What do TCPA rules require for automated service texts?
The line that decides everything is whether the message sells anything. A dispatch notice and a discount code sit under different rules.
Answered in short
6 things that decide this
- 01A purely transactional text, such as a technician-on-the-way notice, needs prior express consent under 47 CFR 64.1200(a)(1), not the signed written kind.
- 02Signed written consent under 64.1200(a)(2) is triggered when a message includes or introduces an advertisement or constitutes telemarketing, so one promotional line changes which rule applies.
- 03Every revocation request made in any reasonable manner must be honoured within a reasonable time not to exceed ten business days from receipt.
- 04Stop, quit, end, revoke, opt out, cancel and unsubscribe are reasonable revocation methods per se, and you may not designate an exclusive way to opt out.
- 05The FCC's one-to-one consent rule was vacated by the Eleventh Circuit in January 2025 and never took effect, so it is not something you must comply with.
- 06Statutory damages run at five hundred dollars per violation, and a court may treble that for a willful or knowing violation.
The upsell that reclassified your whole message
Your appointment reminder is a service message. Someone in marketing adds "book your next clean and save 10%" to the template. That template now advertises the commercial availability of a service, which is what the regulation defines as an advertisement.
The consent you collected was for service notifications. The message you are now sending needs the signed written kind. Nothing in your code changed and your exposure did.
A review request has the same problem in a friendlier costume. So does a referral prompt attached to a completion notice. The FCC's own exemption language for certain notifications is explicit that they must not include telemarketing, solicitation or advertising content.
Which consent applies to which message
Checked 11 August 2026 against the current text of 47 CFR 64.1200 on eCFR. This is the classification question, not legal advice on your programme.
| The message you are sending | What it is under the rules | Consent needed |
|---|---|---|
| Your technician arrives between 2 and 4 | Transactional. No advertisement, no telemarketing. | Prior express consent |
| Your appointment is confirmed for Thursday | Transactional. | Prior express consent |
| Job done. Book your next clean and save 10% | Advertises commercial availability of a service. | Prior express written consent |
| How did we do? Leave us a review | Treat with care. Encouraging future business pulls toward telemarketing. | Take advice before sending on service consent |
| One text confirming you have opted out | Permitted, if it carries no marketing and is the only further message. | Presumed within existing consent if sent within five minutes |
Opt-out handling is an engineering requirement
The regulation says a called party may revoke consent using any reasonable method that clearly expresses a desire not to receive further messages. Seven specific words are reasonable per se, and you may not designate an exclusive means of revocation.
It goes further than a keyword match. Words outside that list count as valid revocation if a reasonable person would understand them as a request to stop. So a system recognising only STOP is not doing what the rule describes. A free-text message saying "please stop texting me" still has to reach a human or a broader classifier.
The clock is ten business days from receipt, whatever route the request arrived by. Opt-outs arriving by support inbox, web form and text reply must all reach the same suppression list. Most of the failures we see are integration failures rather than policy failures.
- 01A one-time confirmation reply is allowed, carrying no marketing, and is presumed within consent if sent within five minutes.
- 02Where you use a one-way texting protocol, the rules require you to disclose that clearly on each message and give alternative ways to revoke.
- Arrive anywhereReply, call, web form, inbox.
- Be recognisedNot just the seven keywords.
- Reach one listEvery channel writes to the same place.
- Confirm onceNo marketing, inside five minutes.
- Take effectWithin ten business days at the latest.
- Stay loggedTimestamped, because you may need to prove it.
The third station is where most builds fail. Two channels writing to two suppression lists produces a customer who opted out and kept receiving messages, which is exactly the fact pattern that gets litigated.
Messaging systems we have shipped
ZhoopZhoop
AI receptionist and parts procurement for a multi-branch auto repair business.
Read the case study →
Maidily
Integrated operations platform for residential cleaning businesses.
Read the case study →
Shift Link
AI workforce compliance and shift management for healthcare and logistics.
Read the case study →
Related questions
01Does an existing business relationship let us text customers?
It helps with solicitation and do-not-call rules. It is not a substitute for the written consent that advertising messages to a mobile number require. The established business relationship in 64.1200(f)(5) runs eighteen months from a purchase, or three months from an enquiry. A do-not-call request to you ends it at once.
02Our texts go out through a vendor. Does that move the risk?
Your consent records are still yours, and so is the decision about what the message says. A platform gives you delivery and keyword handling; it does not tell you whether your template became an advertisement when marketing edited it. We treat message classification as a review step in the build, not a setting.
03How much is a violation actually worth?
The statute provides five hundred dollars for each violation, or actual monetary loss if greater. A court may treble that where the violation was willful or knowing. Per message, across a whole list, is how these become large. That arithmetic is the reason the suppression list is worth engineering properly.
04Can an AI assistant send these messages?
It can, and the classification question does not change because a model wrote the text. If anything the risk rises, because a model given latitude can add a helpful upsell to a service message and reclassify it. We handle that with a refuse list and templates the model fills rather than composes freely.
Related
- Should an AI receptionist handle emergency calls →The other regulated edge of automated contact.
- Field service software →Where these messages are sent all day.
- Field service case studies →Systems we run for trades businesses.
- AI chatbot development →Assistants that message customers, built with a refuse list.
