68% of Agents Use AI. Only 17% See Real Business Impact.

Walk a real estate office and you'll find AI everywhere. It's writing the listing copy, polishing the open-house email, suggesting a caption for the carousel post. Everybody's using it. Almost nobody can tell you what it did to the numbers.

That's not a hunch. The National Association of Realtors' 2025 Technology Survey found that 68% of agents now use AI in some form, yet only 17% report a significant positive impact on their business (NAR, 2025). Another 33% call the impact moderate. And 46%, nearly half of everyone who tried it, report no noticeable difference at all (NAR, 2025). Two-thirds of an industry adopted a technology, and the largest group came back to say they felt nothing.

This is the gap that should keep any operator up at night. The question was never "did we try AI." Almost everyone tried AI. The question is whether trying it moved a single metric you'd put in front of an owner.

Adoption is not the same as impact

The tell is in what agents actually use AI for. The most common tools in that same NAR survey have nothing to do with intelligent automation. They're eSignature, used by 79%, and social media, used by 75% (NAR, 2025). The headline-grabbing generative work, the listing descriptions and the content drafts, sits on top of a tech stack that's still mostly digital paperwork and posting.

That mix explains the 46%. Faster listing copy is a nice convenience that stops well short of a business outcome. It doesn't shorten your response time, it doesn't raise your contact rate, it doesn't pull a deal out of a pipeline that would otherwise have leaked. It saves a few minutes on a task that was never the bottleneck. You can adopt that kind of AI all day and your year-end numbers will look exactly the same, because nothing load-bearing changed.

Generic AI use produces generic results, which is to say no measurable result at all. The 17% who felt something used the same chatbots as everyone else. What set them apart was aiming it at a different kind of problem.

The 17% put AI to operational work

Read the gap the other way and it stops looking like a disappointment and starts looking like a map. The 46% used AI as a faster typewriter. The 17% used it as an operator.

The difference is the work you aim it at. Instant lead response. Routing the inquiry to the right person before it goes cold. Tracking the transaction so nothing falls through the cracks at day nineteen. Qualifying the contact so a human spends time only on the ones worth it. None of that is glamorous and none of it writes a clever caption. All of it touches a number an owner actually watches.

That's the line between the two camps. Content AI sits beside the workflow and speeds up a keystroke. Operational AI sits inside the workflow and changes the outcome. The first feels modern. The second moves the business. Most of that 68% bought the first one and is quietly wondering where the impact went.

You can't see the impact you never measured

Here's the part the survey can't show you, and it's the more uncomfortable explanation for the 46%. Some of those agents probably did get a lift. They just have no way to prove it, because they never wrote down where they started.

This is the quiet failure underneath most disappointing AI rollouts. Without a baseline captured before deployment, the after-number has nothing to be compared against, and the delta is unprovable by definition (Moveworks). The system might be deflecting a third of your routine inquiries, but if you never counted the inquiries last quarter, "a third of what" is a shrug. Impact you didn't instrument reads, on a survey, as no impact at all.

The scale of this is large. McKinsey found that 61% of AI customer-service projects miss their first-year targets, and the failure is largely one of measurement and scoping, with the technology rarely the culprit (McKinsey, via theStacc). The model usually works. The project fails because nobody defined what winning looked like, in numbers, before they turned it on. A tool that quietly performs and a tool that quietly does nothing look identical when neither one is being measured.

So the instruction writes itself. Capture the baseline first: response time, contact rate, the share of inquiries that get a same-day answer, the percentage of leads that get no answer at all. Then deploy. Then watch the same dials move. The two metrics that prove the case fastest are conversion-rate lift and cost-per-acquisition payback (SparkCo). Without them, you're left telling an owner the AI "feels" like it's helping, which is exactly how you end up in the 46%.

The gap is the opportunity

Step back from the numbers and a strange picture comes into focus. An entire industry adopted a technology, and most of it reported back that the technology did nothing. The reflexive read is that AI is overhyped. The sharper read is that 68% of a market is using the easy version of a tool and mistaking that for the whole of it.

That gap between adoption and impact is the most interesting number in real estate right now. It says the early adopters mostly automated the wrong thing, the thing that was easy to automate rather than the thing that was worth automating. The work that actually moves a business, the response, the routing, the qualification, the follow-up, the measurement, is still sitting there, largely untouched, behind a wall of listing copy.

Two-thirds of the industry already pressed the button. Almost none of them aimed it. The technology didn't fail. It has barely been pointed at the work that matters.

Sources: National Association of Realtors, 2025 Technology Survey (via HousingWire), 2025. Moveworks, "Measure and Improve Enterprise Automation ROI." McKinsey customer-service AI project data via theStacc, "AI Customer Service Cost Savings." SparkCo, "Analyze Sales Funnel Conversion Rates."