← All resources
consentpatient-engagementcompliancesmscrm

One word, five places: what a patient's STOP actually has to touch

A texted STOP is not one checkbox. It has to reach a brand-level suppression ledger, the workspace permission record, the member’s preferences, the voice channel and your CRM. Here is what to ask a vendor, and what to check in your own stack.

GGravity Rail TeamHealthcare AI Specialists·7 min read

A patient texts STOP at 9:40 on a Tuesday night. The platform answers. Then what?

If you are the person who has to account for that opt-out — in a vendor review, a security questionnaire, or your own compliance file — the useful question is not whether your platform “honors opt-outs.” Everyone says that. The useful question is where the state lives, who can lift it, and what else it stops.

We build one of these platforms, so we are not a neutral source. The questions below are the ones we would ask any vendor, including us.

Not every opt-out is the keyword

A patient who types “please stop texting me” has asked for the same thing in every way that matters to them, and has not sent the word your system listens for.

A platform that only acts on STOP will report a lower opt-out rate than it has. More importantly, that patient is still reachable: the next reminder, check-in, or payment nudge goes out to someone who asked it to stop. Whether that becomes a complaint, a charge, or a finding depends on your jurisdiction and your luck.

Ask: what does the system do with a stop request phrased in words? In ours, a Workflow that binds our stop-recording tool acts on the sentence itself: it writes an agent-inferred suppression against the brand behind the number, which blocks the next text, ends the conversation as an unsubscribe, and is queued for human review. The same reply still tells the person to send STOP, and the record does not wait for that second message. A Workflow that does not bind the tool only answers, so ask which one you are looking at. The generic answer — “that’s handled by our AI” — is not an answer about state.

STOP is not a checkbox. It is at least three writes

A well-built platform does not flip one field. When a STOP arrives, ours:

  • turns SMS and WhatsApp off on the member record;
  • records a suppression row against each A2P brand the receiving number resolves to, or against the number and sending number when no brand resolves;
  • fires an event when the member’s preference actually changes, so automations, CRM sync and the audit trail can see the change.

Each of those is a different scope. The member record is per workspace. The brand ledger is per sending brand and outlives a workspace setting. The event is what makes the change visible to everything downstream.

Ask: where is the suppression recorded, and against what — the phone number, the sending brand, the workspace, or the member? If the answer stops at “on the contact record,” ask what happens when the same person exists in two workspaces, or when an import re-adds the number.

The layer that outranks the practice

Consent for automated texting is layered, and the order matters. In ours, the brand suppression ledger is read first and unconditionally: a STOP recorded against a brand cannot be lifted by a practice-level permission, because the practice is not the party that received the refusal. Workspace contact permissions come next, then the member’s own preference as a fallback.

Unreadable state should stop the send, not permit it. Consent checks fail closed; a platform that guesses “probably fine” is making a compliance decision on your behalf.

Ask: can a practice-level setting override a carrier or brand opt-out? If yes, that is your finding, regardless of how good the rest of the system is.

The check that has to run twice

Consent is usually checked when a campaign is built. The problem is the gap between building and sending: a campaign can sit in a send queue for minutes or hours, and a patient can opt out in that time.

In ours, the check runs again at the provider boundary, immediately before the message leaves for the carrier. A STOP that lands during the wait still blocks the send.

Ask: what happens to a message that is already queued when a STOP arrives? “It was checked when it was scheduled” is a finding, not an answer.

The stop that has to stop the calls

This is the one that surprises people. A patient who texts STOP has asked for quiet, not for one fewer channel. In our platform, a texted STOP also blocks automated AI voice calls.

The reverse is narrower: a later START lifts the texted stop for SMS — including the voice block that the STOP created — and does not silently re-enable WhatsApp. A patient who opted out of two channels and opted back into one should not be re-subscribed to both.

Ask: is opt-out one state across channels, or one setting per channel? If it is per-channel, that is how a patient who asked for silence still gets a call.

Your CRM is the fifth place

If you run patient texting through a CRM or a care-management system alongside a patient-engagement platform, both hold some notion of “do not contact” — and each is allowed to overwrite the other only if you like surprises.

The safe pattern is to export the platform’s standing STOP state to a dedicated property — “Texted STOP in Gravity Rail,” not the CRM’s own opt-out field. A dedicated property says what it means: this person texted STOP to this system. It leaves the CRM’s own opt-out intact, which may have been set by a phone call, a portal preference, or a conversation a receptionist had this morning.

And the export has to survive a failed read. If the platform cannot determine the STOP state, it should skip that member’s write and retry later. Writing “false” because the lookup timed out is how a suppressed patient gets re-subscribed by a nightly sync.

Ask: in the CRM, how many fields claim to mean “do not contact”? For each one, who writes it, and who is allowed to clear it? If the platform’s value is mapped onto a field the CRM also owns, you have two writers and no single answer.

The six questions

QuestionAn answer that should worry you
Is a stop request phrased in words treated as a signal?“We only handle the keyword.”
Where is the suppression stored, and scoped to what?“On the contact record, in that workspace.”
Can a practice setting override a brand or carrier opt-out?“Yes, an admin can.”
Is consent re-checked after a message is queued?“Once, when the campaign is built.”
Does one opt-out stop every channel, including voice?“Each channel has its own settings.”
What happens when the sync cannot read opt-out state?“It writes the last known value.”

Why this is worth a review, not a glance

None of these mechanisms is exotic. That is exactly the problem: each piece is easy to build in isolation, and the gaps between them are where the message that should not have gone out comes from. The patient experience of a failed opt-out is not a compliance statistic; it is a person who asked to be left alone and was not.

If you are reviewing a patient-engagement vendor, ask for the write path of a single STOP message, end to end: which records change, in what order, what can lift them, and what evidence you get afterward. A vendor who can walk you through it has thought about this. A vendor who answers with a checkbox has not.

If you use Gravity Rail and want the export configured in your CRM, contact us and we will walk through it with you.

See what your team could automate.

Book a walkthrough, or explore the platform that runs healthcare Agents across voice, SMS, email, and web — with human handoffs and a reviewable history.