Realtors, Here's Everything Google Shared About Optimizing for AI Search...
The most quoted document in SEO this month is Google's new "Optimizing your website for generative AI features on Google Search." It is being passed around LinkedIn as gospel.
It also contains, in plain sight, an admission that Google thinks the entire real estate industry has been producing the wrong kind of content for fifteen years.
The guide defines "commodity content" (the type AI search will not reward) with one specific example: "7 Tips for First-Time Homebuyers." It defines "non-commodity content" with another: "Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line."
Both examples are real estate. Of every vertical Google could have picked to illustrate the difference between the content AI rewards and the content it ignores, they picked ours.
Twice.
Sit with that for a second.
The largest search company in the world has now told real estate, in writing, that the dominant content pattern of the last decade no longer works. Google's mythbusting section will get a lot of attention this month. The fact that they picked our industry as the cautionary tale won't.
I think the cautionary tale is more important than the mythbusting. And I think most of the mythbusting, taken at face value, will get real estate brands cited less in 2026, not more.
The guide makes three moves. It tells you the rules haven't changed: optimising for AI search is still SEO. It gives you five things to stop doing:
Then it tells you the one thing that does matter, which is creating unique, expert-led, non-commodity content.
Two of those five myths are right. Three of them are right only if Google is the only AI model that matters to you. And the non-commodity-content principle is the most important thing in the guide by a wide margin, except that the way it's being read in real estate is going to produce more commodity content, not less.
I want to take the guide seriously, because the parts that are right are very right. I also want to be specific about where it stops being useful for realtors, team leads and brokerages, which is roughly the moment Google's interests stop overlapping with yours.
Google's framing is that "AEO" and "GEO" are not new disciplines… They're SEO under different names.
This is the line being amplified loudest, and it has a structural cost that most brokerages haven't priced.
Inside a brokerage, the line item is the argument. When the marketing director walks into the budget meeting and says "AI search is the same thing as SEO," the next thing that happens is the SEO retainer absorbs the work.
The agency keeps the same headcount, runs the same monthly content production, and now also handles citation tracking across five different AI platforms, agent profile optimisation inside Zillow's AI mode, ChatGPT app surfacing for Realtor.com, Perplexity index hygiene, and the entirely separate measurement problem of figuring out which of those is driving inbound business. Same money. Three times the surface area.
The skill set has diverged whether the title has or not. The traditional real estate SEO toolkit is keyword research, MLS feed hygiene, on-page optimisation, schema, local citations, and link building. The work of getting cited in AI search adds passage-level content engineering, entity disambiguation across third-party platforms, brand presence work on sites you don't own (Reddit, Wikipedia, local press, third-party guides), and the ongoing surveillance of how four or five different retrieval systems are interpreting your brand.
There is overlap with classic SEO. There is also vast new surface area that has never appeared on a brokerage SEO scope of work.
This matters because of how brokerage marketing budgets move. A team leader who allocates $4,000 a month for SEO is not going to triple it because Google said the work was the same. They will hand the existing retainer the new work, watch the existing rankings slip because the agency is now stretched across five platforms, fire the agency, and conclude that AI search "doesn't work." Meanwhile the team across town who treated AI search as a separate line, hired a separate specialist, and ran a different playbook will be on the citation list for every buyer query in their market.
This is not theoretical. In FlyDragon's 2026 ‘State of AI Search’ benchmark of 12,400 AI responses across 192 metros, 91% of US agents are effectively invisible in the AI search engines their buyers now use. Buyer-side searches starting in an AI engine rather than a traditional one have hit 61.3%. The agents who started AI search work in early 2025 hold 5.7× the citation share of agents who started the same work twelve months later, despite the latter group spending more.
My read on those numbers: the gap between the cited agents and the invisible ones is not about who is doing better SEO. It is about who treated AI search as a separate problem early enough to build a separate operating budget around it. The "it's just SEO" framing is not a clarification. It is the move that keeps the work uncompensated, which is great for Google and lousy for the agents and brokerages funding it.
I think by Q4 2026 there is a hard split between brokerages with a line-item AI search budget and brokerages without one, and the gap will be visible in lead cost.
Here is what makes the "still SEO" framing especially misleading in real estate specifically: Google's guide describes an AI model that has the lowest AI Overview trigger rate of any major US industry.
