← Writing

Real Estate Schema: What Wasn't Resolved, and How I Resolved It

An AI research agent found my tool, its API and its dataset. Then it said no creator could be verified. For a tool about entity SEO, that's an awkward review.

Pale blue topographic contour lines on a mint background: the backdrop from realestateschema.com

Real Estate Schema is a free structured data tool I built for real estate agents. Most agent websites carry no structured data at all, so Google and AI search have nothing solid to resolve an agent by. The tool is meant to close that gap in minutes, not a consulting engagement.

It does two things:

  • Check tests any page's JSON-LD against a pinned schema.org release (v30.1), with 26 rules. It goes past "is this valid JSON" to "does this type actually have that property", "does every @id resolve", and real estate rules like license credentials and where contact details belong.
  • Create builds connected markup from a profile you set once: agent profiles, articles, neighborhood guides, FAQs, open houses and help articles.

Around that there's a REST API with an OpenAPI description and an llms.txt for agents, a local MCP server, a single-file linter for CI, and a licensing dataset: who issues real estate licenses in every US state, DC and every Canadian province and territory, each entry checked against the regulator's own page. It's CC BY 4.0.

What wasn't resolved

I ran an Exa research agent on "Real Estate Schema" to see what an AI system makes of it. The good news: it found the tools, read llms.txt to describe the API, and found the dataset without being pointed at it. The bad news was one line:

No named creator of the website was verified.

My name is in the footer of every page. The agent also offered Yodata's RealEstateSchema repository as a possible relative. It's a different project with no connection to mine.

Google is a resolution system, and so are AI agents. A name in a footer is a string. What I hadn't given them was a clean way to resolve that string to an entity. Four things were unresolved:

1. Who made the site

The site's markup described a Person with the @id realestateschema.com/#person. This site describes me as timothynishimura.com/#person. Two identifiers, two nodes. The only thing joining them was a URL string that a careful reader might match and an automated one has no reason to.

2. What the relationship was

This site listed realestateschema.com in my sameAs. sameAs means "this is the same thing as". A website is not a person. I'd said the wrong thing in the property meant to be most precise, and nothing anywhere said "made by".

3. The text agents actually read

The agent described the API from llms.txt. Neither that file nor the OpenAPI description named anyone. The pages humans read had a byline. The pages agents read didn't.

4. Which "RealEstateSchema"

The name collides with Yodata's. The site had no alternateName and nothing saying the two are unrelated, so an agent had no basis for telling them apart.

How I resolved it

One fix for each, in the same order.

1. One identifier for one person

Real Estate Schema's markup now reuses my @id from this site instead of minting its own, with the same sameAs profiles. Any system that reads both sites now merges them into one Person.

// before, on realestateschema.com
{ "@type": "Person", "@id": "https://www.realestateschema.com/#person",
  "name": "Timothy Nishimura", "url": "https://timothynishimura.com" }

// after
{ "@type": "Person", "@id": "https://timothynishimura.com/#person",
  "name": "Timothy Nishimura", "url": "https://timothynishimura.com/",
  "sameAs": ["https://www.linkedin.com/in/timothy-nishimura/",
             "https://github.com/timothy-nishimura"] }

2. "Made by", said with the right property

The WebSite now names me as author, creator and publisher, and the dataset names me as creator, all by that shared @id. On this site, realestateschema.com is out of my sameAs, which now holds profile pages only. This site's markup describes the WebSite by the same @id realestateschema.com uses for itself, with me as its creator, so both sites state the same fact about the same node.

3. A byline where agents look

llms.txt and the OpenAPI description now say who made and publishes the site: a name and a link to this site. No email or phone. The site has a no-contact-details rule, and a byline doesn't need them.

4. Saying which one it is

The WebSite has alternateName: "RealEstateSchema.com", the bare domain people actually type. llms.txt says plainly that it isn't affiliated with the Yodata repository, and so does this article.

None of this was new information. Every fact was already on the page somewhere. The fix was making each fact resolvable: one identifier, a property that means what it says, and the same claim made in the places machines read.

Still unresolved, on purpose

I didn't create a Wikidata item for the site. A self-created item for my own project would be self-serving, and there isn't yet enough independent coverage to cite. I'll revisit that once there is.

The real test is the next crawl. When Exa and the others re-read both sites, I'll run the same query and see whether the creator gets named. If it doesn't, that's the next thing to resolve.

realestateschema.com