AI Follow-Up Replies: The 5 Cases Your System Must Handle

“Take me off your list.”

Every team running automated follow-up gets that reply eventually, and what happens next is not a matter of taste. For commercial email, the FTC’s CAN-SPAM compliance guide gives you ten business days to honor an opt-out request (FTC). On the calls-and-texts side, the FCC’s TCPA revocation rule, effective April 11, 2025, holds that a consumer may revoke consent in any reasonable manner, so words nobody put on a keyword list still count, with its own ten-business-day deadline (BCLP summary). Different channel, different rule — and the safe way to read an emailed “stop” anyway.

Tools in this category get graded on the message they write, almost never on the reply that comes back. That is backwards. The outbound message is a guess about someone. The reply is the one moment they tell you the answer.

Five replies, and only one of them is a conversation

Sort a month of replies and they fall into five buckets. The boring ones are the expensive ones.

A sixth thing arrives at the same address and is not a reply at all: the out-of-office auto-responder. Read as engagement, it starts a conversation with a mail server.

Four of the five outcomes are a stop. Follow-up automation sold on send volume is sold on the column that matters least.

The hard part is not classifying. It is seeing.

This is the question to ask a vendor, and it is not the one on the comparison chart: what can your system actually read?

There is a real difference between knowing an inbound email exists and having the text of it, and plenty of integrations get the first without the second — a good reason to know exactly what syncs into your CRM and what doesn’t. A system holding only the metadata can pause a sequence. It cannot tell “take me off your list” from “can you send the disclosures?”, and those two need opposite responses inside the same afternoon.

So the audit question is not “does it handle replies.” Every tool says yes. It is whether the words reached it, and what it did with them.

A four-message test you can run this afternoon

Put a personal address on a test contact record, then reply to one of your own automated messages four times, one behavior per reply.

1. “Please take me off your list.” An hour later: did the sequence stop, and is the stop recorded somewhere a different campaign would see it? Two campaigns each keeping a private list is what produces the angry second email.
2. “Wrong person.” Did anything at all change on the record, or did the next scheduled message go out anyway?
3. An out-of-office auto-responder. Anything that answers it conversationally is answering a robot, and will keep doing it.
4. A real question — “what are the taxes on that one?” Did anyone answer inside the day, and did the agent ever see it was asked?

Run all four before you scale sends. Tests 2 and 3 are the ones that come back empty, and neither shows up in an open rate.

The reply policy worth writing down

Most teams document a follow-up cadence and never document a reply policy, which is odd: the cadence is the part the software already knows. Five lines is enough:

• What counts as a stop. Write the obvious words down, then the sentence saying the list is not exhaustive. Your keyword list is not the standard.
• Where the stop is recorded. One place, readable by every system that can send. A stop that lives inside one campaign is not a stop.
• Who is told, and how fast. An agent hearing about an opt-out a week later is how a contact gets re-added by a well-meaning human.
• What happens to a reply nobody can classify. Pause outreach and put it in front of a person. Silence is the wrong answer; guessing is worse.
• How long you keep their own words. In a dispute, the sentence they typed is the evidence. Your summary of it is not.

There is a business case under the compliance case. A 2021 replication study in the Journal of Interactive Marketing found that building trust has a greater effect than reducing privacy concern on consumers’ willingness to participate in information markets (Swani et al., 2021). It studied consumers and marketers generally, not real estate follow-up, and it never measured a botched opt-out. It does point a direction: trust is the asset, and honoring a “no” on the first ask is a cheap way to keep it.

What Ace does with a reply

Ace Autopilot is the outreach layer that comes with Ace Trove. Once an account turns it on and graduates a workflow out of review mode, it sends without a person reading each message first, which makes the return path the whole job. The companion piece on the leads Autopilot refuses to touch covers who never gets written to at all.

When reply routing is on for an account, automated email from Ace Autopilot goes out carrying a reply address of ours, one per contact. Only the return path changes. Then, in order:

• The sender is verified first. Mail from an address that is not on the contact record never changes consent state and never starts a conversation; it is kept, and a copy goes to the assigned agent. A colleague forwarding a thread cannot opt someone out, and cannot restart a conversation on their behalf.
• The quoted history is stripped, so what gets read and stored is the part the person typed, not our own message bounced back.
• An opt-out is honored immediately. Automated email to that contact is suppressed, an exclude tag goes onto the record in Follow Up Boss, a plain-language note lands on their timeline saying what was said and what was done, and the assigned agent is alerted. Tags your team already uses — unsubscribed, Do Not Email, DNC — stop Autopilot from picking the contact up in the first place, before any of this comes up.
• “Wrong person,” “we moved,” and “not interested” each stop the run, put a note on the Follow Up Boss timeline, and alert the agent. The move alert says out loud that Ace will not follow up and that a referral is the agent’s call to make, not the software’s.
• Auto-replies are stored and ignored. No stop, no answer, no alert. The person has not spoken.
• Anything else is treated as a real reply. Autopilot answers it, the reply itself goes onto the Follow Up Boss timeline as a note, and a copy is forwarded to the agent’s inbox with the contact set as the reply address — so hitting reply takes the conversation over, and Ace steps back.
• A reply Ace cannot read raises a named alert rather than sitting quiet, and outreach to that contact stays paused. A reconcile sweep runs every ten minutes and re-checks open runs against Follow Up Boss, as the backstop.

That list is deliberately boring. Reply handling is plumbing, and plumbing is judged on whether it ever surprises you.

Three limits worth knowing before you trust any of it

• The classifier is narrow on purpose. It favors precision over recall: it would rather miss an unusual phrasing than stop a live conversation by mistake. An idiosyncratic no — “y’all can quit hitting me up” — may be read as a real reply and answered instead of stopping the run. The agent still gets the copy, and a clearer no does stop it. It is not a promise that every wording is caught on the first pass.
• Routing the reply to us changes where it lands. It does not appear natively in the agent’s own mail thread; it arrives as a forwarded copy with the contact as the reply address. That is the price of the software holding the text at all.
• None of this is legal advice. Those deadlines are federal floors; your state, MLS and brokerage may be stricter, and texting carries its own rules. Our communication compliance FAQ and AI compliance guidelines are starting points, not counsel.

Frequently asked questions

Does a reply to an automated email count as an opt-out request?

Treat it as one. The FTC’s CAN-SPAM guide names sending a reply email as one of only two steps you may require of someone opting out. For calls and texts, the FCC’s revocation rule goes further and says consent may be revoked in any reasonable manner rather than only through a method you designate. Practically: plan for sentences, not just clicks on an unsubscribe link. A system that can only count link clicks will miss the ways people actually say it.

How fast do I have to stop sending after someone opts out?

Both give ten business days — CAN-SPAM for commercial email, the FCC’s rule for calls and texts. That is a ceiling, not a target. Move faster anyway: a sequence sends on its own schedule, and every message after a “no” is a separate act the recipient experiences as being ignored.

What should happen when a lead replies to an AI email with a real question?

It gets answered the same day, and the assigned agent finds out it was asked. Those are two requirements, not one. Plenty of setups do the first and never the second, and the agent walks into the next call blind to a conversation that happened in their name. The fuller argument is in our piece on keeping automated follow-up human.

The one thing worth doing today

Send yourself test 2 — the “wrong person” reply — against whatever you have running now, and watch the contact record for an hour. Not the vendor page. The record. Whatever you find is the real answer, and auditing what an AI follow-up tool really did is the next piece to read.

One question worth arguing about with your team: when a lead replies “not interested” to an automated email, how long should a human wait before ever reaching out again — or should they?