It is trained on how your team actually talks, and it stops the moment someone replies or opts out.
Published February 9, 2026
Mostly, and the parts that make automated follow-up feel robotic are not the parts people expect. It is rarely the wording. It is the behaviour.
A message that reads perfectly well still feels like a machine if it arrives after you already replied, ignores what you said, or is the fourth identical nudge in a week. Fixing the tone without fixing those does not help.
Each of those is a state problem rather than a writing problem. The system does not know where the conversation is, so it sends the next scheduled thing regardless.
The first is that the sequence stops on any reply. Not pauses for review, stops. The moment somebody responds, the conversation is live and the scheduled messages are no longer relevant.
The second is that it never sends the same nudge twice. Each follow-up references what has already happened and adds something new, whether that is a floor plan, a newly available unit, or a different time that opened up.
The third is timing that respects the hour. Answering instantly at nine at night is welcome because the person just messaged you. Initiating a nudge at the same hour is not.
The agent is trained on how your team actually talks, from the greeting to the sign-off, and you hear it handle your scenarios before anything goes live. That is the point of the test calls during the build.
It is worth being deliberate here rather than accepting a default. Most stock automation copy is written to sound professional to everyone, which means it sounds like nobody. A short, plain message that matches how your leasing team actually writes will outperform something polished and generic.
If somebody asks whether they are talking to a person, it tells them the truth. That is deliberate, and in practice most people carry on, because the thing they wanted was an answer and they got one at eleven at night.
Trying to pass as human is the wrong goal anyway. It creates an expectation that breaks badly the first time the conversation goes somewhere the agent cannot follow, and it makes the handoff to a real person feel like a bait and switch rather than an escalation.
Long, discursive exchanges are still where a person is better, and the right design is one that recognises that quickly rather than grinding on. A good agent gets somebody from inquiry to booked tour, and gets out of the way when the conversation turns into something that needs judgement.
One detail that gets overlooked while everybody argues about wording. Message length signals more about who is writing than vocabulary does.
People text in short bursts. A four sentence paragraph arriving as a text message reads as marketing regardless of how friendly the words are, because no colleague has ever texted you like that. Matching the medium, short over text, fuller over email, is most of what makes an automated message feel like it came from a person.
During the build you hear it handle your scenarios, and it is worth knowing what to pay attention to rather than forming a general impression.
That last one is a common and fixable weakness. A system that finishes its sentence while the caller is already talking sounds like a machine no matter how well the sentence was written.
On the site
Keep reading