The law still asks the same question it always has: did the professional make an accurate representation and take reasonable care? A chatbot can’t answer that question in a deposition, so the agent and the brokerage will have to.
A listing description that once took twenty minutes now takes about twenty seconds. An agent pastes a few notes into a chatbot, asks for something warm and punchy, and copies the result straight into the MLS. Somewhere in that output there might be a renovated kitchen that was never renovated, a school district the home isn’t in, or a phrase that quietly signals who the neighborhood is “perfect” for. When a buyer relies on that language and gets burned, the chatbot won’t be the one answering the complaint.
Real estate has always run on representations, and the law that governs them hasn’t changed just because software drafted the words. What has changed is how often agents use AI on their own, without a brokerage policy, without training, and without anyone checking the output. That gap between adoption and oversight is where liability is quietly building, and it’s worth understanding before the first claim lands on a managing broker’s desk.
Shadow AI Has Become the Industry Default
Generative AI has moved into real estate faster than most brokerages have moved to manage it. The National Association of Realtors’ 2025 Technology Survey found that 46% of agents report using AI-generated content, most often for listing descriptions, and that ChatGPT led the tools agents reported using, at 58%. Another 32% said they hadn’t used AI in their business yet, which shows how uneven and individually driven adoption still is. In IT departments, tools employees adopt without approval are called “shadow IT,” and the same pattern has taken hold across brokerages.
Industry frameworks treat this as the earliest and riskiest phase of adoption. In the five-stage AI maturity model for real estate, the first stage is defined by individual agents experimenting with chatbots for listing copy, follow-up emails, and pulling details out of agreements, while the firm has no formal policy, no budget, and no training. Each of those missing pieces matters legally. They’re the exact safeguards a plaintiff’s attorney would ask about in discovery: who approved the tool, what review process existed, and what the brokerage did to supervise its use.
The Listing Is Still the Agent’s Representation
Misrepresentation has long been one of the most common grounds for claims against agents and brokers, and AI simply gives agents a faster way to trigger it. A misstatement about a material fact, such as square footage, lot lines, the age of a roof, or recent upgrades, can support a negligent misrepresentation claim whether a person typed it or a model generated it. Large language models are known to “hallucinate,” producing confident, specific details that aren’t grounded in anything the user provided. A chatbot asked to make a listing pop may add hardwood floors, a finished basement, or a short walk to a park that doesn’t exist.
Industry guidance is blunt about where responsibility sits. Guidance from NAR’s legal counsel on AI use warns that AI platforms aren’t 100% accurate, which makes the agent’s oversight critical, and reminds members that the Realtor Code of Ethics prohibits exaggeration, concealment of pertinent facts, and misrepresentation. The same guidance points out that the platforms make no reliable warranties about their output. Agents who aren’t Realtors don’t escape this either, since state license law and common-law fraud principles impose the same basic duty of honesty.
The vendors aren’t eager to share that burden, and regulators are pushing them further from it. Florida’s injunction request against OpenAI includes a demand that the company stop describing ChatGPT as safe, accurate, or reliable. If a state attorney general is arguing in court that a chatbot shouldn’t be marketed as accurate, an agent will have a hard time claiming it was reasonable to publish its output unchecked. The legal profession already learned this lesson the expensive way in 2023, when a federal judge in New York fined two attorneys $5,000 in Mata v. Avianca for filing a brief built on cases ChatGPT invented.
Altered Photos, Fair Housing, and Targeted Ads
Text isn’t the only exposure. California’s AB 723, which took effect January 1, 2026, and added Section 10140.8 to the state’s Business and Professions Code, requires agents and brokers to clearly disclose when listing images have been digitally altered and to give buyers access to the unaltered versions. AI-driven virtual staging, sky replacement, and object removal all fall within its reach when they change how the property is represented. Willful violations can be treated as crimes under California’s Real Estate Law, and the Department of Real Estate can discipline licensees. Other states haven’t passed similar statutes yet, but general false-advertising rules already apply to doctored images everywhere.
