Consent in parent communication: service vs marketing, one-to-one delivery and STOP requests

A deputy head at a private school is sitting with two message drafts. The first: "A reminder for your school tour appointment tomorrow at 14:00." The second: "Enrollment for the new term has opened; here are the early-registration terms." Both look like they are going to the same parent list. But these are two different jobs: the first carries out something the parent asked for themselves, the second is a promotion the institution started on its own initiative. If you leave that difference to the memory of whoever prepares the list, sooner or later it gets mixed up.

This article explains how parent communication consent is built on the product side: why service communication and marketing are kept apart, why consent is stored not per person but per person × channel × purpose, why a bulk announcement goes out as one-to-one messages, and what exactly stops when a parent writes "STOP".

One frame first: this article is not legal advice, it does not interpret regulation, and it offers no compliance guarantee. Everything that touches regulation is subject to current rules and to the privacy notice your institution publishes. What we describe is the operational discipline that keeps that frame standing in daily work once your institution has defined it.

This is the third piece of the school cluster. You will find how the prospective-parent flow is built in the school admissions funnel and how response speed is measured in first response time; for the AI-training side, the series' main article, education institution AI training, is a good starting point.

Quick answer

Service communication or marketing

In the system, the purpose of a message takes one of two values: service or marketing. There is no gray third category, and that is a deliberate simplification. Opening a third box means, in practice, that every debatable message ends up in it.

Here is what the split looks like day to day:

The split is not a loose label; it lives on the record itself, and at send time the gate reads that value to decide. The practical difference shows up here: a parent who has opted out of promotional sends still receives the reminder for their own appointment. Without that separation, a parent who said "do not send me promotions" would also have forgotten their own enrollment meeting the following week.

How a given message is classified at your institution is defined in your privacy notice and your internal policy. The system does not make that call for you; it applies the call you made, consistently, on every send.

The shape of a consent record: person × channel × purpose

Storing consent per person is easy and wrong. A parent may be perfectly happy to get a newsletter by email, while the same newsletter landing on their WhatsApp feels intrusive. Those are two different preferences and they need to be two separate rows.

In inTusell a consent record carries three pieces of information together:

• Person: which parent record.
• Channel: which communication channel — WhatsApp, email, web chat, and so on.
• Purpose: service or marketing.

On top of that sits the state of the record: granted or revoked. So the same parent can have independent rows such as "email + marketing = granted" and "WhatsApp + marketing = revoked".

Two behavioral differences matter:

• On the marketing side, if your institution's policy requires explicit opt-in, a send does not happen when there is no granted row. Silence is not treated as approval.
• On the service side, consent is not sought as a precondition; what blocks here is a revocation. If a parent has explicitly closed service notifications on that channel too, the send stops.

This gate is not applied on a single screen but on every path a message can leave through — campaign sends, scheduled messages, follow-up reminders, win-back flows, and template sends. It is not one list checked from one place — it is a check repeated at every exit. The reason is simple: in systems guarded by a single door, a second door eventually opens and nobody notices.

How a bulk announcement goes out

The job schools hesitate over most is sending a bulk announcement to the parent list. Three separate mechanisms carry that job.

Targeting. You narrow the recipient list in two ways. The first is a segment based on interest intensity: hot, warm, and cold records are separated by the contact's lead score. The second is tags: you mark parents with your own tags and then send to that tag. This is exactly where sending by school level comes in — if you have tagged parents with level tags such as "kindergarten", "primary", or "middle school", you can send to that tag. The level field on the contact record is informational; if you have not done the tagging, that field does not filter the list on its own. Knowing this distinction avoids the "I thought I sent by level" surprise.

One-to-one delivery. Once the list is prepared, the message does not go out as a single broadcast. The recipient list is walked one by one and each parent receives a separate message inside their own conversation. Parents see neither each other's numbers nor how many people the message went to. What the parent sees in their inbox is a message the school wrote to them — not something that landed inside an announcement group. A single run takes at most 500 recipients; for larger lists you split the campaign into more than one run.

