A Text Is a Call in Newark, Not in Chicago. Build the Venue Matrix Before Your AI Presses Send.

By Anna with Oppy

On August 21, a federal court in New Jersey looked at the Telephone Consumer Protection Act, a 1991 dictionary, and a real estate brand, and concluded that a text message can be a "telephone call" under the statute's Do-Not-Call private right of action.1 The case is Owen-Brooks v. Better Homes & Gardens Real Estate. The plaintiff said she registered her number with the National Do-Not-Call Registry in 2013 and began receiving real estate marketing texts around March 2025. The court let the claim proceed, and it let the franchisor stay in the case on an agency theory: the brand allegedly trained franchisees on marketing practices and profited from the resulting business.2

Five weeks earlier, the Seventh Circuit reached the opposite conclusion in Steidinger v. Blackstone Medical Services, holding unanimously that a text is not a "call" under the same subsection, because in 1991 a telephone was an instrument for reproducing sound at a distance.3

Same statute. Same subsection. Opposite answers, forty days apart. A text sent to a consumer in Chicago now sits under different federal law than the identical text sent to a consumer in Newark, and the case that created the split was not aimed at a tech platform or a telemarketer. It was aimed at a real estate brand.

If your AI employee texts prospects, your compliance posture just became a geography problem.

The Map Is the Policy Now

Most outreach programs are built on a single national rule set. One consent script, one suppression list, one quiet-hours table, one assumption about what a text legally is. That architecture made sense when courts broadly agreed with the FCC's position that texts are calls. Loper Bright ended the era of leaning on that deference, and the courts are now reading the statute for themselves, with mixed results.1

The practical consequence is uncomfortable. Your exposure no longer depends only on what your AI says. It depends on where the recipient's area code and address happen to land on a map of federal jurisdictions that disagree with each other.

On August 27, Troutman's Consumer Finance Podcast put the compliance instruction plainly. Stefanie Jackman's message, as the hosts summarized it: Steidinger is not a green light to abandon text message compliance controls. Section 227(b) liability, FCC enforcement authority, and a growing patchwork of state telemarketing statutes remain fully intact, and consent remains the most important risk management tool available.3

Read that again. The ruling that texts are not calls did not reduce your obligations. It relocated them.

The Industry Is Formalizing While You Read This

The courts are not the only institutions redrawing the map. On August 25, MISMO, the mortgage industry's standards body, launched two AI governance certifications at its Fall Summit: a FRAME Advisory Partner Certification for consultants who implement AI governance, and an AI Governance Certification that validates the controls behind AI-enabled mortgage technology, product by product, use case by use case.4

"For mid-sized mortgage companies that may not have extensive AI, risk and compliance resources, FRAME will be a lifeline."

Lindsy Gwozdz, SVP of Compliance and General Counsel at HMA Mortgage.4

A lifeline implies deep water. The same week, at MISMO's Credit Super Session, Equifax's Justin Demola raised the question of when an AI system begins "acting as a loan officer" and triggers licensing requirements.5 Experian's Susan Allen drew the line that should hang over every AI deployment in housing: "There's a big difference between accepting more risk as a way to approve more borrowers versus seeing risk differently, calculating it more effectively."5

And in the Senate, the AI AGENT Act of 2026, S.5051, introduced July 21, is pushing toward verifiable, task-bounded authorization records for AI agents that take consequential actions.6 The direction of travel is consistent across courts, standards bodies, and Congress: the question is no longer whether your AI acts, but whether you can prove what it was allowed to do, where, and under whose authority.

The Data Says Guardrails Decide Who Gets Paid

If governance sounds like a tax on speed, the largest fresh dataset on agentic AI says otherwise, with a twist worth memorizing. Salesforce's State of Agentic AI in the Enterprise, a double-blind survey of 2,025 agentic AI decision-makers across 20 countries, published August 27, found that organizations with lighter governance reached positive ROI faster, in 7.2 months versus 9.3 for heavier governance. But organizations with below-average governance were nearly twice as likely to discover an agent operating outside its parameters only after a consequential error: 32% versus 18%.7

Speed buys you two months. An undetected boundary violation buys you a deposition.

