keeping, because it tells you how to think about edge cases. It pointed out that the broad reading would make the definition cover almost every smartphone in America, since any phone can store numbers and dial them, and Congress could not have meant to regulate ordinary personal phones. It described the broad reading as taking “a chainsaw to these nuanced problems when Congress meant to use a scalpel.”

The practical consequences follow directly. Sending text messages in bulk is not, by itself, autodialing — volume was never the test. A CRM is not an autodialer. Scheduling messages is not autodialing. Dialing from a list you compiled and uploaded is not autodialing, because nothing generated those numbers. The feature the statute names is simply absent.

Be honest about one limit, because glossing it is how people get overconfident. Duguid did not hold that every list-based system is categorically outside the definition for all time. In footnote 7 it noted that an autodialer “might use a random number generator to determine the order in which to pick phone numbers from a preproduced list” and did not decide that scenario. It is a small opening, courts have mostly read it narrowly, but it is there.

What the courts added after Duguid

Three appeals-court decisions since 2021 closed the gaps plaintiffs tried next, and together they sharpen the definition into something you can actually apply.

• Borden v. eFinancial (9th Cir. 2022) — the generator has to produce the numbers themselves. Using a generator only to pick the order in which you message a list you already have is not autodialing. This is the footnote-7 scenario, and the Ninth Circuit read it against the broad interpretation.
• Panzarella v. Navient (3d Cir. 2022) — the statute bans making a call using an autodialer, so it is not enough that an autodialer capacity exists somewhere in a system. The question is whether you used that capacity to place the message in front of you.
• Soliman v. Subway (2d Cir. 2024) — a text message is not an “artificial or prerecorded voice.” That matters because § 227(b) is triggered by an autodialer or a prerecorded voice, and this closes the second route as applied to texting.

Put the three together and the test for whether a system is an autodialer reads: does it have the capacity to store or produce the numbers it dials using a random or sequential number generator, and did you actually use that capacity? If the numbers came from a list you built, the answer to the first half is no, and the analysis usually stops there.

So is your texting tool an autodialer?

This is the question the law-firm write-ups almost never answer for the person doing the sending, so here is a practical way to work through it. It is not a legal opinion about your program — whether any specific system meets the definition is a fact question, and your counsel decides it — but it tells you what to look at.

Ask three things about the tool and the list it sends to:

1. Where did the numbers come from? If you can name the source — a list you built, a public business directory, numbers businesses published themselves — then the numbers were not generated. If the answer is “the tool finds them,” find out exactly how, because that is where a generator could hide.
2. Does anything in the stack produce numbers by pattern? Random dialing, sequential fills, guessing the rest of an exchange, permuting area codes — all of that is the functionality the statute names. A tool that only ever sends to numbers you put into it does not do this.
3. Is the capacity present and used? After Panzarella, a feature you never touch is a weaker basis for liability than one you ran. A tool with no generator at all is cleaner still.

The table below sorts the common setups against the definition the Supreme Court described. “Does not meet the definition” is a statement about the statutory category, not a verdict that a given program is lawful overall — the rest of the law still applies.

The pattern is consistent: the moment anything in your process creates phone numbers rather than sending to numbers that already existed, you are in the category the law was written for. As long as every number traces to a real source you recorded, the autodialer definition is not the thing that reaches your program.

What “not an autodialer” does not settle

Here is the mistake to avoid after reading all of the above: treating “our tool is not an autodialer” as “our texting is legal.” The autodialer definition is one input. Several others sit entirely outside it.

• The prerecorded-voice half of the same rule. § 227(b) is violated by an autodialer or by an artificial or prerecorded voice. A no-autodialer position does nothing for a recorded voice drop or a ringless voicemail, which travel a different route to the same statute. Keep voice out of a text pipeline.
• The Do Not Call rules. 47 C.F.R. § 64.1200(c)(2) is written around the residential telephone subscriber and has nothing to do with whether you used an autodialer. It turns on who the recipient is, not what equipment you dialed with.
• State statutes. Several states have their own telephone-solicitation laws with their own damages, independent of the federal autodialer question.
• Carrier registration. A2P 10DLC is a carrier gate, not a law. A clean autodialer answer does not get your messages delivered if the registration is rejected, and a carrier approval is not a legal ruling either.

In other words, the autodialer definition is the piece people fixate on and often the piece that applies least to modern list-based texting. The parts that actually require day-to-day discipline — opt-outs, send windows, suppression, how you talk to people — live in the operator’s playbook, and the channel-level question of whether you should be texting from your business number at all is separate again.

Why this is the easiest rule to get right

Of everything in this area of law, the autodialer question is the one most under your control, because it is decided by how you build, not by how a court later characterizes a recipient. You cannot control whether a number a business published turns out to be treated as residential. You can control, completely, whether anything in your stack ever generates a phone number. Keep the two apart in your head and the autodialer analysis becomes almost mechanical.

That is why a serious outreach operation treats provenance as infrastructure. Every number is sourced from somewhere real — for us, that means businesses matched on live data and validated as working cell lines before anything sends — and the source and date ride along with the record. Not because the law requires a specific file format, but because when the autodialer question comes up, the answer you want is a documented fact rather than a story. Build so that nothing produces numbers, record where the ones you have came from, and the hardest-sounding word in the TCPA turns into the simplest one to answer.

Frequently asked questions