The consent gate. During the send, the list is filtered once more, and the filtering happens per channel. If the same parent is granted on email but revoked on WhatsApp, they stay on the list for the email send and drop out of the WhatsApp send. This is not a step left to the memory of whoever prepared the list; it runs inside the sending path itself.

Children's data: what is not collected in chat

By the nature of school and nursery conversations, the chat eventually turns to the child. There is a deliberate constraint on the product side here: the fields the AI can extract from a conversation on its own are separated from the fields it cannot.

Fields such as source, school level, target class, decision timing, transport need, and scholarship interest can be extracted from a conversation. By contrast, student name and student birth year are not extractable fields. On the nursery and kindergarten side the same constraint covers child name, child birth year, and any special agenda for the meeting. These fields do not stay empty — but they are filled in only by an authorized user, with the parent's confirmation. The difference is this: a name that came up in a conversation is not a name someone asked to have recorded.

The second constraint concerns health and allergy detail. The AI does not ask for a student's health information on its own. If a parent shares it unprompted, the AI does not record the detail and leaves the subject to the face-to-face enrollment meeting. This is not ignoring the information; it is making sure the information is collected in the right place, with the right person, in the right context.

A third detail concerns someone else's data. The referrer field is defined as a field that requires confirmation: when one parent gives another parent's name, that information is not written to the record silently — it passes through human confirmation first.

On top of all this, in school and nursery setups the AI's reply waits in draft mode by default. On WhatsApp, Instagram DM, Messenger and Telegram the AI does not send its prepared answer straight to the parent; it holds it in an approval queue, and it goes out once an authorized person approves it. Web chat and email do not enter that queue — a question from the chat bubble on your site is answered directly by the AI. You can turn this default off from the panel, but the reason it arrives switched on for schools is exactly the sensitivity described in this section.

STOP and START: what happens when a parent writes

A parent writing "STOP" is a clear request, and it should have a clear consequence.

When STOP (or the local equivalent) is written, marketing permission is revoked. If the channel the message arrived on can be determined, the revocation is applied on that channel. If the channel cannot be matched, the revocation is not kept narrow — it is widened instead: marketing is closed on all known channels. Service notifications continue; the reminder for the parent's own appointment still goes out. Leaving a parent who said "no promotions" unaware of their own meeting would not be honoring their request; it would be punishing them for it.

When START is written, the behavior is deliberately more cautious. Permission is granted only on the channel the message came in on. If the channel cannot be matched, the system does not invent a new permission; it only reopens rows that were previously revoked. In other words, "START" on a channel that never had a permission does not create one.

This asymmetry is not an accident. Revocation is read broadly, re-granting narrowly. If you have to err on one side, erring on the parent's side is the right one.

Duplicate records and merging: revoked wins

Duplicate records are unavoidable in schools. The same parent writes from WhatsApp one year, comes in through a web form the next, and spells their name differently. At some point those two records need to be merged.

There is one rule for merging: the revoked record wins. If either record carries a revocation for a channel and purpose, the merged record counts as revoked for that channel and purpose too. Merging does not bring back a permission that was lost. This makes it impossible to quietly gain consent during a data-cleanup pass.

The second rule concerns who owns the decision: merging is not a background job that runs on its own. A user decides from the panel that two records belong to the same person and starts the merge. There is no mechanism that looks at a similarity score and merges records by itself, because two wrongly merged parent records are among the hardest mistakes to undo.

