AI Visibility for Real Estate Agents: The Fair Housing Risk Inside AI-Written Copy

The listing copy AI writes most fluently is the listing copy that carries Fair Housing risk

Real estate is one of a handful of licensed trades, alongside healthcare, tax, finance, and law, where marketing copy is itself a regulated act, and AI-generated marketing copy can create federal exposure on its own. The Fair Housing Act does not publish a list of banned words. It reaches advertising that indicates a preference, a limitation, or discrimination based on a protected characteristic, and whether a given phrase does that depends on context. This is what makes a language model a liability: it writes fluent, warm, human-sounding copy, and warm human-sounding copy about *who a home suits* is exactly the copy that gets flagged. Liability lands on the agent and the brokerage, not on the model, and penalties in repeat cases exceed $100,000 (CohnReznick, on HUD guidance).

Treat the model's favorite phrases as carriers of risk, not as words a statute forbids. "Ideal for families" points at familial status. "Walking distance to the church" points at religion. "Perfect for young professionals" points at age and familial status. "Quiet neighborhood" is not unlawful on its face, but it has a documented history as a code phrase, and boards and brokerage counsel flag it for that reason. Local boards publish their own advertising word lists on precisely this basis, as cautionary compliance guidance rather than as a recitation of the statute (Long Island Board of REALTORS). The word is not the violation; the preference it signals in context is what a complaint is built on.

The working rule is short: describe the property, never the people who might live in it. Two 2026 statutes widen the exposure. California AB 723, effective January 2026, requires disclosure when a listing photo has been digitally altered; the Colorado AI Act, effective June 30, 2026, requires impact assessments for AI used in "consequential decisions," housing included, though its enforcement and rulemaking status is still developing (CohnReznick). Run every AI-drafted remark through HousingWire's fair housing checklist, and when a phrase is genuinely ambiguous, ask your broker or counsel rather than a chatbot.

Skip the bulk-generated neighborhood pages GEO vendors sell. A machine writing at volume about what a neighborhood is *like* is a machine generating protected-characteristic proxies at volume, and those pages fail a state real estate commission complaint and an AI citation test at once.

If you are a REALTOR, Article 12 follows your AI-optimized pages: firm name and license state, readily apparent

For REALTORS and REALTOR-affiliated websites, Article 12 of the NAR Code of Ethics requires that advertising and representations disclose the REALTOR's firm name in a "reasonable and readily apparent manner," and that REALTOR websites disclose firm name and state(s) of licensure. Articles 2 and 12 together impose a duty to "present a true picture" (NAR). Two limits are worth stating plainly. The Code binds REALTORS, meaning NAR members, and not every licensed agent is one. And separately from the Code, every state has its own advertising and licensee-disclosure rules, which vary in what they require and on which kinds of pages, so the exact wording and placement obligations on your bio page, your city guide, and your social profiles are a question for your broker and your state commission, not for a national article.

The practical takeaway survives both caveats. If you are a REALTOR, an AI-optimized bio page, city guide, or listicle entry is advertising, and Article 12 travels with it. If you are a non-member licensee, your state's advertising rule almost certainly reaches the same page anyway. Either way, attribution is not a formality you can push into a footer image.

NAR's IDX Policy 7.58 governs how one participant displays another's listings. It requires the listing firm's name in a "readily visible color and typeface," prohibits wholesale download and reuse of the MLS database, forbids building a "super-MLS," and lets a seller opt out of AVM estimates beside their listing (NAR IDX FAQ).

In practice: put firm name, license state, and listing attribution in the visible body text of every page built for AI answers, not in a footer graphic. A model reads text; it does not read your logo. The brokerage and the state belong in the same sentence as your name, which happens to satisfy your compliance obligation and your citation strategy with one edit.

Buyers sign with an agent before the first showing now, so the shortlist forms during research

Buyers choose a real estate agent before walking through a single house, because a written representation agreement must be signed before the first tour. A buyer-representation agreement is a written contract between buyer and agent, mandatory across MLSs since August 17, 2024 under NAR's settlement, which also cost NAR $418 million and pulled offers of compensation out of the MLS (NAR).

That signature arrives at the end of a search the agent never sees. In NAR's 2025 Profile of Home Buyers and Sellers, fielded July 2024 through June 2025, 67% of first-time buyers and 76% of repeat buyers interviewed exactly one agent, and 43% found that agent through a friend, neighbor, or relative (NAR). The referral opens the door; the name then goes into a search box, and that box increasingly belongs to ChatGPT.

Commissions did not collapse after the settlement, so pressure moved from price to selection. Redfin measured buyer-agent commission bottoming at 2.36% in Q3 2024 and rebounding to 2.42% by Q3 2025, while AccountTECH, across 224,000-plus transactions, put it at 2.55% by January 2025 (The MortgagePoint). The job is being one of the two names a buyer arrives with.

