The most useful number in real estate’s AI story is not 68%. It is the gap between two numbers from the NAR 2025 Technology Survey: 68% of agents now use AI in some form, but only 17% report significant positive business impact. About 46% see no noticeable difference at all.
Adoption is the easy part. Integration is the hard part. The data is telling agents something specific about why most AI rollouts in this profession produce a flat result, and what the 17% are doing differently.
What the 68% are using
The NAR data names the tools. ChatGPT leads at 58% of agent use. Gemini is at 20%. Microsoft Copilot is at 15%. AI-generated content (listing copy, social posts, email drafts) is used by 46%. CRMs with AI insights sit at 21%. The two motivations agents cite most for adoption are saving time (66%) and enhancing the client experience (64%).
The pattern is straightforward. Most agents are using consumer chat tools to draft text. A smaller share have AI features inside their CRM, which is closer to integrated workflow. A small minority have built AI into the way they actually run their business. The first group is the bulk of the 68%. The third group is most of the 17%.
The work that benefits least from drop-in chat tools is the part of the agent’s job that creates real differentiation: prospecting, market analysis, and the actual conversation with a buyer or seller. Drafting a listing description faster does not move the needle on volume. Writing a Facebook caption faster does not improve win rates. The agents who report significant impact are the ones who have moved AI past content generation into the operational core of the business.
What the 17% are doing differently
The agents reporting real business lift cluster on three behaviours.
They use AI for market intelligence rather than content. Tools like Restb.ai, Revaluate, and Likely.AI surface seller propensity scores and price-trend analytics that change what an agent prospects on Monday morning. Agents who let the model decide where to spend their day, rather than what caption to put on a listing, get a different result.
They use AI to handle the slow parts of the transaction. Document review, disclosure summarisation, and contingency tracking on contracts are work that AI does well and that nobody wants to do by hand. Agents who run a paperwork-heavy practice (REOs, investor work, complex luxury) and who have built AI into the document workflow are reporting real time savings, not the directional kind.
They use AI for client follow-up at scale. Most agents are bad at long-cycle follow-up, which is the single biggest source of leakage in their pipeline. AI-powered nurture sequences (text and email, with the agent reviewing before send) close the leak. The high-performing agents in the NAR data are not always the ones with the best lead generation. They are often the ones with the best follow-up discipline, and AI is finally making that discipline possible without hiring an ISA.
Why most agents stay in the flat group
The honest version of the adoption-impact gap is about workflow integration, and it has nothing to do with the agent being lazy or the tools being bad. Most agents work alone or in a small team, with no operations support, switching between MLS, lockbox apps, e-signature platforms, brokerage tools, and a personal CRM. Adding a chat tool on the side does not reshape that workflow. It just adds a tab.
The agents who get value usually do one of two things. Either they integrate AI into a single high-volume workflow they own (listing prep, buyer onboarding, follow-up) and rebuild it end to end, or they pay for a brokerage stack that has done the integration for them. Both require thinking about AI as a process change rather than a tool change.
Most brokerages are not yet at the level of integration their agents need. NAR notes that adoption is being driven by individuals, not by brokerage rollouts. This is the inverse of the pattern in larger industries, where corporate IT pushes tools out to the workforce. In real estate, the agent is the IT department. Most agents are not equipped to run that department.
The market layer most agents are ignoring
The real estate buyer is using AI before they meet the agent. Listing search, neighbourhood research, mortgage comparison, school-zone analysis, and even initial offer drafting are all happening through ChatGPT and Gemini before the buyer ever picks up the phone. Agents who do not know what the public-facing AI tools are saying about their listings, their territory, and their reputation are working with a different information set than their clients.
This is the quiet structural shift in the buyer-agent relationship. The agent’s information advantage was always the core of the value proposition. AI has flattened it for the parts of the work that used to be hidden behind a paywall (the MLS, the disclosure databases, the comp analysis). The information advantage that remains is the agent’s local relationships, transactional craft, and judgement about edge cases.
That is a different value proposition than “I have access to data you do not.” Agents who have updated their pitch to match the new reality close better. Agents still pitching the old advantage are starting to feel friction in the conversation.
What the next twelve months change
Two things move through 2026 and into 2027.
Brokerages catch up on integration. The big franchise brands have all announced AI strategies. Most are still in pilot. The 2026 to 2027 period is when those strategies either ship as real workflow integration or quietly disappear. Agents at brokerages that ship will see a different curve than agents at brokerages that do not.
Buyer-side agents face the harder pressure. The post-NAR-settlement compensation environment, combined with AI tools that handle initial buyer education and search, is squeezing the buyer-agent value proposition from two sides at once. The buyer-side agents who survive the next two years will be the ones whose offer involves judgement, negotiation, and access that AI cannot replicate.
The 68% adoption number will keep rising. Whether the 17% impact number rises with it depends on whether the profession solves the integration problem, not the access problem. The tools are not the constraint anymore. The workflow is.