In 2023, 5% of commercial real estate firms were piloting AI.
Today it’s 92%.
Only 5% say they’ve achieved most of what they set out to do.
Those three numbers come from JLL’s 2025 Global Real Estate Technology Survey, which polled more than a thousand senior decision-makers across sixteen markets. They tell you the interesting question isn’t whether AI arrived in real estate. It’s where it’s actually working.
In pre-development, five specific things changed in the last eighteen months. Not predictions – each one has a date on it.
1. The rules now change faster than anyone can check them
Picture a developer pricing a site in Los Angeles last spring.
California’s SB 79 was coming on 1 July 2026, allowing significantly taller housing by right near major transit. The maths worked. The offer made sense.
Then Los Angeles adopted two ordinances on 23 June, effective 30 June, that paused SB 79 citywide and replaced it with a narrower local programme covering 57 Opportunity Station Areas – pushing full implementation to roughly 2030.
The state law took effect eight days after the city opted out of it.
And this wasn’t a one-off. LA codified ED1 into the Municipal Code in December 2025 and replaced its entire transit incentive structure with CHIP in February 2025. Three structural changes to what a parcel allows, in one city, inside eighteen months.
What it means: a zoning answer has a shelf life now. The question is no longer “what does the code say” but “what does it say today, and which ordinance established that?” Platforms have had to move from static records to daily updates – ArchiWise refreshes from official sources every day – because anything slower is a pro forma built on last quarter’s rules.
2. The address bar stopped being the starting point
Every traditional tool asks the same question first: which parcel do you mean?
Assessor portals, city GIS, listing platforms – all of them need an address before they do anything. Which quietly means you can only evaluate what someone has already handed you, and everyone else got the same list.
The shift is to screening an entire market against a buy box. Lot size, use, value, ownership type, zoning, hazard exposure – stack the criteria, get back every parcel that matches, listed or not.
Scale is what makes this new rather than aspirational. ArchiWise covers 160 million parcels across 20,000 cities and 3,000 counties in all 50 states, representing over 90% of US development activity. At that coverage, “search the market” stops being a figure of speech.
What it means: your pipeline is no longer capped by how many people think to call you. We compared both approaches step by step in AI site selection vs manual parcel screening.
3. The bad news arrives before the offer instead of after it
Hazards used to be a due diligence problem – which really means they were an earnest money problem. Each one sat in a different agency portal, and nobody checks six portals for a site that might not survive the week.
That order has flipped. Flood, fire severity, wetlands, past fire footprints, earthquake faults, seismic shaking and methane now screen at the parcel in seconds, and as map layers across a whole corridor. The corridor view answers a question a single lookup can’t: is a clean site normal around here, or the exception?
The most expensive constraints are the ones zoning never mentions. In Los Angeles, methane mitigation lives in LAMC Division 71, Section 91.7101 – and the building department can withhold a permit until adequate protection is designed. Pipeline consultation zones of 660 to 1,000 feet apply wherever a local ordinance adopts PHMSA’s recommended practice. Not federally mandated, not a setback, but a real step that’s dramatically cheaper to absorb at screening than under contract.
What it means: the cheapest moment to walk away is during screening. More of what used to kill deals in month three now shows up in minute one. More on one of them in the methane hazard most developers miss.
4. Ownership became something you search, not something you look up
An assessor record answers exactly one question: who owns this? It needs the address first, and it tells you nothing about whether that owner would ever sell.
Now ownership is a filter. Hold period, portfolio size, linked debt, absentee status – plus a separate screen for pre-foreclosure, tax delinquency, liens and probate.
Read two parcels side by side. One has been held thirty years by an owner carrying almost no debt who doesn’t live there. The other has been held thirty years with a tax delinquency against it.
In a listing, those two look identical. In practice they’re completely different conversations.
What it means: outreach stops being spread evenly across owners who will never sell. That’s the mechanism behind most off-market deal flow, which we unpacked in why off-market deals stay off-market.
5. Feasibility moved to the front of the process
This is the newest shift and the one with the furthest to run.
Feasibility used to come after site selection. You picked a site, then paid someone to find out what actually fit on it – meaning the expensive analysis happened on sites nobody had yet proven were worth it.
Early massing and pro forma are moving into screening itself. In our October release, the AI Visualizer takes a parcel from building type through massing, placement, layouts and financials in 3D on the real lot, then saves the whole thing as a report.
What it means: the question changes from can I build here to what would this actually be, and do the numbers work – asked earlier, and on far more sites, because the cost of asking collapsed.
What ties all five together
McKinsey Global Institute puts the potential value of generative AI in real estate at $110 to $180 billion. The JLL figures explain why so little has landed yet: adoption is nearly universal, results are not.
The difference is where the tool sits in the workflow. And the through-line across all five trends is the same – information that used to arrive after commitment now arrives before it.
Rules that shift mid-process. Hazards that used to surface in diligence. Owners whose situation you couldn’t read. Feasibility you couldn’t justify testing on a maybe.
None of this makes a bad deal good. It makes the no cheaper and faster – and since most of site selection is rejection, that’s where the time actually goes.
Know what a site allows, and what it would cost you, before you make the offer.