The share of real estate companies that have embedded artificial intelligence tools in their business operations has doubled over the past year, significantly outpacing adoption in most other sectors of the economy, government data shows.
But an extensive survey conducted by Inman Intel reveals that real estate agents are still largely using external AI tools — not ones provided by their brokerage — and defaulting to non-enterprise versions of the models that are less suited for complex tasks.
The latest Intel Index results from August reveal that most real estate agents aren’t using AI tools for much more than generating text for social media posts and property descriptions.
But elsewhere in the brokerage, a small but growing share of their colleagues are turning to more powerful versions of today’s models to unlock higher productivity returns in areas ranging from lead management to market data and image analysis, an Intel analysis suggests.
This first installment of the Intel AI series digs deep into how these tools are actually being used in today’s real estate industry — and how significantly things have changed in the past nine months alone.
New capabilities and rising adoption
In the middle of 2024, roughly 7 percent of real estate business representatives told the government that their firm regularly used AI to produce goods and services, according to the U.S. Census Bureau’s Business Trends and Outlook Survey.
But that growth accelerated from the start of 2025 through the spring, reaching nearly 15 percent of real estate companies by late August.
Chart by Daniel Houston
Over that same period of time, the top generative AI models also underwent substantial updates as companies like OpenAI, Google and Anthropic jockeyed to release a new series of higher-performing “reasoning” models.
These models remain limited and unreliable in many domains. But the upgrades from the spring to now have leveraged greater computing resources, more human feedback and a new approach of breaking prompts and responses into a series of “thinking” or “reasoning” steps in order to achieve better results on many tasks, including math, coding and summary of information from documents and images.
But there’s a catch: Many of these improvements remain reserved for paid or enterprise subscribers. And by their own account, most real estate agents who took Intel’s August survey are opting instead for free-tier models.
- 22 percent of agent respondents and 43 percent of brokerage leader respondents told Intel in August that they used paid- or enterprise-tier AI models in their work.
This tier of model includes OpenAI’s o3 reasoning model and its recently debuted GPT-5 Thinking model. It also includes Google’s Gemini 2.5 Pro, Anthropic’s Claude Opus and reasoning models from xAI and other providers.
By contrast, most agents are using the free tier of models for most of their requests.
- 63 percent of agents and 47 percent of brokerage leaders said the models they used most often for work were free-tier models such as the default ChatGPT model, Gemini 2.5 Flash or Claude Sonnet.
As a result, the tasks for which agents most use AI today are mostly text-related.
- 69 percent of agent respondents in August said they used AI to generate text for property descriptions, up from 59 percent in December.
- 57 percent of agent respondents in August used AI to generate text for social media or marketing materials, up from 48 percent in December.
Those shares were twice as big as the next-closest uses of AI tools. But in more complex, higher-stakes tasks like market-data analysis, image and document summary, and lead management, agent use has been on the rise — especially among the agents who say they’ve gotten the most out of the tools.
Emerging uses
As major brokerage networks develop their own internal AI-driven solutions and agents continue to test the limits of what the tools can be used for, Intel asked hundreds of agents for a detailed assessment of what they use AI for, and how much it’s affected their productivity.
And so far, the agents with access to the highest-performing “reasoning” models are getting the most out of the tech.
- Agent respondents who pay for reasoning models were three times as likely to report AI has made them “significantly more productive” over the past year — 33 percent, compared to 10 percent among free-tier or other users.
These numbers should be interpreted with caution. An agent who’s willing to pay for a service probably already comes from a group that feels they are getting a lot out of it in the first place. In other words, not every agent would see similar results by simply gaining access to more effective models.
But it’s worth keeping track of the specific areas where agents with access to the best models are getting more out of their AI tools than free-tier users.
Paid users were also more likely to lean on AI for every task across the board. But there was a particularly wide gulf between users of reasoning and non- or limited-reasoning models in the following tasks:
Summarizing contracts or other documents
- 52 percent among paid users, 10 percent among others
Analyzing or summarizing market data
- 59 percent among paid users, 22 percent among others
Extracting data from property photos
- 29 percent among paid users, 5 percent among others
Generating contracts or other documents
- 29 percent among paid users, 5 percent among others
Generating text for social media or marketing materials
- 78 percent among paid users, 53 percent among others
Top executives at major publicly traded brokerage companies have told investors they are developing more elaborate brokerage-provided tools that touch on a number of these tasks, with some set to roll out in the months to come.
Next week, Intel goes deeper into how the habits of today’s AI power users have evolved alongside the new models.
