← All articles
AIPropTechTechnology

Generative AI in Real Estate: Past the Hype, Into the Workflow

C
Castellan Team
February 22, 2025 · 6 min read

Separating the demo from the workflow

Generative AI has had a strange couple of years in real estate. Every conference has a panel, every vendor has a roadmap slide, and every operator has seen a slick demo that generates a listing description in five seconds. The demos are impressive. The question operators actually care about is narrower and harder: where is generative AI doing real, repeatable work that moves a number on the P&L?

That is the line this piece walks. There is genuine value here, but it is not evenly distributed, and the highest-value uses are not the flashiest. The useful applications tend to be unglamorous, embedded in daily workflow, and measured in vacancy days and hours saved rather than novelty. Here is where the value is real, where it is still emerging, and where it remains mostly hype.

Where it is delivering real value today

Prospect communication at scale

The clearest, largest win is in handling the constant stream of prospect communication. Generative AI is genuinely good at natural conversation, which is exactly what leasing first response requires. An AI agent can answer an inbound call, respond to a Zillow inquiry, or reply to a prospect's text in seconds, with a coherent, on-brand conversation rather than a canned auto-reply.

This is not a demo. It is a workflow. Castellan's agents handle inbound leasing communications across phone, email, and SMS around the clock, qualifying prospects and booking showings end to end. The value shows up directly in vacancy, because Harvard Business Review's lead-response research ties sub-five-minute response to dramatically higher conversion, and an AI agent hits that window every time, including nights and weekends when offices are closed.

Drafting and summarizing routine text

Generative AI quietly saves real hours on the writing-heavy parts of the job. Listing descriptions, follow-up emails, renewal notices, and resident communications can be drafted in seconds and then reviewed by a human. Equally useful is the reverse: summarizing a long call transcript, a thread of resident messages, or a maintenance history into a few lines a manager can act on. These are small wins individually but large in aggregate across a portfolio.

Structured extraction from messy inputs

A lot of property data arrives as unstructured text: a Zillow inquiry email, a maintenance request typed in a hurry, an application note. Generative AI is well suited to pulling structured fields out of that mess, the prospect's move-in date, the unit they asked about, the nature of the repair, so the information lands in your systems clean and actionable instead of requiring a human to read and re-key it.

Where the value is still emerging

Maintenance triage and coordination

Applying conversational AI to maintenance is promising but earlier in maturity than leasing. The intake works well: an agent can take a request through any channel, ask clarifying questions, and gauge urgency. The harder part is the downstream coordination with vendors and the judgment about prioritization, which is improving but still benefits from human oversight on anything non-routine.

Owner and resident reporting

Generative AI can draft owner updates and resident notices, and that is genuinely useful. But reporting that involves judgment about what to highlight, or sensitive communications, still needs a human editor. The technology drafts well and decides poorly, so the value is in acceleration, not autonomy, here.

Where it is still mostly hype

It is worth naming the overstated parts, because believing the hype leads to disappointment and wasted spend.

The pattern behind what works

Look across the genuinely valuable applications and a pattern emerges. Generative AI delivers when the task is high-volume, language-based, and repeatable, with a human available for the exceptions. It struggles when the task is low-volume, high-stakes, and judgment-dependent.

That is why leasing communication is the standout. It is enormous in volume, entirely language-based, highly repeatable in its structure, and the rare exception, a complex negotiation, a sensitive situation, can be cleanly escalated to a person. The task fits the tool. By contrast, the applications still labeled hype are exactly the ones that are low-volume and judgment-heavy, where the tool fights its own nature.

Compliance is not optional

Any generative AI that talks to prospects inherits the same legal obligations a human leasing agent has. Fair housing law applies in full. An AI agent must avoid any line of questioning that touches protected classes, and in jurisdictions with source-of-income protections it cannot ask about or react to a prospect's use of a housing voucher.

The encouraging part is that a well-built agent makes compliance more consistent, not less, because it applies the same vetted conversation to every prospect without improvisation. But this only holds if compliance was designed in deliberately. Bolting a generic language model onto your phones without that discipline is how you generate a fair housing problem at scale. The technology is a tool, and like any tool it reflects how carefully it was built.

How to actually capture the value

For an operator trying to move past the hype, the path is concrete:

Generative AI in real estate is neither a revolution that replaces your team nor empty marketing. It is a genuinely useful tool that, applied to the right workflows, takes real work off your plate and recovers vacancy days you were losing. The operators winning with it are not chasing the flashiest demo. They are quietly putting it to work on the highest-volume task they have and measuring the result.

See Castellan respond in under 2 minutes

We'll map your real lead flow and show you exactly where the response gap is costing you leases.

Book a demo