Generative AI is software that produces new text, images, or code by predicting what should come next, based on patterns it learned from an enormous body of examples. That is the whole definition. When you type a request into ChatGPT, Claude, Gemini, or Microsoft Copilot and it writes back a letter of intent, a lease summary, or a market paragraph, it is not looking up an answer in a database and it is not thinking the way you do. It is generating the most statistically likely continuation of your request, one word at a time, fast enough that it feels like a conversation. For a commercial real estate professional, the useful version of this explanation is not the computer science. It is what the tool does to the specific documents on your desk — and, just as important, the two places it will let you down if you trust it blindly. This piece covers both.
A word on where this sits before the definition. Generative AI has gone from a curiosity to a line item that commercial real estate leaders now plan around — Deloitte’s Commercial Real Estate Outlook has tracked a steady rise in executives who expect these tools to reshape how their firms operate. The instinct at a small firm is to react by buying something. The better first move is to understand what the tool is well enough to decide what to hand it, which is the entire argument of the small CRE firm manifesto and the reason the CRE AI training playbook starts with fluency rather than software. This article is the vocabulary that makes those next steps make sense.
What Generative AI Actually Is
Strip away the hype and generative AI is a very good pattern-completion engine. It was trained by reading a staggering volume of text — books, articles, code, transcripts — and learning, statistically, which words tend to follow which. When you give it a prompt, it uses those learned patterns to generate a response that fits. The “generative” part is the point: unlike a search engine that retrieves an existing page, or a spreadsheet that computes a fixed formula, it produces something new each time.
The kind of generative AI that matters for your work is the large language model, or LLM — the technology inside ChatGPT, Claude, Gemini, and Microsoft Copilot. An LLM works with language, which is exactly what most CRE work is made of. A letter of intent is language. A lease is language. A market write-up, a prospecting email, a listing description, the summary of a call — all language. That is why a tool most people first met as a chatbot turns out to be so useful to a brokerage: your day is full of documents, and this is a machine built for documents.
You interact with it by prompting — writing a plain-English instruction of what you want. “Summarize the key rent and term details in this lease.” “Draft a cold email to an industrial tenant whose lease expires next year.” “Rewrite this listing description to lead with the loading-dock count.” There is no query language and no menu of features to learn. The interface is a sentence, which is the reason a 15-person firm with no technical staff can adopt it at all.
Generative AI vs the AI You Already Use
You have used AI for years without calling it that, and the distinction between that older AI and the generative kind is the single most useful thing to understand — because it tells you what to trust the tool for and what to double-check.
The AI already embedded in the tools you use every day is mostly analytical AI: it recognizes patterns, ranks, predicts, and calculates against structured data. CoStar’s comps, an automated valuation estimate, the recommendation of which lead to call next, your email spam filter — these look at numbers and categories and produce a number or a ranking. They are built to be precise about known quantities.
Generative AI is different in kind. It produces fluent, plausible language rather than a verified calculation. Ask it to summarize a lease and it excels, because summarizing is a language task. Ask it to compute the effective rent across a 10-year term with three months of free rent and a 3% annual escalator, and it will confidently return a number that may be wrong, because arithmetic is not what it is built to do — it is predicting what a plausible answer looks like, not calculating one.
| Analytical AI (the older kind) | Generative AI (the new kind) | |
|---|---|---|
| What it does | Recognizes, ranks, predicts, calculates | Produces new text, images, or code |
| Good at | Precise answers on structured data | Drafting, summarizing, rewriting, explaining |
| CRE examples | Automated valuations, comps ranking, lead scoring | LOIs, lease summaries, market write-ups, emails |
| Trust it for | The number, when the data is clean | The first draft, which you then verify |
Hold that line in your head and most of the confusion clears. Generative AI is a brilliant first-draft writer and an unreliable calculator. Use it where the work is language and you win time. Reach for it where the work is exact math on money and you check every figure — or you keep that step in the spreadsheet where it belongs.
What It Does to a Broker’s Actual Workday
The generic explanations of generative AI list use cases like “writing” and “images.” That is true and useless. Here is what it does to five documents a commercial real estate professional produces every week.