According to the 2026 Luxury Real Estate AI Discovery Report from Stepps and 5WPR, real estate triggers Google AI Overviews 0.14% of the time. Health triggers them at 13%. Finance at 4.2%. Retail at 2.1%. Real estate, the largest single asset class in the American economy, sits at one-thirtieth of retail.
Whatever Google's AI Overviews are doing for other industries, they are not yet doing for ours at a meaningful scale.
Which is the first reason this document should not be read as advice for ChatGPT, Perplexity, Grok, or Claude. It is conveniently aimed at Google. The other large language models — the ones your buyers are increasingly typing into before they ever open Google — retrieve from entirely different substrates, with entirely different incentives, and Google's guide does not speak for any of them.
Here is what the retrieval map looks for different LLMs:
ChatGPT uses Bing as its primary search partner, supplemented by direct content deals with publishers. When a buyer asks ChatGPT for an agent in your market, the system is sending rewritten sub-queries to Bing's index and pulling from OpenAI's contracted publisher set.
Optimising for Google does very little for you here. Optimizing for Bing (the engine the SEO industry has ignored for fifteen years, including me) matters again, suddenly, for a different reason than it ever mattered before.
Perplexity uses a blend of Bing, third-party Google data, and a proprietary index it now exposes as a developer API.
The substrate is partly Google-adjacent and partly its own crawler stack. Citations on Perplexity routinely surface content that does not rank in either Google or Bing's top 10, because Perplexity's reranking treats source diversity as a feature.
Grok retrieves from two distinct layers: a real-time web search with no public dependency on Google's index, and a live feed of public X posts.
For real estate, the X layer is the part most agencies are not thinking about — agents publicly active on X with named transactions and identifiable thread history have a substrate advantage on Grok that no amount of website SEO will replicate.
Claude uses Brave Search as its primary live retrieval layer. Brave's index is independent of both Google and Bing, with its own crawler and its own ranking signals. Content that ranks well in Google can be invisible to Brave, and therefore invisible to Claude.
Five LLMs. Four different retrieval substrates.
Optimising for Google's index is optimising for exactly one of them, in the vertical where Google's AI surface fires least often.
Stanford's 2026 research captured the directional shift directly: in July 2025, 76% of AI-cited URLs ranked in the organic top 10. By February 2026, only 38% did. The rest came from positions 11–100 and beyond.
AI is finding authoritative content from across the web, not just the top of Google's index — and "the web" here means four different indexes operated by four different companies with four different definitions of authoritative.
A guide that describes one of those indexes, written by the operator of that index, telling you that index is the same thing as the other four, is not neutral guidance. It is product positioning.
Read it as that.
Google's most-quoted line in the mythbusting section is that you don't need to chunk content. The implication, repeated approvingly across SEO LinkedIn this month, is that you should write naturally for humans and trust the systems to figure out the rest.
The technical reality is different, and it matters more in real estate than in almost any other vertical.
RAG (Retrieval Augmented Generation) systems do not retrieve pages. They retrieve passages. A passage is a chunk of text. A few sentences, a paragraph, a section that the retrieval system scores against the user's search.
The chunking happens regardless of whether you optimised for it. The question is whether your content survives the chunking process with meaning intact, or whether it shatters into incoherent fragments that lose retrieval comparisons to your competitor's tighter content.
Bing has been publishing the opposite of Google's position on this for months.
In their May 2026 update on the evolving role of the index, Bing's team stated plainly that "chunking and transformations must preserve meaning and claims used in the answer." The unit of value, they said, is shifting from documents to groundable information — discrete, supportable facts with clear provenance.
The form factor where this matters most in real estate is the neighborhood guide. The neighborhood guide is the workhorse asset of every brokerage content marketing program in the country.
It’s also where most real estate content fails passage-level retrieval, because the standard format is a single page that tries to cover ten neighborhoods with three sentences each, then loops the same generic structure for the next city. When a buyer asks ChatGPT "Is Mueller a good neighborhood for families with kids and a dog?" The retrieval system pulls passages, scores them on semantic matches, and selects the passage that most tightly answers the specific question.
A page covering ten neighborhoods at the same density loses every time to a page covering Mueller specifically.
