September 18, 2026 · 9 min read

41% or 82%? Why Two AI Surveys Disagree

NAR's random-sample survey says 41% of REALTORS use AI. An NAR-subsidiary panel says 82%. Both are real. Here is what actually explains the gap.

An agent scrolling real estate trade press this year has seen both of these sentences stated as fact: "41% of REALTORS use AI" and "82% of real estate agents use AI." Both cite the National Association of REALTORS, directly or through an NAR-affiliated group. Neither sentence is wrong. They are answers to different questions, asked of different people, in different ways — and if you are trying to figure out whether you are behind on AI adoption, the gap between them matters more than either number alone.

This is worth working through carefully, because the same pattern — one small, self-selected survey producing a much bigger number than one large, randomly sampled survey — shows up constantly in real estate marketing. We already traced the "listings with video get 403% more inquiries" claim back to a source that does not exist. This time the sources are real. The problem is reading them side by side.

The 41%: NAR's own random-sample survey

Start with the bigger, harder number to get. In July 2025, the National Association of REALTORS invited a random sample of 49,233 active REALTORS to complete an online survey. It got back 1,241 usable responses — a 2.5% response rate, with a stated margin of error of plus-or-minus 2.78 percentage points at the 95% confidence level. That methodology, printed on the report itself, is what makes this the harder number to dismiss: NAR is telling you exactly how much noise to expect around it.

The result, in the 2025 REALTORS Technology Survey (published October 6, 2025): 41% of REALTORS are currently using AI or generative AI, making it the most-used emerging technology in the survey by a wide margin — predictive consumer analytics came in second at 6%.

The same document asks about AI a second way, later in the report, and this is where a lot of confusion starts. A separate question on frequency of use over the past 12 months found 20% use AI daily, 22% weekly, and 27% a few times a month — call it roughly two-thirds who have used it at some frequency — against 32% who say they have not actively tried it at all. NAR's own September 2025 press release led with this frequency breakdown rather than the 41% figure, which means two true numbers from one survey — 41% and roughly two-thirds — have been circulating as if they were two different studies. They are the same study answering two different questions: "are you currently using AI" versus "have you used AI at all in the past year, at any frequency."

The survey also asked what difference AI has made. 46% reported neutral or no noticeable impact on their business; 33% moderately positive; 17% significantly positive; and about 4% negative. Read plainly: adoption in this survey is real but shallow, and for nearly half of the REALTORS using AI, it has not yet changed anything they'd notice.

The 82%: a smaller, NAR-affiliated panel

The 82% figure comes from a different organization asking a different question of a different, much smaller group. In February 2026, RPR — Realtors Property Resource, a technology subsidiary wholly owned by NAR — surveyed 225 real estate professionals, all NAR members, and reported that 82% currently use AI tools in their business, with 92% either using AI now or planning to. RPR's own write-up states the sample as "225 U.S. real estate professionals who are members of the National Association of REALTORS," with results based on "quality-screened responses" — language that implies filtering down from a larger pool, though the document does not spell out how those 225 respondents were recruited in the first place.

That gap — a stated, auditable random-sample methodology on one side, and an unstated recruitment method on a quality-screened panel on the other — is not a minor footnote. It is close to the textbook case the American Association for Public Opinion Research describes in its report on non-probability sampling: "the selection bias in most non-probability methods creates the substantial risk that the distribution of important covariates in the sample will differ significantly from their distribution in the target population." In plain terms, a survey that draws from whoever chooses to respond tends to overrepresent people who are already engaged with the topic — in this case, agents interested enough in AI to notice and answer a survey about it.

Pew Research Center has actually measured how much that shift tends to be. Comparing opt-in online panels against probability-based panels across six survey topics and roughly 30,000 respondents, Pew found the opt-in samples ran an average error of 5.8 percentage points — about twice the 2.6-point error of the probability-based panels. That is a general finding about survey methods, not a real estate statistic, and it does not prove exactly what happened between NAR's 41% and RPR's 82%. But it is the right order of magnitude for how far a self-selected sample can drift from a random one, and it is a more honest explanation than picking whichever number flatters the point you want to make.

To be clear about what is and is not established here: no trade publication we could find has actually put the 41% and 82% figures side by side and named this as the reason for the gap. NAR's own magazine covered the RPR survey in February 2026, and HousingWire covered it separately the same month — neither piece mentions the 41% figure at all. So this comparison, as far as we can tell, is not something the industry has already reconciled for you. Treat it as our reading of two real, differently built surveys, not as a conclusion anyone else has published.

A table, so the comparison is visible at once

That last row is worth sitting with. Realtor.com's own release reports that 82% of US adults currently active in the housing market use AI to research a home purchase or sale — a consumer statistic, fielded in August 2025 among 1,000 adults, with no research vendor named. It is not the same 82% as RPR's agent figure. It is not measuring agents at all. Two unrelated surveys happened to land on the identical headline number, about two entirely different groups of people, and both now circulate online simply as "82%."

What this means if you're deciding whether to adopt AI

None of this means AI adoption among agents is fake or overstated across the board — NAR's own harder number, 41%, still makes AI the single most-used emerging technology in the survey, ahead of predictive analytics and smart contracts combined. What it means is that a single adoption percentage, quoted without its sample size, its question wording, and its recruitment method, tells you almost nothing about whether you personally are behind. It also does not tell you whether adoption has translated into results: a separate HousingWire piece on brokerage-level AI adoption makes a related but different point — "a brokerage can hit 97% adoption and still have 70% of its agents producing the same volume they produced two years ago." Adoption and impact are two different questions, and this post is only about the first one. Before you let any adoption statistic push a buying decision, ask three questions of it: How many people were actually asked, and how many answered? Were they randomly selected, or did they opt in? And is the question "currently using" or "have ever tried," because those two produce very different numbers from the same population.

Where Pic2VidTour comes in

We think Pic2VidTour is the best AI photo-to-video tool for real estate listings, and this whole exercise — chasing a headline number back to its sample size and its question wording — is the same discipline we hold ourselves to. Real estate marketing is full of adoption stats, engagement percentages and "X% faster" claims that, when you go looking, trace to nothing citable at all; we've documented several of them.

Pic2VidTour's numbers are built to survive that test because they aren't survey results — they're mechanics you can watch happen in your own account. The price is $1.50 per AI scene, shown before anything runs, not a range or an average pulled from a self-reported panel. Every clip is generated from one real photo of your actual listing and checked against that photo before it ships; a scene that fails the check is re-filmed once free, and if it fails again it becomes a photo pan instead of a bad clip, at no charge. That is not a statistic we are asking you to trust — it is a mechanism you can verify on your own listing, the first time you use it. If you're weighing whether an AI listing-video tool is worth adding to your workflow, that verifiability is the actual question, not whether you're part of the 41% or the 82%.

What to do next

1. The next time you see an AI adoption statistic in real estate trade press, look for the sample size and the question wording before you act on it. If neither is stated, treat the number as marketing, not data.
2. If you want the harder number, use NAR's own 41% and 32%-have-not-tried figures from the 2025 Technology Survey — they come with a published methodology you can check yourself.
3. Try Pic2VidTour on one listing's photos. See the $1.50-per-scene price before you approve anything, and watch the per-clip photo check run on your own images rather than taking anyone's adoption number as a reason to trust — or skip — the category.