Letters of intent. An LOI is mostly boilerplate wrapped around a handful of deal-specific terms. Generative AI drafts the boilerplate in seconds and slots in the terms you specify, turning a 40-minute task into a five-minute one — draft, then read carefully and adjust. It does not replace your judgment on the terms; it removes the typing.
Lease abstraction. A 90-page lease has maybe two pages of information you actually need: the rent schedule, the term and options, the CAM structure, the assignment and default clauses. Ask the tool to pull those into a summary and it does in a minute what used to take a careful hour. You still verify the numbers against the source, but you start from a structured summary instead of a blank page.
Market write-ups. Feed it the vacancy, absorption, and rent figures you have gathered and it will draft a clean market paragraph in your firm’s register. The facts have to be yours and correct; the prose is the part it handles.
Listing and marketing copy. A property description, a broker’s opinion of value narrative, an email blast to your buyer list — all of it drafts faster and reads better when a strong writer produces the first version and you edit for accuracy and voice.
Prospecting and correspondence. The tenant whose lease expires next year, the owner you have been meaning to call, the follow-up you keep postponing — generative AI drafts a specific, personalized message in the time it takes to describe the recipient. You are approving and sending, not composing from zero.
The pattern across all five is the same. The tool is fastest and safest where the work is drafting and summarizing language, and where a human reads the output before it goes anywhere. What a firm looks like once this habit is in place across the whole team — not one enthusiast, everyone — is described in the anatomy of an AI-fluent brokerage.
The Two Things It Gets Wrong
An honest explanation of generative AI has to cover where it fails, because for commercial real estate the failures are not footnotes — they are the whole reason to use it carefully rather than not at all.
The first is hallucination. Because the tool generates plausible-sounding text rather than retrieving verified fact, it will sometimes state something false with complete confidence — invent a statute, misremember a figure, fabricate a citation. It is not lying; it has no concept of true or false, only of likely. The practical consequence is a simple rule: generative AI writes the draft, a human owns the facts. Never send a number, a legal term, or a market claim it produced without checking it against a source you trust. Used this way, hallucination is a manageable quirk. Ignored, it is how a wrong cap rate ends up in front of a client.
The second is confidentiality, and it is the objection most brokers raise first. Your work runs on NDA-bound offering memoranda, seller financials, and rent rolls with tenant names — can that go into an AI tool at all? The short answer is that it can, with the right handling, but the handling is not optional. There are two separate questions: whether the vendor trains its model on what you type (business and enterprise tiers of the major AI products state they do not, by default), and whether your NDA permits disclosing the material to a third party at all. Getting this right is a firm-level decision, not a per-broker guess, and it is worked out in full in the confidentiality framework for safe AI use with client deal data. The one-line version: put the team on a paid business tier, and classify each document before it goes near a text box.
Neither failure is a reason to avoid the tool. Both are reasons to use it with a rule, which is exactly the difference between a firm that gains from generative AI and one that gets burned by it.
Why This Is a Training Problem Before a Software Problem
Notice what the two failure modes have in common: they are not solved by buying a better tool. They are solved by a person who understands the tool well enough to know what to hand it, what to double-check, and what to keep out. That understanding is the actual asset, and it is why generative AI adoption at a small firm is a training question before it is a purchasing one.
This is the point most firms get backwards. They buy a subscription, hand it to the team, and wait for productivity that never arrives, because a powerful tool given to people who do not know its shape produces a few cautious experiments and then quiet disuse. The firms that win spend their first effort on fluency — teaching the team to prompt well, to recognize a hallucination, to classify a document — and only then worry about whether they need anything beyond the general-purpose tools everyone can already buy. The full argument for that sequence is the case for training before tooling, and the ordered ninety-day version of it is the CRE AI training playbook.
The strategic prize is real. Generative AI is one of the few tools that lets a lean firm produce work at a volume and polish that used to require headcount a small shop cannot afford — which is precisely how a 15-person brokerage starts to out-operate a much larger competitor. But the prize goes to the firms whose people understand the tool, not the ones with the biggest software budget. The definition you just read is the first step toward being one of them.
Where to Start
You do not need to become a technologist to put generative AI to work, and you do not need to guess which tasks at your firm are the right first ones. A free AI-readiness assessment is a short working session that looks at your firm’s actual mix of brokerage, management, and acquisitions work, identifies the handful of document-heavy tasks where generative AI pays off fastest, and points you at the right tool and tier for your existing Microsoft or Google setup. Book a free AI-readiness assessment and you will leave knowing exactly where to start — with a plan matched to your deals, not a generic tool you have to figure out alone.
