Your leasing inbox is full of people who aren't renting
Post a unit on a major listing site and the inquiries roll in. Some are real prospects ready to tour. Many are not. Mixed into every leasing inbox is a steady stream of noise: outright spam, phishing attempts, automated scrapers, recruiters pitching services, wholesalers, and tire-kickers who will never set foot in a unit.
The cost of this noise is sneaky because it's distributed. No single junk lead wastes much time. But across a portfolio, your team spends real hours every week reading, sorting, and sometimes replying to inquiries that were never going anywhere. Worse, the noise buries the signal. When a genuine prospect's email is sandwiched between ten junk ones, response time slips, and response time is the single biggest lever in leasing.
The answer is to filter the noise before it reaches a human, accurately enough that you trust it not to drop a real lead.
What junk actually looks like
Spam and junk in a leasing context aren't all the same, and good filtering treats them differently.
- Outright spam and phishing. Generic blasts, malicious links, payment scams. These should be filtered hard and never seen by your team.
- Bots and scrapers. Automated inquiries that follow templated patterns, often with mismatched or nonsensical details. No human on the other end.
- Solicitations. Vendors, recruiters, and service pitches dressed up as inquiries. Not prospects, but not malicious either.
- Tire-kickers. Actual humans who are curious but not in market: browsing for next year, asking about a unit they can't afford, or fishing for information with no intent to rent.
The first three categories you want gone. The fourth is trickier, because a tire-kicker today can be a prospect in three months, and you don't want to slam the door. Good filtering sorts rather than just deletes.
Deterministic checks catch the obvious cases
A lot of junk is obvious and can be caught with fast, deterministic rules before any heavier analysis runs. Known-bad senders, malicious link patterns, blatant template spam, and inquiries that fail basic sanity checks can be filtered immediately and cheaply.
This first layer matters because it's reliable and explainable. You know exactly why something was filtered, and the most clear-cut garbage never consumes another second of attention or analysis. It's the equivalent of a spam folder that actually works, tuned to the patterns that show up in leasing specifically.
But deterministic rules have a ceiling. They catch what looks obviously bad. They struggle with the inquiry that's well-written but pointless, or the message from an unknown sender that could go either way. That's where a smarter layer comes in.
Where judgment is needed, language models help
The hard cases are the ones that look plausible. A message from a sender you've never seen, phrased like a real inquiry, that may or may not be a genuine prospect. Rigid rules either let too much through or block real people. This is exactly the kind of ambiguous, language-heavy judgment that language models handle well.
A well-designed spam gate uses deterministic checks first, then falls back to an LLM for the unknowns. The model reads the actual content and weighs whether this reads like a person who wants to rent a home or like noise. It's looking at intent and coherence, not just keywords, so it can pass a real prospect with an unusual email address while catching a polished-looking solicitation.
The critical design principle is to bias toward caution on real people. The cost of filtering one genuine prospect is far higher than the cost of letting one borderline message through to a human. A good gate is aggressive on clear junk and conservative on ambiguous humans, and it logs its decisions so you can audit and tune them.
Don't let the filter become a fair housing problem
Filtering has to stay strictly about whether a message is genuine, never about who the prospect is. A spam gate sorts on legitimate, neutral signals: sender reputation, malicious patterns, coherence, and stated intent to rent. It must not infer or act on anything tied to a protected class.
This is a real line. "Is this a real prospect?" is a fair question. Anything that drifts toward filtering people based on how they write, their name, or assumptions about their background is not. A responsible system is built to evaluate the legitimacy of the inquiry, applied uniformly, and to escalate anything ambiguous to a human rather than quietly screening people out.
The payoff: a clean funnel that moves fast
When junk is filtered upstream, two things improve at once.
First, your team's time goes to real prospects. The leasing inbox becomes a list of people who actually want to rent, not a sorting chore. Agents spend their hours qualifying and showing, not triaging noise.
Second, and more importantly, your response time to genuine leads drops. When the AI handles the noise and surfaces only real inquiries, the prospect who emailed at 8 PM gets a fast, relevant reply instead of waiting behind a queue of garbage. Given that responding within minutes dramatically raises the odds of converting a lead, a clean funnel directly improves lease velocity.
Platforms like Castellan run this as the first stage of the inbound pipeline: a spam gate combining deterministic checks with an LLM fallback for unknown senders, so only legitimate prospects reach the qualification and scheduling steps. Your team never sees the bots. They just see the renters. That's the point of filtering, not to add a step, but to remove the dozens of pointless ones your team has been doing by hand.