The TCPA Has Two Buckets - A Debt-Relief Robocall Case Just Showed Why That Matters for AI Voice.
A federal court in Michigan just handed operators a clean teaching example of how the TCPA actually works. And everyone's favorite TCPA plaintiff, Mark Dobronski, was standing right in the middle of it.
In Dobronski v. Credit Swag Ventures, Inc. d/b/a Credit & Debt, the Eastern District of Michigan split the case in two. Some of Dobronski's claims survived a motion to dismiss. Others did not. The line the court drew between them is the single most important distinction in TCPA compliance, and most operators get it wrong.
Here is the short version. You can beat the Do-Not-Call registry and still lose on the robocall. Solving one bucket does nothing for the other.
Let me lay out the two buckets, walk through what happened, and give you the Monday-morning fix.
The TCPA's two buckets, in plain English
The TCPA is not one rule. It is two different regimes living in the same statute, protecting against two different things.
Bucket one is § 227(b). This is the robocall bucket. It governs calls made with an automatic telephone dialing system or an artificial or prerecorded voice. The FCC confirmed in its February 2024 Declaratory Ruling, CG Docket No. 23-362, that AI-generated and cloned voices count as "artificial or prerecorded voice." So yes, your AI voice agent lives in this bucket. The § 227(b) question is simple: did you have consent to make the call the way you made it?
Bucket two is § 227(c), the Do-Not-Call bucket. This is the national registry, the internal do-not-call list, and the caller-ID transmission rules that ride alongside them. But, and this is the part that decides cases, § 227(c) only bites when the call is a "telephone solicitation" or "telemarketing."
That word carries the whole load. Under § 227(a)(4), a "telephone solicitation" is "the initiation of a telephone call or message for the purpose of encouraging the purchase or rental of, or investment in, property, goods, or services." The FCC's telemarketing definition at 47 C.F.R. § 64.1200(f) runs on the same engine: the purpose has to be encouraging a purchase.
No purchase pitch, no solicitation. No solicitation, no DNC claim. That is the whole ballgame in Dobronski.
What happened in Dobronski v. Credit Swag Ventures
Dobronski pled a fact pattern that reads like a compliance nightmare:
- A number on the National DNC registry since 2003.
- Spoofed caller IDs.
- An artificial voice named "Jessica," calling from "Debt Relief Group."
- A live transfer.
- A Credit & Debt closer who picked up already holding his intake data.
On paper, that is a stack. Spoofing, a robocall, a long-registered number, a warm transfer to a company that somehow already had his information. Graded as a consumer complaint, it looks bad for the defendants.
But the motion to dismiss did not grade the vibe. It graded the elements. And the elements sorted these claims into two different buckets.
What survived, and what died
The artificial-voice claim survived
The court found Dobronski had sufficiently pled that Credit & Debt used an artificial or prerecorded voice. It helps when an employee admits to using prerecorded voice. That admission, plus the pled facts, carried the § 227(b) claim past the pleading stage.
Read that again. The robocall claim survived without any finding that the calls were trying to sell anything. § 227(b) does not care about your sales pitch. It cares whether you used a regulated technology to reach someone without the consent that technology requires.
The DNC and caller-ID claims did not
Here is where the split happens. The court held that Dobronski "did not sufficiently allege that the calls at issue were 'telephone solicitations' within the meaning of the TCPA." There was no encouragement in the calls to pay for anything. At most, the court said, the calls were "encouragement to use their free services."
Free is not buy.
The caller-ID claim under 47 C.F.R. § 64.1601(e) went down for the same reason. That rule rides on the telemarketing predicate. No telephone solicitation, no telemarketing, no § 64.1601(e) violation, spoofed number or not. And the conclusory complaint language, the "designed to induce purchase" boilerplate, did not save it. A court reads the pled facts, not the labels you paste over them.
No solicitation. No DNC claim. Hence, the split.
The individual closer walked too
There was also an individual defendant, the closer who took the transfer. The court let him out. Joining a multi-step transfer train and placing a later callback is not "initiating" the robocall within the meaning of the statute. The "who initiated the call" analysis traces back to the FCC's DISH Network ruling (In re DISH Network, LLC, 28 FCC Rcd 6574 (2013)), which is where direct and vicarious liability for calls gets sorted out. The treble-damages and forward injunctive-relief requests were dismissed at the pleading stage as well.
If your compliance model assumes everyone who touches a call is equally on the hook, this is a useful correction. Initiator liability is narrower than that. (Whether that helps you or hurts you depends on which end of the transfer you are sitting on.)
Why this matters for AI voice, lead gen, and debt relief
Here is the part most operators are misreading. They treat DNC compliance as the finish line. Scrub the list, honor opt-outs, transmit clean caller ID, and assume the TCPA is handled.
It is not handled. DNC compliance is bucket two. Your AI voice agent lives in bucket one.
Dobronski is the cleanest illustration you will get. The defendants may well beat the DNC claim, because the calls were not solicitations. But the artificial-voice claim, the one that actually turns on consent, survived and is heading toward discovery. The "free offer" framing that dodged the solicitation problem did nothing for the robocall problem.
Translated for 2026: building your script around a free consult, a free membership, or a "see if you qualify" screen may keep you out of the telephone-solicitation definition. It may pull you out of the DNC and caller-ID rules. But, it does not give you a pass to run an AI voice or a prerecorded voice without the consent § 227(b) requires. Those calls need consent regardless of whether they are trying to sell anything.
That is the trap. Operators engineer the pitch to escape "solicitation," then assume they have engineered their way out of the entire TCPA. They have solved the wrong bucket.
One honest caveat. This is one district court ruling on a motion to dismiss, not a circuit-level rule, and the case is early. But, the two-bucket logic it runs on is bedrock TCPA, not a Michigan quirk. You should not bet your calling program on the idea that a court somewhere will collapse the two buckets back into one.
What to do this week
- Separate your two buckets on paper. List every calling program you run. For each one, answer two questions independently: do I have § 227(b) consent for the technology I am using, and is this call a telephone solicitation that triggers DNC and caller-ID rules? Different questions, different answers.
- Audit your AI voice and prerecorded scripts for consent, not just for DNC posture. If an AI voice or a prerecorded message opens the call, the consent question is live whether or not you are selling anything. Pull the consent record for every campaign that uses one.
- Stop treating "free offer" language as a TCPA shield. It may narrow your solicitation exposure. It does nothing for § 227(b). If your playbook calls a free offer a non-solicitation, make sure your consent file tells the same story, and make sure that framing does not lull you into skipping consent.
- Map your transfer trains and callbacks against the initiator question. If you take warm transfers or place callbacks, know where you sit in the "who initiated the call" analysis, and document accordingly.
- Write down your consent architecture. Do not improvise it. The reason § 227(b) claims survive motions to dismiss is that consent is a record, and most operators cannot produce the record on demand. Build the record before a plaintiff like Dobronski builds the complaint.
The bottom line
The two-bucket problem looks handled right up until a complaint lands and you realize you solved half of it. If you are running AI voice or prerecorded calls and you are not certain your consent holds up independent of your DNC posture, that is worth a look before a plaintiff's lawyer takes the look for you.
