Inman’s story last month on Local Falcon’s research, which found that 91.5 percent of established, top real estate agents never once turned up across 37,500 artificial intelligence searches, struck a nerve with readers.
Local Falcon’s research also raises a harder question. If showing up in AI search results on platforms like Claude and ChatGPT is now table stakes for real estate agents, what actually moves the needle?
Inman caught up again with Sam Mehrbod, CEO and co-founder of Roomvu, the AI marketing platform behind the new Answer Engine Optimization (AEO) tool Found, to dig into that question.
The conversation covered AI content watermarking, why broad keywords are a dead end, how agents’ data is starting to move between apps through MCP servers and the connective layer tying tools like Roomvu to large language models.
The following has been edited for length and clarity.
Inman: Anthropic recently started watermarking AI-generated text, following ElevenLabs on voice and Google on images and video. If more of what’s published online is visibly AI-made, does that work for or against agents who lean on AI to produce content?
Sam Mehrbod: I think it depends entirely on what’s in the content, not that it was AI-assisted. Watermarking will help discovery algorithms sort out what’s original from what isn’t. And probably most editors and marketers today use AI to help create content in some form. That alone doesn’t make it unworthy.
The real differentiator is whether the content has the real estate agent’s own DNA. That means it’s customized to how they actually talk to their audience, not a one-sentence prompt. Agents who feed AI well-written, personal material — their own past posts, their tone, their specific verbiage — get output that still ranks, even with a watermark.
Agents who post generic AI output are the ones who get penalized, watermark or not. We’re seeing that play out right now.
You mentioned engagement matters here, too, not just how the content was made.
Mehrbod: Right. If you post something that’s just a template, AI-generated or not, it doesn’t rank. If it gets real engagement — shares, comments, likes — that helps its visibility across the board, because Google is now ranking Instagram content directly. Most large language models are looking at similar signals.
Shifting to AEO more broadly. What’s actually different about it from traditional SEO, given Google Search already used AI in its rankings?
Mehrbod: The front door to an agent’s business is changing. It used to be a website. If you owned [yourname].com, you could be fairly confident it would rank. Now Google Business Profile plays a much bigger role in discovery, along with how much an agent has posted on YouTube, Instagram and Facebook.
Email is becoming a factor, too. How many emails an agent sends, and to how large a list, matters because email servers communicate with each other and read sender and domain reputation. But the content behind all of that has to be hyperlocal.
It can’t be “first-time homebuyer tips,” because everyone covers that. It has to be something like what a stadium renovation means for housing values in a specific neighborhood. That’s the kind of post that gets picked up, and it should go on Google Business Profile too, since that’s one of the sources LLMs scrape most.
Is there a way to actually track what’s working, or is this still trial and error the way early SEO was?
Mehrbod: It’s mostly trial and error, the same way SEO always was. There’s no published algorithm from Anthropic or OpenAI, unlike Google, which provides some transparency. What we’ve found tracking agents over time is that frequency and specificity both matter, such as how often someone mentions a specific neighborhood, how many listings they’re marketing, and how many local market updates and testimonials they’re posting. One agent we track mentioned their neighborhood by name 16 times across a stretch of content, and that showed up in how often they surfaced.
The bigger lesson is that long-tail terms rank far more easily than short-tail ones. Every agent in San Francisco wants to rank for “top agent in the Bay Area.” That’s nearly impossible. “Top agent in Berkeley who speaks Spanish” is achievable.
Even “Berkeley” alone can be too broad. Going down to a specific neighborhood like Noe Valley gets better results. It’s the same “go niche” advice that’s always applied in real estate. It just now has to run through every piece of content an agent puts out, mixing personal material, AI and interviews, all pointed at the same specific term.
You’ve talked before about building a Model Context Protocol (MCP) layer into Roomvu’s tools. Where does that fit into this?
Mehrbod: MCP is essentially an advanced API. It connects an agent’s data to their LLM, so a tool like Roomvu can talk to Claude or ChatGPT directly. We built an app on Canva, unofficially, that lets agents pull their listing photos, testimonials and tone of voice straight into templates. We had more than 200 installs in the first week.
The bigger point is that brokers are sitting on a lot of data spread across disconnected software, and whoever can move that data between tools rather than keep it siloed to one platform is going to win. If you’re on Inside Real Estate, you can use their MCP. If you’re not, what happens to your data? That’s the problem this is starting to solve.
What can an AI engine actually verify about whether an agent is legitimate?
Mehrbod: Not much, honestly, which is why the few things it can verify matter so much. Testimonials from other people are one. Listings, especially sold ones, are another, particularly if there’s a written backstory behind the sale, because that signals expertise rather than just activity. And hyperlocal content like neighborhood-specific FAQs rounds it out.
Posting on only one channel doesn’t move any of this. Posting consistently across YouTube, Google Business Profile and a written blog does.
Where does that leave agents who feel like they’re already behind on all of this?
Mehrbod: Cookie-cutter content won’t help going forward. It’s going to read as AI slop, and there’s less attention to go around relative to how much content is being produced. Agents either need someone dedicated to doing this daily or a tool that automates it while keeping it personalized.
But the bottom line hasn’t changed. Personalized content will outrank generic content, whether or not it started with an AI prompt.