The same study found that the strongest predictors of agent success were clean, accessible data at the moment the agent acts and a tightly bounded use case, each cited by 36% of successful deployers. Model quality ranked near the bottom. Among organizations whose AI initiatives stalled or failed, 38% named stronger governance frameworks and escalation protocols as the thing they would do differently.7

"Every boardroom is asking whether it's moving fast enough. Two years into the agentic shift, the answer from the data is that the advantage was never in starting first; it's in starting deliberately."

Shibani Ahuja, SVP of Data & AI Strategy, Salesforce.7

Deliberate, in 2026, has a specific meaning for any business that texts consumers: your agent must know which legal universe each recipient lives in before it presses send.

Build the Venue Matrix

The answer to a circuit split is not a memo. It is a data structure. A Venue Matrix is a living table that maps every contact to the rules that govern messaging them, evaluated before every send, not once at list import.

Two design notes matter more than the schema. First, the matrix is read by machines and signed by humans. Every row carries a source and a review date, because "the AI thought it was fine" is not a defense posture, it is a headline. Second, venue follows the person's verified address, not the phone number's area code. In a country where 60.1% of home searches in the 100 largest metros now cross market lines, per Realtor.com's Q2 2026 Cross Market Demand Report, area codes have never been less probative of where a human actually lives.8 A 917 number belongs to a Manhattanite who moved to Columbus three years ago. Route your legal logic by residence, not by digits.

Three Oppies, One Gate

Do not hand one agent the keys to the dialer and the statute book. Split the work into three Oppies with deliberately incompatible powers.

The gate logic is deliberately boring:

Before any SMS or call to a contact:

1. Resolve venue_state from the verified address on file.
   If unknown or stale, default to the strictest applicable rule set.

2. Load the Venue Matrix row and its rule_source citations.
   If the row is unsigned or past its review date, BLOCK and escalate.

3. Check consent_scope against the message class.
   Transactional consent never authorizes marketing.
   If consent is ambiguous, do not ask again in a way that
   manufactures revocation; route to a human.

4. Apply text_is_call_dnc and state_mini_tcpa overlays.
   Where courts conflict, follow the plaintiff-friendly reading
   unless counsel has signed a venue-specific exception.

5. Apply quiet_hours_local and registry status.

6. Emit ALLOW, ALLOW_WITH_DISCLOSURE, or BLOCK with reason codes,
   matrix version, and citations. The Evidence Keeper logs all three.

This architecture also answers the franchisor problem in Owen-Brooks. The theory that kept the brand in the case was control: training franchisees on marketing and benefiting from the results.2 If your brand sets the outreach playbook, your brand owns the compliance of the plays. A shared matrix with per-franchisee gates turns that liability surface into an audit trail that shows exactly which rules were applied to which message.

The Source of Truth Stays Human

One more fresh voice deserves the floor, because it comes from an operator rather than a vendor. On August 28, ROOST Real Estate Co., a brokerage and property management firm in Ohio and Florida, published its AI Manifesto and a companion podcast episode. Founder Chris McAllister's framing is the cleanest summary of where the industry is heading:9

"AI does not execute. It doesn't make the call. It doesn't have the hard conversation... or take responsibility when something goes wrong."

And on systems of record: "If AppFolio says one thing and AI says another, AppFolio wins, period."9

That is the correct instinct, and it generalizes. Your consent ledger is the source of truth for permission. Your Venue Matrix is the source of truth for jurisdiction. Your counsel is the source of truth for what the rows should say. The Oppies enforce, record, and escalate. They never originate law, and they never get the final word on an ambiguous consent.

McAllister's manifesto adds a second line that belongs in every AI policy in residential services: "AI does not fix poor habits. It magnifies them."9 A sloppy suppression process, automated, becomes a sloppy suppression process at scale, with timestamps.

Why This Is the Build, Not the Warning

It would be easy to read the August docket as a reason to slow down. The better reading is that the industry just received a specification. Courts are telling you that venue matters. MISMO is telling you that governance will be certified, product by product.4 Salesforce's data is telling you that bounded scope and clean records are what separate the 30% in production from the pilots that stall.7 And the AI AGENT Act is telling you that task-bounded authorization records are the shape of coming federal expectations.6

A brokerage, lender, title company, or property manager that builds the Venue Matrix now owns something its competitors cannot buy off a shelf: proof, per message, that its AI knew the law of the recipient's land before it spoke.

The courts split the map. Build the agent that reads it.

References