A tenant rep opens ChatGPT and types a question: what office buildings near the convention center have full-floor availability and covered parking? Somewhere else, Perplexity is being asked about flex space with dock doors in a specific submarket. These conversations end with a shortlist, and your building is either on it or it is not. There are no blue links to climb and no page two to settle for. There is the answer, and there is absence.
This is the discovery shift behind generative engine optimization, or GEO: making your property findable, legible, and quotable by AI systems, not merely rankable in classic search. The good news for building owners is that GEO is not a new dark art layered on top of the old one. It rewards exactly what a well-run property website should be doing anyway, plus a few new and very specific habits.
How AI search actually finds a building
AI assistants compose answers from what they can read: pages they crawl or fetch live, structured signals on those pages, and the broader web they learned from. When a broker asks about buildings in a submarket, the engine looks for sources that state facts plainly, address, suite sizes, amenities, availability, and attributes what it finds. Ambiguity is fatal. If your building's facts live inside a brochure PDF or behind a listing platform's login, the engine either cannot see them or ends up citing someone else's page about your property.
So the core GEO move is the same as the core SEO move: publish the truth about your building, in plain text, on a domain you own. But AI engines add two twists of their own. They favor content that is structured, with clear headings, labeled facts, and consistent naming, and they read machine-oriented signals that human visitors never see.
Structure like you want to be quoted
State the basics in plain text: building name, address, class, and available sizes
One page per question: availabilities, amenities, location, contact
Name the building identically everywhere it appears, on and off the site
Write liftable factual sentences, like: Suite 300 is 4,200 square feet and available now
Keep it current, because an engine quoting stale facts hurts worse than silence
Floorplate sites are built this way by default. Block-based pages with dedicated Availabilities, Amenities, Location, and Contact sections give AI engines the same thing they give brokers: unambiguous, current, well-labeled facts about the property, in a structure that is easy to parse and safe to quote. A building described clearly is a building recommended accurately.
llms.txt: the new front door
There is also a direct signal you can send. An llms.txt file is an emerging convention that gives AI systems a plain-text guide to a website: what it is, what is on it, and where the important pages live. Think of it as a site map written for language models instead of crawlers. Every Floorplate site ships with llms.txt included, so your property introduces itself to AI systems on purpose rather than hoping to be pieced together from fragments.
The timing matters more than it might appear. Classic search rankings took years to consolidate, and the winners became hard to displace. AI-driven discovery is consolidating now, and the sources these systems learn to trust early will enjoy the same stickiness. A building whose facts are already structured, current, and machine-readable is teaching every AI assistant in the market what to say about it. The ones that wait will be described by whoever got there first.