A seller opens ChatGPT and types: Who’s the best listing agent in my neighborhood? Three names come back, a sentence about each, a few links. No page two. No ads to buy your way onto. Every other agent in that market is out of the conversation before it starts because of real estate AI visibility.
That’s the fear driving a whole new spend category in real estate marketing: Answer Engine Optimization, Generative Engine Optimization, real estate AI visibility — pick your acronym. Some of what’s sold under those labels is real work. Cleaning up your business data, fixing your site’s structure, publishing verifiable transaction evidence, tracking which sources the models actually pull from. Some of it is people are getting paid to post about you on Reddit and pretend they’re a happy client.
Two things happened in the last eight months that should change how brokerages buy this stuff.
The Reddit collapse
On Aug. 20, the AI visibility firm Promptwatch reported that Reddit’s share of ChatGPT Search citations fell from about 3.83% to about 0.52%. That’s an 86% drop, and it happened over roughly four days in mid-August. Reddit’s citation share on Google’s AI products slid too, but gradually. The ChatGPT drop was a cliff.
Promptwatch was careful about what it did and didn’t know. The data shows when the shift happened, not why, and the firm said a collection problem couldn’t be ruled out. OpenAI hasn’t explained it. Reddit’s communications officer, Adam Collins, told Axios the change has no meaningful impact on the business, and added a warning worth reading twice: “If you’re a brand looking at Reddit merely as a way to hack your presence on an AI platform, you risk frustrating your consumers here.”
Now think about the agent who spent the first half of this year paying someone to seed Reddit threads with their name.
The lesson isn’t that Reddit is dead. It’s that any strategy built on one outside platform holding one particular level of influence is a strategy you don’t own. Steve Rubel at Burson put it plainly to Axios: There’s no single source to chase. “It’s a long game, and [brands] need to be engaged on many different channels.”
The other thing that changed
The FTC’s Consumer Reviews and Testimonials Rule has been on the books since 2024, and on Dec. 22, 2025, the agency sent warning letters to 10 companies over possible violations. The flagged conduct reads like a menu of what some AI-visibility vendors sell: misrepresenting whether a reviewer actually used the service, conditioning payment on the reviewer saying something positive, and failing to disclose reviews written by company insiders or their relatives.
Penalties run up to $53,088 per violation.
Separately, the FTC’s endorsement guidance requires clear disclosure of any material relationship a consumer wouldn’t expect. Reddit’s own rules require authentic participation and ban content manipulation. Google’s guidance on AI features says scaled or automated content produced mainly to manipulate rankings or generative responses can violate its spam policies.
And for Realtors there’s a layer most compliance conversations skip entirely. Article 12 of the Code of Ethics requires presenting a true picture in advertising and disclosing your status as a real estate professional. A paid post written in the voice of a neighbor who “just sold with Sarah” is not a true picture, and it doesn’t disclose anything.
Whether a specific campaign crosses a specific line depends on the facts. But no brokerage should approve a third-party AI-visibility contract without written answers to these:
- Are the comments labeled as advertising, or do they read as independent consumer opinion?
- Does the account posting have real experience with your business?
- Who writes the content, and who approves it before it goes live?
- Are performance claims substantiated?
- Who owns the accounts and the published assets when the contract ends?
- What happens when a platform removes the posts?
- Who carries the regulatory and platform-policy risk? Name the party.
Get it in writing. A vendor who won’t put it in writing has told you something.
AI answers are assembled, not awarded
There is no master list of the best agent in every ZIP code. Web-connected AI systems go find whatever they can and reconcile it. OpenAI’s own documentation describes web search as a tool that returns current information with sourced citations, and notes that the full set of URLs consulted is usually larger than the handful shown as inline citations.
What’s in that source mix is not what most agents assume. BrightLocal tested 20 local searches across 10 industries and found ChatGPT pulling from Foursquare’s database for somewhere between 60% and 70% of local results, especially in smaller markets and narrower categories. Business websites showed up as a source 58% of the time. Yelp appeared in 33% of searches overall. Google’s models leaned hardest on Google Business Profile. Industry-specific directories mattered a lot: dental queries surfaced ten different dental directories.
So the agent obsessing over blog cadence while their Foursquare and Yelp records sit wrong or empty has the priority backwards.
The prompt wording changes everything too. “Who’s the best listing agent” rewards reputation and review consensus. “Who sells the most homes here” needs transaction evidence. “Who should list a $5 million property” needs proof of that price point specifically. “Which agent knows Wellington Park” rewards geographic depth nobody else has published.
Most agents don’t have a content problem
They have an evidence problem.
Take an agent with 22 years in the business, 400 closings and a reputation any neighbor would vouch for. A model still can’t confirm any of it, because the proof is scattered. The website says “The Miller Group.” The Google Business Profile says “Sarah Miller Real Estate.” Two dead brokerage profiles from 2019 are still indexed. The sold listings are attributed to the brokerage, not to her. Reviews are split across four platforms. And the community knowledge, the stuff that actually makes her the right answer, exists only in her head and in listing appointments.
Machines can’t infer what was never published or consistently attributed. That gap is also the opportunity, because closing it is entirely within your control and nobody can deindex it out from under you.
A 30-day audit
Week 1, benchmark.Write down the 20 questions your buyers and sellers actually ask. Run each one through ChatGPT Search, Gemini, Perplexity and Google AI Mode. Log every agent, brokerage and URL that comes back. Sort the sources into three buckets: things you own, things you earned, things you rent. Note which competitors keep showing up.
Week 2, fix your data.Pick one canonical name for the agent or team and use it everywhere. Standardize brokerage, phone, website and service area. Correct Google Business Profile and Bing Places, then check Foursquare, Yelp and Apple Business Connect, since those feed more AI answers than most agents realize. Kill duplicate and obsolete listings. Confirm license and leadership info is current.
Week 3, publish proof.Two seller case studies with real numbers. A page for each community you actually serve. Documented listing and transaction history attributed to you by name. One piece of original market data nobody else has. Written answers to the five seller questions that decide whether you get hired. Transcripts on your videos and podcast episodes, because that text is retrievable and the audio isn’t.
Week 4, get corroborated.Ask recent clients for honest reviews, on the platforms your market’s answers actually cite. Go after local and trade press. Show up in community and professional organizations in ways that leave a published record. Write your disclosure rules for staff and vendors and put them in the handbook. Then rerun the same 20 questions and compare the source mix.
Measure recommendations, not just traffic
Rankings, clicks and sessions still matter. They just miss the part that’s growing, which is whether you were in the answer at all before anyone visited a website.
Start tracking how often you’re recommended, where you land inside the answer, what share of voice competitors hold, which sources get cited, whether the summary about you is accurate, how it varies by neighborhood and price band, and how many inquiries mention an AI tool by name.
The goal isn’t to make a model say something flattering that isn’t true. It’s to make sure that when the model goes looking, there’s enough accurate, current, corroborated evidence that leaving you out would be the wrong answer.
The agents who win this won’t be the ones with the most manufactured mentions. They’ll be the ones whose closings, market knowledge, client results and community record are documented well enough that a person and a machine reach the same conclusion.
Tim and Julie Harris are the co-founders of Premier Coaching, bestselling authors and hosts of Real Estate Coaching Radio. Premier Coaching provides real estate professionals with practical systems for lead generation, listing acquisition, business planning and profitable execution. The views expressed are their own.
This column does not necessarily reflect the opinion of HousingWire’s editorial department and its owners.
To contact the editor responsible for this piece:[emailprotected]