Five steps to make it stick

Whether the mechanisms work depends on a few habits settling in on the institution's side. Here is what we see in the field:

1. Write the text first, then set up the system. Your privacy notice and internal policy determine which message you place on which side. The system does not make that call for you; it applies the call you made. Do not start promotional sends before completing this step with your adviser.
2. Match the place you collect consent to the channel. A permission taken on a web form and one taken on WhatsApp are not the same row. Mark which channel a record was taken for, so that no "which channel did they consent on" argument comes up later.
3. Establish tag discipline early. Treat distinctions such as level, campus, and area of interest as tags. When the day of the bulk announcement comes, you send to a ready tag instead of trying to filter the list by hand.
4. Review duplicate records at regular intervals. Merging is a human decision and the revoked record wins. Cleaning up before enrollment season starts is better than writing to the same parent twice during it.
5. Leave draft mode on for the first weeks. In school and nursery setups the AI's reply on the messaging channels (WhatsApp, Instagram DM, Messenger, Telegram) already waits in an approval queue. Reviewing that queue once a day settles both the tone and the boundaries quickly; you can switch it off from the panel when you feel ready.

These five steps are not a compliance project but an operating habit. Settling them in a quiet period is far easier than trying to build them once enrollment season has started.

What it is not

Let us draw the boundary clearly:

• It is not legal advice. Neither this article nor the product interprets regulation; your institution and your advisers define your compliance frame.
• It offers no compliance guarantee. The mechanisms described provide discipline; none of them stands in for a compliance certificate.
• It does not produce privacy notices or consent forms. You prepare and publish your own texts; the system only records and applies the outcome.
• It is not an official student records system. Grades, report cards, attendance, and official registration are outside the product's scope.
• It is not a group-messaging tool. Sends go out one-to-one; parents do not see each other's information.
• It is not a tool that makes sending without permission possible. The consent gate runs inside the sending paths; it is not a check bolted on afterwards.

Frequently asked questions

Does an appointment reminder count as marketing?

In the system an appointment reminder is recorded as service communication: it carries out something the parent asked for themselves. A promotional newsletter, a campaign announcement, and a win-back message sit on the marketing side. How that split is interpreted at your institution is subject to current regulation and to the privacy notice you publish; the system only keeps the distinction on record and filters accordingly.

What happens if a parent grants consent on email and writes STOP on WhatsApp?

Because a consent record is kept as person, channel, and purpose together, the revocation on WhatsApp closes only that channel; the email permission stays standing. At send time the gate filters separately for each channel, so the same parent remains on the email list while being left out of WhatsApp promotions.

Can parents see each other in a bulk announcement?

No. A bulk announcement does not go out as a group message; the recipient list is walked one by one and each parent receives a separate message inside their own conversation. Parents see neither each other's numbers nor how many people the message went to. A single run sends to at most 500 recipients.

Can I send by school level (kindergarten, primary, middle)?

Targeting works through tags. If you tag parents with level tags such as kindergarten, primary, or middle school, you can send to that tag. The level field on the contact record is informational and is not on its own a targeting criterion, so do not skip the tagging step.

Does the AI collect the student's name or health information in chat?

No. The student name and birth-year fields are not among the fields the AI can extract from a conversation on its own; they are filled in only by an authorized user, with the parent's confirmation. Allergy and health detail is not collected in chat either: if a parent shares it unprompted, the AI does not record the detail and leaves the subject to the enrollment meeting.

If the same parent exists twice, which consent applies?

When you merge duplicate records, the revoked record wins. If either record carries a revocation, the merged record counts as revoked too, and a permission lost that way does not come back. The merge decision is not automatic; a user makes it from the panel.

Next step

Consent discipline is not a standalone module but a layer that touches every step of parent communication. You will find the stages a prospective parent passes through from first contact to final enrollment in the school admissions funnel, and how long the first reply takes and how that is measured in first response time. To set what the AI will and will not say, the series' main article, education institution AI training, is the right starting point; the channel setup is covered in Instagram DM and WhatsApp automation.

You can see the full solution for education institutions on the education page and the packages on the pricing page; for common questions see the support center, and for the whole series the blog.

If you would like to see how it works on your own parent list, write to us from Request a demo or at hello@intusell.com; we will set up your consent policy together and send the first bulk announcement side by side.