I have been using AI tools in my CRE workflow for 18 months. Some of them have changed how I work. Some of them have wasted my time. Most of the AI in CRE conversation comes from consultants and vendors. This is what it looks like from a working broker.
Here is what is real and what is hype as of mid-2026.
What works: research and synthesis
The biggest practical win has been research speed. Looking up zoning. Pulling lease comparable summaries from public sources. Summarizing long emails into action items. Drafting first-pass market memos that I then edit by hand.
I use Claude and ChatGPT roughly daily for these tasks. The output is usually 70 percent right. The 30 percent that needs editing is faster to fix than writing from scratch.
What works: structured document review
Reading a lease draft against a checklist of terms to look for. Comparing two LOIs side by side. Flagging clauses that vary from a standard. These are well-defined tasks where the AI does the tedious read and surfaces the points I actually need to think about.
I do not let the AI make legal calls. I let it surface what my attorney and I should discuss. There is a difference.
What works: drafting
First draft of a market update. First draft of a proposal cover note. First draft of a landlord outreach email. These are tasks where the structure is roughly known and only the specifics change.
The AI gets me to 60 percent of a draft fast. I bring it to 100 percent. The total time is less than writing from blank. The quality is the same or better because I am editing instead of facing the blank page.
What does not work: cold outreach automation
AI-generated cold emails are easy to spot. Recipients know. The response rate on AI-generated outreach is approaching zero in my experience. The recipients filter them the way they used to filter generic mass mail.
I have written separately about what cold outreach actually requires. Specificity beats automation.
What does not work: market predictions
AI tools that promise to predict rent movements or vacancy trends from historical data are not reliable for NYC retail. The local market is too idiosyncratic. Foot traffic patterns change. Subway routing changes. Building conversion drives sudden shifts. Models trained on aggregate data miss these.
Human judgment about specific blocks beats AI projections about citywide averages for every decision I actually make.
What does not work: client meetings
AI cannot have the conversation. Clients want a human who has walked the block, knows the landlord, has done the deal. AI summaries of those things are not the same as having them.
I have not seen any AI tool that meaningfully replaces broker-to-client conversations. The augmentation tools that help me prepare for those conversations are useful. The replacement tools are not.
What I am watching
Two areas where AI could matter more in the next year. Building data extraction from unstructured sources. Pulling specific data from building records, court filings, and public documents at scale. The current tools do this poorly. Better ones may emerge.
And voice agents for routine call screening. I do not use one yet. The technology is close to ready. The trust is not. I will probably try one within the year.
The broker who wins with AI
Not the broker who automates the most. The broker who uses AI to do more of the work humans cannot easily do at scale, while keeping the human work that humans uniquely do. Relationships. Judgment. Reading a room. Knowing which landlord to call first.
That broker spends less time on summaries and drafts. More time on conversations and walks. The output of those conversations and walks is the part of the job that pays.