How AI Is Changing Real Estate Keyword Research (and the Monthly Workflow to Keep Up)

Real estate agents ask some version of the same question every week: what real estate keywords should I actually be going after? It used to have a straightforward answer. Pull a keyword tool, sort by volume, build a page. That approach still has a place, but it's no longer the whole job. Buyers and sellers aren't just typing three words into Google anymore. A lot of them are having a conversation with ChatGPT first, and that conversation looks nothing like a keyword list.

Kyle Whigham, SEO Manager at Luxury Presence, and Thomas Gregorich, Manager of Search and AI Visibility at Luxury Presence, break the work into three parts: the traditional keyword research that still works, the newer job of figuring out what people are asking AI tools, and the monthly habit that ties both together. 

Both built their careers around the same question, long before AI search existed: what is someone actually trying to find, and how do you get in front of them first.

The short version

• Keyword volume still matters, but it no longer tells the whole story. AI Mode queries are three times longer than traditional searches, and people are asking multi-part, conversational questions instead of typing three words.
• The basics haven't disappeared: search intent, tools like Semrush and Ahrefs, Google Trends, and Google Search Console are still where research starts.
• There's no dedicated tool that shows what people are typing into AI platforms. Thomas Gregorich calls it "educated guesswork," and that's an accurate description of where things stand right now.
• The clearest patterns show up in specific, multi-part questions, not single keywords. If a piece of content still reads fine after swapping out the city name, it isn't specific enough to win.
• A repeatable monthly check, covering Search Console, Google Trends, and a look at who's getting cited by AI platforms, keeps this from becoming a one-time project that never gets revisited.

Why keyword research alone isn't enough anymore

The way people search has changed shape. According to Google's own data, the average AI Mode query is now three times longer than a typical Google search, and conversational queries starting with "Tell me about" are up 70% year over year. 

One in six AI Mode searches in the US now includes an image or another format alongside text. Luxury Presence has covered that shift in more depth in Google's AI search changes, but the short version is this: the raw volume number sitting next to a keyword in Semrush or Ahrefs is going to carry less weight going forward. Depth and coverage carry more.

Gregorich put it this way on a call recorded specifically on this topic: "Traditional keyword research in the sense that it applied to SEO has changed slightly. It still matters to understand what demand looks like, but now you need to understand and try to figure out the prompt level: what are people actually typing into ChatGPT when they're locating somebody's market."

The keyword research basics every agent needs

Before touching AI prompts, the fundamentals still need to be in place. This part hasn't changed much.

Search intent matters more than raw volume

A keyword like "homes for sale in Los Angeles" and a keyword like "properties for sale in Los Angeles" can look like two different opportunities on paper. In practice, Gregorich has found they often aren't. "I always do a manual search in Google incognito mode to see what keywords are, let's say, variations of the same search intent and which ones actually require unique landing pages," he said. If Google shows the same results for both, building two pages just creates duplicate content. If a variation like "condos for sale in Los Angeles" pulls a different set of results entirely, that one earns its own page.

Tools that show real demand

For straightforward keyword volume and competition data, tools like Semrush and Ahrefs remain the standard. Gregorich uses them to check monthly search volume for a market before recommending an agent target it: "We can look at the data and recommend which city they want to target. We can also help them understand which cities are too competitive and which ones have maybe a little less traffic than the competitive one but are easier to rank in." For a free option, Google Trends allows entering a name or a market to see how demand is trending over time.

For real questions in people's own words instead of a keyword database, Reddit threads are worth searching directly. Search a keyword like "buying a home in Los Angeles," find a relevant thread, and pull the actual language people use to describe their problem.

Check what Google is already showing

Before writing a single page, Google Search Console shows which queries are already driving impressions to existing content, including the ones nobody planned for. It's free, and it's the same tool Gregorich recommends checking before assuming a page needs to be built from scratch.

How to research what people are actually asking AI tools

This is the part that doesn't have a clean answer yet, and Luxury Presence's SEO team is upfront about that.

Why there's no keyword tool for AI prompts

"In terms of finding AI prompts, there aren't a ton of tools that just show that, like Semrush shows you traditional SEO keywords," Gregorich said. "It's really easy. But with AI prompts, it's kind of almost guesswork." 

Kyle Whigham agrees, and adds one more layer to it: no AI platform is giving agents visibility into what real users are searching for, and there's no guarantee that will change. Google has talked about eventually adding an AI visibility section to Search Console, but even then, it sounds like it will show the queries without showing impressions. It's also not fully rolled out yet.

Where to look for clues right now

Gregorich has a couple of workarounds in the meantime. The first: "You go in Google Search Console and if you look at longtail keywords that have high impressions but not a lot of clicks, those might be the kind of things people are searching for in LLMs." High visibility with low click-through can be a sign that people are getting their answer somewhere else first, and that somewhere else might be an AI summary.

The second workaround borrows from a familiar rule: great artists steal. "You could always look at, with a tool like Profound or the cheaper tool Peec, the competitor pages that are being cited a lot, and just look at their pages and see what kind of headings they're using, what kind of sections they have, and kind of cover it the same, but try to go more in depth in those sections." 

