A multifamily chatbot is a scripted tool that answers renter questions on your property website or by text. It works by matching what someone types against a list of things it was built to recognize. When it finds a match it gives the scripted answer. When it doesn't, it either guesses, loops, or asks the renter to fill out a form.

That last part is why so many leasing teams say they tried AI and it didn't work.

What a leasing chatbot actually does

Under the hood, most of these are decision trees. Someone sat down and mapped the questions renters ask most: pet policy, parking, office hours, application fees, availability. Each one gets a scripted answer and a few expected phrasings.

Build enough branches and it feels like a conversation. It isn't. It's a very fast lookup table with a friendly tone applied on top.

For a narrow set of jobs, that's completely fine. Pet policy at 11pm doesn't need reasoning. It needs an answer.

Where they break

The renter phrases it like a person

"Do you take dogs" is in the script. "My golden retriever is 80 pounds, is that going to be a problem" is the same question and a different string of text. A decision tree sees no match and falls back to something generic. The renter reads that as being ignored.

The question has more than one part

"I need a two bedroom by October 1st under $2,200 and I have a co-signer." That's four constraints in one sentence. Scripted tools answer whichever piece they recognized and drop the rest.

The conversation moves channels

A renter texts on Saturday, emails Monday, calls Tuesday. To most tools that's three unrelated people. The renter explains their situation three times and starts drawing conclusions about how your property is run.

The question is one it shouldn't answer

Ask a scripted tool whether a building is family friendly, or what the neighborhood is like, and you find out fast whether anyone thought about Fair Housing when they wrote the branches. A tool that improvises here is a real risk, not a rough edge.

Nothing happens next

The renter asks two questions, gets two answers, and leaves. The tool has no reason to follow up. It answered what it was asked. The lead is now sitting in a log somewhere waiting for a person to notice it.

Why the fatigue is earned

A lot of operators bought one of these between 2019 and 2023, and a lot of them had the same experience. Response times looked great on the report. Leasing didn't move. Onsite teams stopped trusting it and started answering everything manually again, which meant the property was paying for a tool and doing the work anyway.

So when someone brings an AI leasing tool into a meeting now, the room is skeptical. That skepticism is a reasonable read of the evidence, not resistance to change.

The problem is that it makes the current generation harder to evaluate honestly, because the demos look similar from the outside.

What actually changed

Scripted tools match text against a list. Reasoning models work differently. They read the whole message, hold the context of everything said earlier, and work out a response rather than retrieve one.

Practically, that means the 80-pound golden retriever question and the "do you take dogs" question land the same way, because the model understands them as the same question. It means four constraints in one sentence get four answers. It means a conversation that started on text and moved to voice is still one conversation.

It also means the tool can recognize what it shouldn't handle. A scripted tool has no concept of its own limits, because it has no concept of anything. A reasoning model can be built to hand off cleanly, which turns out to matter more than almost anything else.

This is worth asking about directly on any demo. Which model, running where, and who's accountable for renter data. "Proprietary AI" isn't an answer. We put that question, and eight others, in a guide to comparing AI leasing tools.

How to tell from the outside

You don't need to know anything technical. Three tests, five minutes:

• Ask the same question in an odd way. Not the phrasing on the demo script. Your phrasing.
• Put three things in one message. See how many come back answered.
• Switch channels mid-conversation. Then see if you have to repeat yourself.

A scripted tool fails at least two of those. It'll fail politely, which is what makes this worth testing rather than assuming.

What this means for your team

The version of this that worries people is a tool that talks to renters instead of the leasing team. That's not what good looks like, and honestly it isn't what the technology is best at.

What it's good at is the volume that never needed a person: the same eleven questions, at all hours, at every property. What it's bad at, and should be bad at, is the conversation where a renter is anxious about their application or working through something complicated. That's leasing, and leasing is a skill.

The design question isn't how much the AI can take. It's how cleanly it gets out of the way.

Where Vida fits

Vivian is an AI leasing agent, not a chatbot. She answers across text, email, and voice as one conversation, qualifies leads, books tours, and follows up on the ones that go quiet. Response is 100% and averages around 38 seconds. She resolves 84.5% of conversations on her own.

The other 15% is the part we care most about. When Vivian brings in a person, Leasing Intelligence gives that leasing agent the renter's history, a summary of what's been discussed, and current property details in one view at the moment they take over.

Not AI instead of your team. AI + your team.

Frequently asked questions

Is a multifamily chatbot the same as an AI leasing agent?

No. A chatbot matches text against a script. An AI leasing agent reasons through the conversation and can act on it: qualify, book, follow up. We wrote about the difference between an assistant and an agent separately.

We already have a chatbot on our website. Is it worth replacing?

Depends what it's costing you. If it's answering basic questions after hours and nothing more was expected of it, it may be fine. If leads are going quiet after a conversation with it, that's the signal.

Do renters mind talking to AI?

Less than people assume. What they mind is a tool that clearly doesn't understand them, or having to repeat themselves. Being upfront that it's AI doesn't hurt conversion.

What about Fair Housing risk?

It's real either way and worth asking about specifically. Ask any vendor to show you live what happens when a renter asks about schools or neighborhood demographics.

How long does it take to get live?

Most operators pilot on a handful of properties, look at the numbers, then expand. Plan around a pilot rather than a portfolio rollout.

The short version

Chatbots break on the questions that matter because they were never reading, just matching. If your team stopped trusting one, they were right to. Test the next tool with your own phrasing before you believe the demo.

If you want to run those tests against a real conversation, we'll set it up. Bring the hard questions.