Do homebuyers actually ask ChatGPT about agents? What has been measured, and what has not

Homebuyers verifiably use AI for housing research, but nobody has measured how many picked their real estate agent out of an AI answer. In a survey of 1,000 US adults fielded August 7 and 8, 2025, Realtor.com found 82% of Americans actively buying or selling a home use AI for housing-market information, ChatGPT used by 67% and Gemini by 54% (Realtor.com).

Agents still edge AI on trust in that same survey: 62% said an agent makes them smarter about the market, against 61% for AI. Buyers are checking the machine, not obeying it.

Now the gap, stated plainly. NAR's 2025 Profile carries no line on buyers using AI to find an agent, and no independent study has measured what share of buyers shortlisted an agent from an AI answer. The one figure claiming to, 67% of buyers in Q1 2026 naming AI their primary agent-research tool against 17% in October 2024, comes from a February 2026 survey of 4,180 buyers by FlyDragon, a vendor selling GEO services to real estate agents (FlyDragon). Treat it as a hypothesis with a business model attached.

Two vendors measured how often real estate searches trigger an AI Overview and got 4.5% and 0.14%

Nobody reliably knows how visible real estate agents are in AI search: the two published estimates differ by a factor of thirty, and both come from vendors. FlyDragon's "2026 State of AI Search in Real Estate" (12,400 AI answers, 192 metros, January to March 2026) reports 4.5% of real-estate queries trigger a Google AI Overview (FlyDragon). The "2026 Luxury Real Estate AI Discovery Report" from 5WPR and Haute Residence, released April 23, 2026, puts that rate at 0.14%, last among all industries (5WPR). There is no consensus number here.

The alarming statistics about invisible agents trace to one FlyDragon report: an 8.4% average citation share, and 91% of agents effectively never named in an AI answer. FlyDragon sells AI-visibility services to real estate agents, and HousingWire relays those figures without independently verifying them (HousingWire).

The report's most useful claim is its least frightening one: the top 1% of agents capture 47% of citation share, yet in 71% of US metros no single agent holds more than 15% of it, which describes an unoccupied position. Adoption is less contested, at 82% daily AI use among agents in Q1 2026 per RPR against 68% in NAR's 2025 Technology Survey (HousingWire). FlyDragon, GeoProof, and Atlas Unchained already sell agents this.

The sources AI actually cites when someone asks for a real estate agent near them, ranked

Your own site comes first, because it is the only source where a claim about you originates: the deals you actually closed, the subdivisions you work, the license state, the price bands. Everything else, portal profiles, portal apps built into ChatGPT, third-party "best agent" roundups, repeats that claim, or contradicts it once it goes stale. No study breaks down AI citation share by domain for agent queries, so the third-party sources below are ordered by how well each is documented, and the site sits above them by function, not by measured volume (the mechanics).

• 1. Your own site. It is the only place your closed-deal record, your neighborhoods, your relocation or probate experience, your firm name and your license state are stated rather than echoed, and a model reads text, not an MLS back end. In the three markets we measured ourselves, dental, med spa and plastic surgery in Dallas, July 2026, 73% to 79% of the citations behind AI recommendations pointed at the businesses' own websites rather than at directories (our benchmark). Real estate was not in that sample: direction, not your number. Dallas, July 2026 — one market, one day, and a primary measurement: the raw answers are stored, the method is published, and the run repeats against any city. Action: write the page before you pitch the roundup that would cite it.
• 2. Portal apps inside the assistant. Zillow became the first real-estate app in ChatGPT (October 2025), Redfin added conversational AI (November 2025), Realtor.com launched its ChatGPT app on March 30, 2026, and Google switched on AI Mode for real estate in March 2026 (RealEstateNews). Action: treat the portal profile as a model's data feed that should agree with your site, not replace it.
• 3. Zillow agent reviews. Zillow's 1-to-5 rating is a "likelihood to recommend" score across local knowledge, process expertise, responsiveness, and negotiation skills, and Zillow advises at least five recent reviews (Zillow). Caveat: "Best Match" placement is driven heavily by what an agent pays, not by deals closed. Action: ask five recent clients this month.
• 4. The five portals as a training-data block. Zillow, Realtor.com, Redfin, Trulia, and Homes.com account for roughly 61% of real-estate URLs in publicly available LLM training sets, and Zillow's share of agent discovery fell from 41.2% to 33.8% year over year (FlyDragon, a GEO vendor, via HousingWire). Action: keep the profile; stop treating it as the only door.
• 5. Third-party "best agent in [city]" roundups. Roughly 44% of AI citations go to "best…" listicle content, a figure attributed to Ahrefs, an SEO vendor with a commercial interest in AI-search visibility; note also that the link below is a third-party summary rather than Ahrefs' own study, so it is second-hand (summary). The working list: Expertise.com, ThreeBestRated, Agent Pronto, FastExpert, Yelp (per Atlas Unchained, a vendor selling AI-visibility services to agents). Action: pitch two local roundups this quarter; you cannot cite yourself in.