Gregorich also uses Claude for a gap check once he knows what a competitor covers: "Here's what my competitor is covering, I want to cover this, but what are the gaps? What aren't they covering that I should be covering that my audience is interested in?"

Start with the conversations agents already have

When asked how to actually do AI prompt research, Whigham's honest answer is that agents have to know their audience, because no platform is going to hand over that data. “If there was an agent trying to do this on their own, I would start with what do you know best and how can you answer those questions? How would you have a conversation with somebody who is looking to move to your area? What kind of questions would they ask? Are you answering that on your website?"

It's also worth knowing how volatile this space still is before treating any single data point as gospel. 

On that call, Whigham referenced a study he believes came from SparkToro, where a hundred volunteers ran a hundred prompts each across a dozen industries. The most cited brand only showed up about 70% of the time, and it was less than 1% of the time that the same brands appeared in the same order across every query. 

That's a reason to stop thinking in terms of "ranking" and start thinking in terms of visibility over time. Not showing up one day isn't the end of the world. The same agent could show up for someone else down the road.

What patterns to look for, and how to turn them into content

Once real prompt language replaces keyword lists, a few patterns show up consistently.

Conversational, multi-part questions beat single keywords

In Google, someone might type "Scottsdale homes for sale." In an AI tool, the same person is far more likely to type something like "best neighborhoods in Scottsdale, Arizona for families with a budget of 800,000 or less, but also have good schools." That's a completely different intent than the broad keyword, and it signals something specific about who's asking: a family, a budget range, a priority on schools. 

"Everybody wants to rank for Scottsdale homes for sale," Whigham said, "but if you're somebody who wants to work with families, if you're somebody that wants to work within a particular parameter or property type, you need to be able to showcase that expertise on your website rather than a broad page."

The move here is to map that pattern to a page that actually answers the compound question: budget, schools, and neighborhood character together, not three separate generic posts.

The "swap the city name" test

Once a pattern is mapped to a page, there's a simple gut check for whether the content is specific enough to earn it. Describing what a strong hyperlocal page looks like, Whigham put it this way: "A hyperlocal page is going to talk about realities of your daily commute, which streets flood, which homes might be in a flood zone, the actual price bands for those homes, the fluctuation in inventory... Just to do some gut checks: if you swap out the city name for a different city and it still reads just fine, then it's not unique enough."

If a page about Scottsdale would read exactly the same with "Boca Raton" dropped in instead, it hasn't done its job. This same principle is behind the topic clustering approach: a pillar page on a market, backed by genuinely specific supporting pages, tells Google and AI tools that a site actually covers that ground.

Gregorich and Whigham also frame this as an experience test as much as a keyword test. Could a competitor have written the exact same page? If the answer is yes, the page is commodity content, and that's exactly the kind of content an LLM can already generate on its own without needing to cite anyone. 

Hyper-specific details a competitor genuinely couldn't produce, like the reality of navigating a specific HOA or which park a neighborhood actually uses, are what push a page out of commodity territory.

A repeatable monthly workflow for real estate keyword research

None of this works as a one-time project. Here's the cadence Whigham and Gregorich recommend running every month, combining the traditional side with the AI side.

Check Search Console data. Look at impressions, clicks, and average position for target pages. Flag any longtail queries with high impressions and low clicks. Those are worth a second look, since they may point to content that's being surfaced without earning the click.

Run a competitor citation check. Search the questions buyers are actually asking in ChatGPT or a tool like Profound or Peec, and note which pages keep getting cited. Look at their structure as closely as their topic.

Check branded search volume. A tool like Semrush or Ahrefs shows how often a name gets searched each month, or use Google Trends for a free option. Rising branded search is one of the more reliable signals that visibility work, on Google and in AI answers, is compounding.

Update, don't just add. Market reports need fresh data on a regular cadence since prices, days on market, and inventory shift constantly. Neighborhood guides are more durable and don't need a touch-up on a schedule. Update them when something in the neighborhood actually changes. Either way, a real refresh means new data, a new perspective, or an emerging question people are asking, not swapping out a date in the title or tweaking a paragraph. If a page is genuinely years old and outdated, it's often better to build a new page and point a redirect or canonical tag at it than to patch the old one. 

Gregorich suggests automating the market report piece using Claude's Cowork feature: connect it to a data source, or even just a Google Sheet that gets updated, and schedule the report to refresh on that cadence. "You want to make the data and the data points, the perspective, as unique and based on your expertise as possible," he said, "not just something that you would find on a hundred other different websites."

Pull directly from agents. For a team or a brokerage, a short interview or a simple questionnaire with each agent, asking about the parks they actually recommend, the streets that flood, the HOA quirks buyers always ask about, turns into the specific material that makes a pillar page defensible. That's usually a better source of original content than asking any one person to freewrite a blog post from scratch.

Revisit patterns alongside rankings. Once a quarter, go back to the actual language buyers are using. Markets shift, school ratings change, and the multi-part questions people are asking this year won't be identical to last year's.

This isn't a complicated system. It's a short checklist, run consistently, so keyword research stops being something done once and starts being something maintained.