Most "AI in real estate" is a summary machine. That's not what agents actually need.
The AI tools flooding our industry right now mostly do one thing well: they summarize. Summarize a listing description. Summarize a comparative market analysis. Summarize an inbox thread. Rewrite a follow-up email in a friendlier tone. Useful, but not transformative — because the summary is where the real work starts, not where it ends.
Between the two of us, we've written and re-written AAR forms as a designated broker, and we've watched agents at every experience level try to reason their way through those forms in the middle of a live transaction. Neither of those problems is a summary problem. Neither one is well served by an assistant that reads a document back to you in a friendlier voice.
What "AI in real estate" mostly is right now
Look under the hood of most of the AI features currently marketed to agents and brokerages, and the shape is the same: a general-purpose language model wrapped around whatever the vendor could scrape or ingest, tuned to be pleasant. It writes listing copy. It suggests subject lines. It transcribes voicemails. It's genuinely helpful on the edges of the day.
But it doesn't know what a Cure Period Notice is, when to use one, or how it differs from an Amendment. It doesn't know that line 28 of the Additional Clause Addendum is the operative language for making earnest money non-refundable. It doesn't know that the SPDS obligation reactivates when a house comes back on the market after a failed escrow. And when it tries to answer questions in that domain, it hallucinates plausibly-worded answers that a broker has to spend time correcting. That's not helpful; it's a liability.
What the next step looks like
The version of AI that actually changes the daily broker workflow reads the brokerage's compliance corpus — the current AAR form library, Arizona statutes (Title 32, Title 33, A.A.C. Chapter 28), the brokerage's own policy manual, the designated broker's curated guidance — and then acts on that knowledge. Two capabilities, working together.
The first is retrieval-grounded answering: every answer traces back to a specific line of a specific form, statute, or policy document. Not "here's a paragraph about earnest money" but "here's line 28 of the ACA, quoted verbatim, with a link to the form itself and an explanation of how it interacts with the RRPC." No hallucination, because the model isn't making it up; it's telling you what the corpus already says.
The second is structured document generation. When an agent describes the deal — "seller credits buyer $3,000 for termite treatment, extend COE from 8/28 to 9/11, buyer originated" — the AI doesn't reply with a summary of what an addendum is. It produces the addendum itself: the right form, the right fields, the right parties in the right slots, checkboxes checked, ready for the file. Same corpus, different mode.
Why the two pieces need to work together
A chatbot that answers compliance questions accurately but can't take any action is a research tool. It saves time on the front end of a decision, but the broker still has to open the form, transfer the answer into the right fields, and produce the artifact by hand.
A document generator that fills forms but isn't grounded in a real corpus is a liability. It'll happily produce a filled form based on plausible-sounding defaults that don't match Arizona practice, and now the broker has an artifact in the file that reads confident but is quietly wrong.
Both together — grounded knowledge, structured output — is what turns AI from a productivity gadget into a compliance layer. That's the leap that matters.
What this changes for designated brokers
Arizona brokers carry real supervisory weight. A.R.S. § 32-2151.01 puts the burden of policy, training, and record-keeping squarely on the designated broker. Most brokerages we've talked to know they're behind on that burden, and none of them think adding another AI chatbot to their agents' toolbelt solves it.
Grounded AI, applied honestly, actually strengthens the supervisory posture. Every AI-produced answer traces to a source the broker can audit. Every AI-filled form is deterministic given the described inputs. Every question and answer is logged and reviewable. The audit trail is stronger than what most agents produce manually today, because the AI's work is inspectable in a way that a hand-drafted addendum isn't.
The interesting question isn't whether AI can chat about real estate. It's whether AI can do the work — and whether we're going to hold the tools our industry uses to that standard.
— Sarah Richardson (CEO) & Mike McGowan (CRO), reTEQ