Google is right that you shouldn't break content into 50-word answer blocks. That advice was always dumb, and it produces content humans don't want to read. But the deeper principle — that passages are the unit of retrieval, and that one tight, self-contained passage beats three loose ones every time — is the most important format change in real estate content since the move to mobile.
Calling it "chunking" makes it sound like a tactic.
Calling it passage-level structure makes it sound like what it is: a basic constraint of how the systems work.
Google says you don't need to write differently for AI. AI systems, the guide claims, understand synonyms and general meaning. Just write naturally and trust the system.
I disagree with this for one specific reason in real estate, and I'd put real money on it: most agent content fails AI retrieval not because of word choice, but because of entity ambiguity.
Read a hundred agent blogs at random. Count how many times you see the phrase "great schools" without a named school district. "The luxury market" without a defined price floor. "This neighborhood" without a named neighborhood. "We've helped many buyers" without a named transaction, ZIP code, or year.
The content is technically natural language. It is also, from a retrieval system's perspective, almost entirely ungrounded (and identical).
AI retrieval is entity-anchored. The query is parsed into entities. The candidate passages are scored on entity match. A blog post that says "I help families find homes in great school districts in the Austin area" has roughly three retrievable entities: families, Austin, school districts.
A blog post that says "Last month I closed a $785,000 four-bedroom in Mueller for a family relocating from San Jose, where the buyer specifically needed walking distance to Maplewood Elementary" has fifteen.
When the retrieval system is choosing which passage to surface for "Mueller homes near good elementary schools," the second post wins by a margin that isn't even close.
This is what Google's "don't rewrite for AI" framing obscures.
You don't need to write in a special AI dialect. You absolutely do need to write with named entities, specific transactions, and dollar amounts woven into prose at a density that most real estate content has never approached.
Three principles I would tell every internal writer or editor at any brokerage are:
Google's guide will not tell you any of this, because the guide's authors are not building real estate agent websites for a living.
The retrieval math is the retrieval math regardless.
Non-commodity content is the most important sentence in the guide. AI systems are trained on the open web. Content that just restates what's on the open web does not survive synthesis. The systems are not stupid; they will pick the source that adds something the other ten sources can't.
Where the real estate industry is misreading this — including most of the AI SEO advice currently being sold to agents — is in thinking "non-commodity" means "long" or "comprehensive" or "with original photos."
It doesn't.
A 4,000-word "Ultimate Guide to Buying a Home in Austin" with original drone photography is commodity content if everything it says could have been written by anyone with an MLS feed and a search engine. Length is not non-commodity. Imagery is not non-commodity. Even original data is only sometimes non-commodity, depending on whether the data is publicly derivable.
Non-commodity for an agent means content that could not have been produced by anyone who hasn't done the transaction. Three tests I use to score whether a piece of agent content earns the non-commodity label:
Google's own contrast example — "Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line" — passes all three. It is grounded in a specific transaction. It is dense with named entities (sewer line, inspection, dollar saved).
It cannot be replicated from public data because the decision logic only exists inside the head of the person who made it.
Most agent content marketing programs running today do not produce content that passes a single one of those tests, much less all three. The work to fix that is much harder than the work most brokerages are commissioning their content agencies to do, which is why the brokerages getting cited are the brokerages who pivoted their editorial brief in early 2025, not the ones still ordering "10 things to know about the Austin market in May."
I don't want to end on a checklist, because the work is not a checklist. The work is a reframing.
Treat AI search as five interfaces, not one. Build a separate operating budget for it, with a separate scope from your existing SEO retainer.
Use Google's own "non-commodity content" definition as a hiring filter for whoever writes your blog, not a tagline for your homepage. Stop spending on content production into a substrate that doesn't retrieve, because the retrieval substrate — entities, structure, brand presence across third-party sites — is doing the work the content is taking credit for.
The hardest part of this is the part Google's guide cannot help you with at all. The retrieval systems are now plural. The optimization surface is now plural. The opinions about what works are now plural. The era of one document, one platform, one set of best practices is over for real estate, and the brokerages who keep operating as if it isn't are funding the gap their competitors will exploit.
Google's guide is one opinion. It is the opinion of the company with the most to lose from a multi-platform world.
Read it. Take what is useful. Apply it where it applies.
The buyer asking ChatGPT about your market this week doesn't know Google wrote a guide. They want a name. Either yours is in the answer or someone else's is. The next twelve months decide which.