CallataGuides

AI Text Messages: Consent and Carrier Rules

What consent you need before an AI agent texts customers, how carrier 10DLC rules apply to AI replies, and which AI texts are safe versus risky.

An AI agent can text customers who started the conversation or who consented to texts for that purpose, and it can't send marketing texts without marketing consent. On top of the law, US carriers require registered business numbers (10DLC) and expect the messages to match what you registered. The safest AI texting uses are replies to inbound texts and short confirmations during calls the customer agreed to.

This guide covers both layers: the law and the carrier rules. It's general information, not legal advice.

Two rulebooks apply

Layer Source What it requires
Federal law TCPA and FCC rules Consent for automated texts; marketing texts need written consent
State law Mini-TCPA laws (Florida and others) Often stricter consent and timing rules
Carrier rules CTIA guidelines and 10DLC registration Opt-in, clear identity, STOP handling, registered use cases

The carrier layer is the one small businesses feel first: unregistered or non-compliant traffic gets filtered or blocked, regardless of whether any law was broken.

What federal law says about texts

The FCC treats text messages as calls under the TCPA. After the Supreme Court's 2021 decision in Facebook v. Duguid, which narrowed what counts as an "automatic telephone dialing system," fewer texting platforms clearly fall under the federal autodialer rules. That doesn't mean consent is optional:

  • Carriers require consent regardless.
  • State laws, such as Florida's Telephone Solicitation Act, define automated systems more broadly.
  • Marketing texts carry the most risk and need express written consent in practice.

The FCC's 2024 proposal on AI-generated calls also included AI-generated texts, proposing that they disclose AI involvement. It hasn't been finalized as of this writing.

Which AI texts are low risk

AI text Why it's low risk
Reply to a customer's inbound text The customer started the conversation
Text sent during a call because the caller asked ("text me the address") Requested on the call
Appointment confirmation the customer booked Informational, expected
"We got your message, someone will call you" Informational reply to their contact

Which AI texts need care

AI text What you need
Follow-up offer days after a quote Marketing consent
Review request Consent; carriers often classify it as marketing
Re-engagement to past customers Marketing consent
Text to a number from a lead list Don't

Carrier expectations for AI texting

The CTIA Messaging Principles and Best Practices and 10DLC registration through The Campaign Registry shape what's allowed:

  • Registered campaign: your 10DLC campaign describes your message types, such as customer care, appointment reminders, or marketing. AI replies should fit what you registered.
  • Identify the business: messages should make clear who's texting.
  • Honor STOP: standard opt-out keywords must stop messages. See handling opt-outs on AI calls and texts.
  • Respond to HELP: with your business name and contact info.
  • No prohibited content: carriers restrict certain categories (for example, some lending, cannabis and third-party lead generation content).
  • Consistent content: if your campaign says "customer care," don't let an AI agent start sending promotions.

Registration involves a one-time brand and campaign vetting process. AI texting doesn't need a special registration, but your sample messages should reflect how AI replies look.

Writing AI texting instructions with compliance built in

When texting:
- Only reply to customers who texted us, or text a caller during
  a call if they ask.
- Keep replies short and factual. Use only the business facts.
- Never send offers, discounts or promotions.
- If someone texts STOP or asks to stop, don't reply.
- If someone asks if you're a person, say you're an AI assistant
  for [Business] and offer to have a team member follow up.

Good and bad AI texts, side by side

Situation Good AI text Problem AI text
Customer texts "are you open Saturday?" "Yes, we're open Saturday 9 to 1. – Ava, Corner Bike Shop" "Yes! And this weekend only, 20% off tune-ups!" (adds marketing)
Caller asks for the address during a call "Corner Bike Shop: 118 Main St. Parking is behind the building." Sending three more texts later with offers
Customer texts a question the facts don't cover "Good question. I'll have someone from the team text you back today." Guessing an answer
Customer texts "stop" (no reply beyond an allowed confirmation) "Sorry to see you go! Before you leave..."
Customer texts a complaint "Sorry about that. I've flagged this for the manager, who will reach out." Arguing or offering refunds the AI can't authorize

The good versions share three traits: they answer only what was asked, use only facts you've provided, and hand anything else to a person.

Texting during business hours vs after hours

Replies to inbound texts can go out any time, because the customer chose when to text. Texts your business starts, including AI follow-ups, should stay within normal hours. Several state laws restrict when solicitation texts can be sent, and even informational texts at 11 p.m. annoy people. See AI calling hours rules for the windows that apply to outbound contact.

Disclosure in texts

Even without a federal rule, saying who's texting avoids confusion. Options:

  • Sign AI texts with the agent name and business: "– Ava, [Business] assistant."
  • Answer honestly when asked if they're talking to a person. Utah's AI law, for example, requires disclosing generative AI when a consumer clearly asks. See AI disclosure laws for phone agents.

A starter policy for AI texting

  1. AI replies only to inbound texts and to callers who ask for a text.
  2. No marketing texts from AI agents.
  3. Marketing texts, if any, go only to people with written consent, sent by staff.
  4. Every AI-texted thread is visible to the team, and a human can take over at any time.
  5. Opt-outs are automatic and reviewed weekly.

How Callata handles AI texting

In Callata, an AI agent can be set to reply to inbound texts on your business number until someone on your team replies; it stays quiet for 12 hours after a teammate sends a message in that thread. AI replies are limited to short messages based on your business facts, and when a person needs to follow up, the thread is marked unread for your team. AI agents on calls can text the caller a confirmation, address or link only after the caller agrees. Messages with STOP, START or HELP keywords aren't answered by the AI, and opted-out numbers are blocked from texts.

Business texting requires carrier registration, which Callata files for a one-time $49 fee. Each AI text reply uses a quarter of an AI minute (about 6 cents at $0.25 per minute). Callata Office is $99 per month with five users included. Set up texting in Callata.

Frequently asked questions

Do I need consent for an AI to reply to a customer's text?

A reply to a text the customer sent you is a conversation they started, which carriers and courts generally treat as consent to respond about that topic. Marketing follow-ups need separate consent.

Do AI texts have to say they're from an AI?

No federal rule requires it today. The FCC proposed AI disclosure for texts in 2024, and some state laws require disclosure when asked. Identifying the business and offering a human is good practice.

Does 10DLC registration cover AI texting?

Your 10DLC campaign describes the kinds of messages you send. If AI agents send customer care or appointment texts, make sure your campaign description covers those uses.

Can an AI agent send marketing texts?

Only to people who gave consent for marketing texts. Carriers expect express written consent for marketing, and many state laws require it.