Fair housing is the quieter risk. Chatbots trained on decades of real estate copy can reproduce language that signals preferences about families, religion, or disability, such as describing a home as ideal for a young couple or noting its distance from a particular house of worship. GAO’s review of AI in property technology found that chatbots and AI-assisted search tools could steer buyers and renters toward only some listings or neighborhoods, which may be illegal under fair housing laws. GAO also noted that federal agencies have already pursued legal action and obtained settlements over allegedly discriminatory advertising on rental platforms. Under the Fair Housing Act, an ad is judged by how an ordinary reader would understand it, so “the software wrote it” won’t work as a defense.
Targeted marketing raises its own questions. Many agents now use AI-powered CRMs to decide which prospects get which messages, and that kind of automated profiling is drawing scrutiny well beyond real estate. A proposed class action over AI-driven targeting of vulnerable consumers accuses DraftKings of violating Massachusetts law by failing to tell customers it was using AI that way, and the company denies the allegations. Brokerages running lead-scoring tools should know which signals those tools weigh, because a model that sorts prospects by zip code or inferred family status can recreate steering at scale.
Confidentiality and the Unauthorized Practice of Law
Agents routinely handle information clients would never want public, including a seller’s divorce timeline, a buyer’s maximum offer, and financial details from pre-approval letters. Pasting that material into a free consumer chatbot can expose it, since the terms of many free AI tools let providers use inputs to improve their models unless users opt out. A listing agent who drops a seller’s bottom line into a prompt to plan a counteroffer strategy may be breaching a fiduciary duty of confidentiality without realizing it. Brokerages that haven’t vetted which tools are approved, and on what data terms, can’t easily show they took reasonable steps to protect client information.
The second trap is drafting. NAR’s guidance tells agents to avoid using AI to draft contracts, modify standard forms, or give legal advice, noting that Article 13 of the Code of Ethics and many state laws prohibit the unauthorized practice of law. An AI-written addendum or contingency clause can look polished while leaving out protections that a state-approved form includes or a real estate attorney would insist on. If that clause fails, the agent who inserted it faces a negligence claim and potential discipline from the state licensing board.
Brokerages Carry the Supervision Burden
In most states, a managing or designated broker has a statutory duty to supervise the licensees working under them, and courts regularly hold brokerages vicariously liable for misrepresentations agents make within the scope of their work. That means an individual agent’s unsanctioned chatbot habit becomes the firm’s problem the moment it produces a false statement. A brokerage with no AI policy can’t point to training, review requirements, or approved tools to show reasonable supervision. That absence can turn a defensible claim into an expensive settlement.
A workable policy doesn’t need to be elaborate. At minimum, it should name approved tools and the data terms that come with them, bar agents from entering confidential client information into consumer chatbots, require a human fact check of every AI-drafted listing against the property’s records, and prohibit AI-drafted contract language. It should also require disclosure of altered images in every market, rather than only where a statute demands it, and route fair housing questions about AI-generated copy to a designated reviewer. Firms that take those steps move out of the experimental phase and into a practice they can actually defend.
What Buyers and Sellers Should Watch For
Consumers don’t need to know which tool wrote a listing to protect themselves. Buyers should verify square footage, room counts, upgrades, and school assignments against public records, the seller’s disclosure statement, and an independent inspection before relying on marketing copy. When photos look unusually polished, it’s fair to ask whether they’ve been altered and to request the originals. Sellers should ask their agent directly whether AI tools are used in their marketing, what data goes into them, and who reviews the output before it’s published.
AI isn’t leaving real estate, and nobody’s seriously suggesting agents give up tools that save them hours every week. The law still asks the same question it always has, though: did the professional make an accurate representation and take reasonable care? A chatbot can’t answer that question in a deposition, so the agent and the brokerage will have to. The firms that build real oversight now will be the ones with a good answer ready.
Join the conversation!