Check whether ChatGPT names you: six queries a buyer types, and 15 minutes between showings

An agent can run this between two showings, with no tool, no vendor, and no payment. AI visibility is the share of AI answers naming you when a buyer or seller asks an assistant for a recommendation. Open ChatGPT, Gemini, and Perplexity, paste each query below with your city and neighborhood filled in, and count how often your name comes back, and whose name comes back instead. Six queries, three engines, eighteen answers.

Read one run the way you would read one month of MLS data: real, but not a trend. Answers drift between sessions, between metros, and between models, so a single pass tells you where you stood on a Tuesday, not where you stand for the quarter. There is a second limit worth knowing before you over-read a bad result. Profound, an AI-visibility monitoring vendor whose product this figure helps sell, analyzed roughly 700,000 real ChatGPT.com conversations in the US between October and December 2025 and found web search fires in only 18% of conversations (Profound). If that holds, the model is usually answering from what it absorbed months ago, not from the page you published last week. Your new bio page does not fix Tuesday's answer; it changes what gets absorbed for next season.

Price it against the lead channel you already fund. Zillow's published cost per connection is $223 in large metros and $139 elsewhere, on non-exclusive leads that industry reporting converts at 1% to 3% (The Close). At a 2.42% buyer-side commission, a single named mention that turns into one closing pays for a lot of portal months, and an AI answer naming you costs nothing per mention and does not resell you to three competitors on the same address. See also AI visibility versus local SEO.

• "I inherited a house in [city] and live out of state. Which real estate agent should I call?"
• "My buyer-representation agreement is with an agent I do not trust. Who else works [neighborhood]?"
• "Which agent has actually closed sales in [subdivision] in the past 12 months?"
• "Relocating to [city] in six weeks. Which realtor handles corporate relocations?"
• "Which agent in [city] will take a listing at 2% now that compensation is out of the MLS?"
• "Is [Your Name] at [Your Brokerage] a good real estate agent?"

Measuring your name in AI answers touches no MLS feed, no IDX license, and no client file

Checking how often AI names a real estate agent reads only public AI answers, so it needs no MLS data, no IDX license, and no client records. That matters because the MLS question is unsettled: IDX and VOW rules were written for broker websites and portals, not for feeding listing data into language models, and MLSs are closing that gap with their own licensing terms (RealEstateNews; WAV Group). Realtor.com's ChatGPT app shows a listing preview only: training the chatbot on MLS data is "strictly prohibited" (RealEstateNews).

What moves citations is narrow and known. Aggarwal and co-authors, in "GEO: Generative Engine Optimization" (ACM KDD 2024), found that adding statistics, adding quotations, and citing sources lift visibility by up to 41%, while keyword stuffing does nothing (arXiv). For an agent that means dated market numbers, named neighborhoods, and closed-deal specifics. NAR's 2025 Profile found 91% of clients would recommend their agent again, and the raw material is in the building (NAR).

Run the six queries yourself first; the manual pass costs you fifteen minutes and tells you whether you have a problem at all. What it will not give you is the same check run every month, across your whole metro, against the specific agents whose signs you keep parking behind. That is what RecoSignal's free AI Visibility Snapshot does: it asks ChatGPT, Gemini, and Perplexity a set of buyer-style questions and reports how often your name comes back, whose names come back instead, and which sources those answers lean on. It touches no MLS feed and no client file, because it only reads public answers. It measures. It does not promise you a citation, a lead, or a listing, and no tool that does is telling you the truth about how these models work.

What real estate agents ask before touching any of this

Five questions come up in every brokerage conversation about AI answers. Take the answers to your broker before you publish anything.

• Can I let ChatGPT write my MLS remarks? Draft with it, publish as a licensee. The Fair Housing Act reaches advertising that indicates a preference or limitation based on a protected characteristic, and liability for AI-drafted copy falls on the agent and the brokerage, not the model. So every remark needs a human pass, checking not for banned words but for whether the sentence describes the property or the people. Board word lists are a useful prompt for that judgment, not a substitute for it (Long Island Board of REALTORS).
• Is this worth doing if AI will replace agents anyway? In NAR's 2025 Profile, 88% of buyers purchased through an agent or broker and 91% of sellers used one, an all-time high (NAR). What moved is the moment of choosing, which happens before anyone calls you.
• I only work with first-time buyers. Does this apply? More than before. The first-time buyer share hit a record low of 21% in NAR's 2025 Profile, median buyer age 40 (NAR). Fewer first-timers, chased by the same crowd of agents, and they research hard before calling.
• Do I have to be on every directory? No, and none of them comes before your own site, which is where your closed deals, neighborhoods, firm name and license state are stated rather than repeated. After that: Bing Places, your portal profile, and the two local roundups already ranking for "best real estate agent in [your city]," then stop. A directory nobody cites is a directory no model reads.
• How often should I re-run the check? Quarterly, and after any brokerage change, since your entity record moves with it. On what such a check does and does not prove, what an AI visibility audit measures sets the boundaries.