The Court Said a Text Is Not a Call. Build Your AI Agent Like It Is.
By Anna with Oppy
On July 14, a unanimous Seventh Circuit panel ruled in Steidinger v. Blackstone Medical Services that a text message is not a "telephone call" under the TCPA's Do-Not-Call private right of action. One of the most popular theories fueling text-message class actions just lost its footing in Illinois, Indiana, and Wisconsin.
Cue the celebration in every sales bullpen from Chicago to Milwaukee. Now put the champagne down.
The ruling is real, it is narrow, and it is a trap for anyone who reads only the headline. If you run outbound texting for a brokerage, a lender, a title agency, an insurance office, a property management firm, or the dental practice that texts recall reminders at 8:47 pm, the smart move this week is counterintuitive: build your AI agent as if the court ruled the other way.
What the Court Actually Said
The plaintiffs got marketing texts about home sleep tests that kept arriving after they replied STOP and joined the National Do-Not-Call Registry. They sued under Section 227(c)(5), which lets a consumer recover for "more than one telephone call within any 12-month period" in violation of DNC rules.
Judge Kirsch, writing for the panel, went full textualist. In 1991, when Congress wrote the statute, a "call" meant communication by sound. A text does not reproduce sound. Congress said "call" in the private-suit provision and "call or message" elsewhere in the same statute, and that variation has to mean something. Dismissal affirmed. The Ecommerce Innovation Alliance's same-day analysis called it a leading authority on the question and a natural vehicle for eventual Supreme Court review.
Even the panel could not resist noting that repeated unwanted texts "are undoubtedly a nuisance." A win with an asterisk the size of a courthouse.
Why This Changes Almost Nothing for You
Here is the part the celebratory LinkedIn posts will skip.
| What survived Steidinger | What it means for your outreach |
|---|---|
| Section 227(b) | Consent rules for autodialed and prerecorded communications still apply to texts nationwide |
| FCC enforcement | The agency's authority to police spam texts is expressly undisturbed |
| State mini-TCPAs | Florida, Oklahoma, Washington and others keep their own private rights of action |
| Every other circuit | Outside IL, IN, and WI, courts are still split, and plaintiffs pick the forum |
| Statutory damages | $500 to $1,500 per message under surviving theories, per Hunton Andrews Kurth |
One ruling, one circuit, one subsection. Your lead list does not check ZIP codes against circuit boundaries, and neither does the plaintiff's bar. Hunton's compliance guidance, published the same day as the ruling, notes a recent uptick in plaintiffs testing the bounds of "reasonable" opt-out language and a surge in quiet-hours suits. The hunting season is open. Only the ammunition changed.
The How-To: An AI Agent That Treats Every Text Like a Call
The practical question is not whether to text. Speed to lead is still the whole game. The question is whether the system doing your texting can prove, message by message, that it behaved. Here is the four-layer build, whether you assemble it from scratch or launch it as an Oppy in an afternoon.
Layer 1: The consent ledger. Telemarketing texts require prior express written consent, and dual-purpose messages that mix information with promotion count as telemarketing. Your agent's first instruction should be blunt: never initiate an outbound message unless a consent record exists with a timestamp, a source, and the scope of what was agreed to. Purchased lists without documented consent are not leads. They are exhibits.
Layer 2: The seven-word reflex. The FCC treats seven reply words as per se revocations: stop, quit, end, revoke, opt out, cancel, unsubscribe. Your agent must process these instantly and also catch the long tail, because "please don't text me anymore, I sold the house" revokes consent just as surely as STOP. This is where a conversational agent quietly outperforms a drip campaign. A keyword filter matches strings; an agent that actually reads understands intent, confirms the opt-out politely, and writes it to the suppression list before the next campaign fires.
Layer 3: The clock and the scrub. Solicitations are restricted before 8:00 am and after 9:00 pm in the recipient's time zone, not yours. Give your agent standing orders: queue anything outside the window, scrub against the National DNC Registry and the Reassigned Numbers Database on a schedule, and maintain the internal suppression list as a living document. A reassigned number means your carefully obtained consent now belongs to a stranger with a lawyer's phone number in their contacts.
Layer 4: The audit trail. BCG's new piece on scaling agents, also published July 14, warns that unmanaged "agent sprawl" turns experimentation into thousands of ungoverned agents, and finds only about 15% of companies have the governance backbone in place. The fix at any scale is the same: every message your agent sends or receives gets logged with content, consent basis, and timestamp. When a demand letter arrives, the difference between a nuisance and a settlement is whether you can produce that record in minutes.
The Trust Curve Runs Both Ways
Addy AI CEO Michael Vandi told HousingWire this week that AI adoption is "a trust curve rather than a learning curve." People learn the tools quickly; what takes time is trusting them with real responsibility. He added that organizations "are beginning to hand over entire responsibilities to AI agents," which changes where humans spend their time.
The same curve applies to the person on the other end of the message. A lead who says stop and is instantly heard trusts you more, not less. Compliance and conversion are not opposing forces. They are the same discipline wearing different name tags.
There is an economic argument too. OpenAI's guidance on managing AI investments in the agentic era, published the same day as everything else in this unusually busy news cycle, urges leaders to measure "useful work per dollar" and track cost per accepted outcome rather than chasing the cheapest tool. Run that math on outreach: the cheapest texting blast that produces one certified class action has the worst unit economics in your entire stack. A governed agent that books three extra appointments a week while generating its own audit trail is the outcome you actually wanted to buy.
The Bottom Line
Steidinger will be cited, appealed around, and possibly headed for the Supreme Court. None of that timeline helps you this quarter. What helps you this quarter is an AI employee with consent verification, a seven-word reflex, a time-zone clock, and a memory that would satisfy a federal judge.
The court said a text is not a call. Build like it is, and you will never need to care who wins the circuit split.
Sources: Steidinger v. Blackstone Medical Services, No. 25-2398 (7th Cir. July 14, 2026) · Ecommerce Innovation Alliance analysis · Hunton Andrews Kurth, TCPA Compliance for Marketing Campaigns (National Law Review, July 14, 2026) · BCG, Scaling Agents Without Creating Your Next Tech Legacy (July 14, 2026) · HousingWire, Michael Vandi on mortgage AI (July 14, 2026) · OpenAI, How to manage AI investments in the agentic era (July 14, 2026)