The AI Trust Gap in CRE: 66% Use It, Only 5% Trust It on a Deal

Most coverage of AI in commercial real estate is still counting adopters. The more useful number came out in May, and it is the one almost nobody is building around.

A study from First American Data & Analytics and DealGround, fielded March 31 to April 8, 2026 among 255 CRE professionals across brokerage, lending, capital markets, development, and asset management, found that 66% use AI weekly or daily while only 5% trust it enough to inform a real deal decision. Matt Key, vice president of property data at First American, put it plainly: "AI adoption in CRE is no longer the question—trust is."

That spread, 66 to 5, is the most important fact in broker tech right now. It says the adoption argument is over and a different contest has started.

The trust gap is the distance between how often CRE professionals use AI and how much they will let it touch a real decision. The same study fills in the middle. 53% say they use AI for support only and keep it out of final decision-making. Another 17% use it only with heavy verification. Add those up and roughly seven in ten professionals are holding AI at arm's length from the moment that actually matters, which is the recommendation a client pays for.

So this is not a slow adoption curve. Usage is already mainstream. What has not arrived is confidence. People are drafting emails, summarizing leases, and running first-pass research with AI, then quietly redoing the part that carries professional risk.

Adoption is mainstream. Confidence is not. That gap is the whole story.

The reasons are operational, which is good news, because operational problems have fixes. Among professionals who have not leaned in further, 34% say they do not know which tools to use and 32% cite accuracy concerns. Only 5% point to cost or unclear ROI. Brokers are not resisting AI on principle. Dan Mosher, CEO of DealGround, framed it well: "CRE professionals are not resisting AI—they're pressure-testing it ... they will not stake a deal on outputs they don't trust."

Pressure-testing is the right instinct. An AI tool that produces a polished market survey with one wrong rent comp does not save time. It moves risk from the keyboard to the conference room, where a broker has to defend a number they never actually checked. The polish makes it worse, because polish reads as confidence, and confidence on a wrong number is how you lose a client.

Here the national number gets sharper for a tenant-rep broker working this market. AI is most reliable where data is dense and most likely to hallucinate where data is thin, and Seattle has plenty of thin. Pull recent comparable office comps in South Lake Union, First Hill, Capitol Hill, or pockets of the Eastside and you are often working with a handful of relevant transactions rather than a deep table. That is the exact condition under which a model fills the gap with something plausible and wrong.

So a Seattle broker's exposure is inverted from the marketing. The tasks AI vendors demo best, fast comps and instant surveys, are the tasks most likely to be unreliable in the submarkets where a tenant rep earns the fee. Here the trust gap carries a dollar value: a comp set you can defend to a CFO versus one you cannot.

We dug into the local picture in Seattle's Two-Speed Office Market.

Now the part the study does not say, because that is a broker's job to say, not a data vendor's.

The 5% trust number usually gets read as a problem the model companies will eventually fix. Read it instead as the opening. In a market where everyone has the same chatbots, using AI stops being a differentiator and becomes table stakes. The next edge belongs to the broker who can do what most cannot, which is show the work. Hand a client output that is sourced and checkable and you have converted AI from a private speed trick into a public trust signal.

So the differentiator moves from "do you use AI" to "can you prove your AI output is right." The brokers who win the next cycle will build verification into the workflow itself: every comp tied to its source, every survey number checkable in front of the client. The machine raises the floor on speed for everyone. The ceiling moves to whoever can be trusted on the output.

Using AI is table stakes. Proving it is right is the moat.

This also reframes what to buy. The reflex in a trust gap is to add more AI. The stronger move is to add trust infrastructure, meaning the data foundation and verification discipline that make an output defensible. Key said as much from the data side: outputs matter "only if the outputs are verifiable and the data is reliable." More features do not close the gap. A stronger data foundation does. We make the same argument in The AI Advantage in Tenant Rep Isn't Speed. It's Credibility.

We built BrokerHQ around the belief that the tenant-rep edge in 2026 comes from trustworthy intelligence density, not from another chatbot. The idea is a command center where tenant rosters, lease-maturity signals pulled from public filings, and live leasing activity sit in one place, each traceable to where it came from, so a broker can put a number in front of a client and stand behind it.

That design is a direct answer to the 66-to-5 gap. The same study found the next wave of wanted value is in execution: respondents named automating manual transaction work (33%) and delivering reliable comps in thin or opaque markets (25%) as the priorities. Reliable comps in thin markets is the Seattle tenant-rep job, almost word for word. The real product question is whether a broker can sign their name to every output, not whether they can generate more of them faster. We treat that sourcing discipline as a security and trust property of the product, not a footnote.

One honest boundary: the value here is in the discipline, sourced data a broker can verify, not in any claim that software removes broker judgment. It does not. It makes judgment defensible.

A fair operator names the other side. Three caveats.

First, the survey is vendor-commissioned. First American sells property data and DealGround sells a broker command center, so both benefit from a story where trust and data quality are the bottleneck. The 66-to-5 finding has been reported independently and it matches what brokers say out loud, but read it knowing who paid for it.

Second, the gap may close on its own. Models get more accurate every year, and some of that 5% becomes 15% with no broker doing anything special. If trust arrives as a free upgrade, "verification as moat" is a shorter-lived edge than it looks. The answer to that worry is timing: brokers who build the trust workflow while the gap is open compound client relationships now, and those relationships do not reset when the models improve.

Third, do not weaponize the fear. The point is not to scare clients about AI errors to sell a tool. It is to hold a higher standard for your own work. A broker who quietly verifies and shows clean, sourced output will win more than one who turns every meeting into a warning about hallucinations. For a related angle, see How AI Agents Automate CRE Deal Management.

Stop competing on whether you use AI. By the numbers almost everyone does, and the client cannot see it anyway. Compete on what the 66-to-5 gap leaves wide open, which is being the broker whose output a client will actually trust on a real decision. In Seattle's thin submarkets, where the data is sparse and the stakes are concrete, that trust is the product. Build the workflow that makes every comp and every recommendation checkable, and you will own the part of the job the models cannot hand to anyone else.

The adoption race is finished. The trust race just started, and it is the one worth winning.