Frequently Asked Questions
What is generative AI in simple terms?
Generative AI is software that creates new content — text, images, or code — by predicting what should come next based on patterns it learned from a huge body of examples. When you ask ChatGPT or Claude to draft an email or summarize a lease, it is generating the most likely response one word at a time, rather than looking up an answer in a database. For commercial real estate, the useful mental model is a very fast, very fluent first-draft writer: excellent at producing language, and something whose facts you always verify before you rely on them.
How is generative AI different from the AI already in my tools?
The AI already in tools like CoStar or an automated valuation model is analytical — it ranks, predicts, and calculates against structured data to produce a precise answer. Generative AI produces new, fluent language instead. The practical difference is what you trust each for: analytical AI for the number when the data is clean, generative AI for the first draft of a document you then edit. Generative AI is a strong writer and an unreliable calculator, which is why you check any figure it produces.
What can generative AI do for a commercial real estate firm?
It accelerates the language-heavy parts of the job: drafting letters of intent, summarizing long leases into the few terms that matter, writing market paragraphs from your data, producing listing and marketing copy, and drafting prospecting emails. In each case it produces a strong first version in seconds that a person edits and approves. It does not replace judgment on deal terms or verify facts for you — it removes the typing and the blank-page problem so your time goes to the work that needs a broker.
Can generative AI make mistakes?
Yes. Because it generates plausible text rather than retrieving verified fact, it will sometimes state something false with confidence — a wrong figure, an invented statute, a fabricated citation. This is called hallucination. It is not a sign the tool is broken; it reflects how the technology works. The rule that manages it is simple: let generative AI write the draft, but have a human own the facts. Never send a number, legal term, or market claim it produced without checking it against a trusted source.
Is it safe to use generative AI with confidential deal data?
It can be, with the right handling. Two questions matter: whether the vendor trains its model on your input, and whether your NDA allows disclosing the material to a third party at all. The business and enterprise tiers of the major AI products state they do not train on business customer content by default, but the NDA question is separate and governs whether you may hand the document over regardless. The safe approach is to put the team on a paid business tier and classify each document before pasting it — public and internal material freely, NDA-bound or personal data redacted first or kept out.
Which generative AI tool should a small CRE firm use?
For most small firms the answer is one of the general-purpose business tiers — ChatGPT, Claude, Gemini, or Microsoft Copilot — chosen to match your existing setup. A firm already on Microsoft 365 will land naturally on Copilot; a firm living in Google Workspace on Gemini. Standardizing the whole team on one tool matters more than which one you pick, because it makes training and data rules enforceable. Specialized proptech tools can come later, once you know which tasks justify them.
Do I need to know how to code to use generative AI?
No. You interact with generative AI by writing plain-English instructions, called prompts — there is no code, query language, or menu of features to master. Writing a good prompt is a learnable skill closer to briefing a capable assistant than to programming: say what you want, give context, and specify the format. That low barrier is exactly why a small firm with no technical staff can adopt these tools, and why the first investment is training people to prompt well rather than hiring anyone.
Will generative AI replace commercial real estate brokers?
No, but it changes what a broker’s time is spent on. Generative AI is good at the routine language work — drafting, summarizing, first passes — and poor at the things that define the job: judgment on deal terms, relationships, negotiation, market instinct, and accountability for the facts. The realistic outcome is that brokers who use these tools handle more volume with the same headcount and spend less time on paperwork, while the client-facing and judgment-heavy work stays firmly human. The competitive risk is not the machine; it is the broker down the street who has learned to use it.
How does a small firm actually get started with generative AI?
Start with understanding, not a purchase. Learn the distinction between what the tool drafts well and what it should not be trusted to calculate, put the team on one business-tier tool, and agree on a simple rule for what deal data can go into it. Then pick two or three document-heavy tasks — LOIs, lease summaries, market write-ups — and build the habit there before expanding. The first investment that pays off is training the team to fluency, not buying specialized software; the tools most firms need are the general-purpose ones they can already subscribe to today.
Arthur Wandzel
