When I was in college, there were multiple girls in my dorm who shared my first name, one of whom was in my actual apartment (apparently Alyssa was a very common name for 90s babies, who knew?). Let me tell you, it was VERY confusing keeping us all straight. So everyone fell back on other ways to tell us apart. But even then, it took a lot of context to figure out which of us was which. One spelled her name with an "i" and was a twin, two of them were tall, three of us were brunette, two were from Nevada… Sometimes we went by our last names, or where we were from, or even our apartment number, but usually it took some combination of context to distinguish one of us from the others.
Which is honestly exactly the problem AI has with brands right now. At BrightonSEO, one line stuck with me more than almost anything else: brands don't lose to competitors in AI, they lose to confusion. To a language model, your brand is just another "Alyssa" in the dorm. Unless you give it the last name, the hometown, and the apartment number, etc, it has to guess which one you are. And do we really want that?
So what is an entity, anyway?
An entity is a thing, not a string. It's a person, place, brand, product, or concept that can be uniquely identified and described. A keyword is just the word someone types. "Alyssa" is a keyword. Alyssa-with-an-i, the twin from Nevada in apartment 4B is an entity.
A knowledge graph is what you get when you connect entities to their attributes and to each other: this brand makes this product, is headquartered in this city, was founded by this person. Google has been building one since at least 2012, when it started talking about moving from "strings to things." LLMs build their own understanding in a similar way. They're trying to figure out which Alyssa you are, using every clue they can find.
Entity mapping is doing that work on purpose. You figure out which entities your brand should be connected to, check whether AI actually makes those connections, and fix what's missing or wrong.
I recently attended BrightonSEO, and two talks came at this from different angles. They ended up being two of my favorites and the most useful sessions I sat in on.
Grant Simmons: entities speak louder than words
Grant Simmons of Waikay opened with the line I already shared, but definitely bears repeating: brands don't lose to competitors in AI, they lose to confusion. He also called disambiguation "identity hygiene" rather than branding, which I love, because it makes it sound less like a creative exercise and more like brushing your teeth. (Unglamorous, but skip it and things get bad, just ask my kids.)
He used a Simpsons-themed deck (a man of culture), and he used an analogy on one of his slides that I’ve co-opted and used a few times to clarify entity mapping for clients. He showed the sentence "Homer is known for living in an imaginary town called Springfield, his love of eating donuts and his excessive consumption of Duff beer at Moe's Tavern," first plain, then with every entity highlighted. On its own, "Homer" could be the Greek poet. Springfield, Duff, and Moe's Tavern are what tell a machine which Homer you mean.
His framework broke down into three kinds of "actionable entity" work:
- Entity data: find the gaps. Compare the entities on your site against what LLMs and competitors associate with your topic, then fill the gaps with relevant, related entities.
- Entity facts: find the "Dohs." Ask the LLMs what they know about your brand, check your Knowledge Graph and Google Business Profile, and fix errors by repeating the "truth" consistently everywhere.
- Entity presence: analyze and improve. Look at which entities actually show up in LLM responses, then improve consistency, connections, and coverage.
My favorite tactical idea was building an entity canonical page, basically an About page on steroids, so AI has one authoritative place to land. He also flagged query drift, which happens when a page wanders into outlier topics and dilutes what it's actually about.
Martha van Berkel: from search to AI agents
Martha van Berkel of Schema App took the same idea further down the funnel. Her framing was that we're no longer just optimizing to be found. Now we're optimizing for understanding, influence, and action, because AI systems don't just read your site. They interpret it, synthesize it, decide, and act on your customers' behalf.
Her core argument was that schema isn't a rich-results trick anymore. It's how you build a knowledge graph machines can trust. Your website is becoming a semantic data layer for marketing. Schema connects the dots, tells your story, and reduces hallucinations. And the more specific you are, the more visibility you earn.
The piece that tied back to Grant's talk for me was governing entities in your content. That means connecting your internal business context to external authoritative sources like Wikipedia, Wikidata, and Google's Knowledge Graph (think sameAs and mentions markup). Going back to my dorm example, it's the difference between "Alyssa" and "Alyssa, apartment 4B." You're handing the machine the unique identifier so it doesn't have to guess.
She closed on what "agent ready" looks like:
- A knowledge graph that's accessible, correct, and complete
- AI governance built on trust, accuracy, compliance, and open standards
- A registry of actions agents can take in your business, through emerging protocols like MCP, Microsoft's NLWeb, and Agentic Resource Discovery (ARD)
How we're putting this to work at 97th Floor
Hearing Grant and Martha was honestly a little validating, because entity mapping and schema have been a big part of our work and internal conversations lately. Here’s what that looks like for us in practice.
Step one: figure out if you actually have an entity problem
Not every brand needs a full entity mapping project. We usually recommend it when:
- The brand's positioning isn't clear, to humans or machines
- The company operates across a lot of different verticals
- It offers services that are unusual for its industry
- It's pioneering a new product, service, or category
- It's seeing slow results in AI search despite doing everything else "right"
That last one is often the key decision maker. If your content is good, your technical SEO is clean, and AI still describes you vaguely (or is very clearly confused), there's a decent chance it's an identity problem, not a content problem.
Step two: map it before you mark it up
Before jumping into implementation, we map out what the entity should be and make the connections intentionally. We look at two sides of the map:
- Internal: everything the brand owns, like blog and resource articles, product pages, use cases, and solutions pages
- External: everywhere else the brand shows up, like social profiles, third-party listings, public forums, local citations and reviews, listicles, and press releases
We also separate entities from topics. For example Nike is an entity; athletic gear is a topic. The goal is to connect the entity to the topics it should own, so AI makes those associations without having to guess.
Step three: say it in schema
Schema is next. We’re using a few different kinds for this generally.
sameAs is how you take (or retake) ownership of your entity. It tells search engines and LLMs that your website, your LinkedIn, your Wikipedia or Wikidata entry, and your YouTube channel are all the same organization.
It's been especially useful for a couple of situations we keep running into:
- Rebrands and acquisitions. One software client had grown through acquisitions, and its legacy product names were still floating around the web as though they were separate companies. sameAs helped consolidate them into one entity.
- People as entities. For founder-led and professional-services brands, we map the founder or attorney's bio page to their LinkedIn or directory profiles, so the person and the brand reinforce each other.
isPartOf / hasPart establishes how pages relate to each other. Each child page declares its parent with isPartOf, and each parent declares its children with hasPart. Stating the relationship in both directions is what makes the hierarchy unambiguous. Suddenly your service pages read as one topical cluster instead of a pile of unrelated URLs.
subOrganization / parentOrganization solved one of our most complex projects: a global company that had acquired dozens of regional brands across multiple countries, many operating under completely different names. Without it, AI had no reason to connect those brands to the parent at all.
We also use BreadcrumbList to reinforce site hierarchy and containsPlace to group multiple office locations under one organization's footprint, so they don't read as isolated addresses.
Step four: hand over something ready to use
The final deliverable is an implementation doc: ready-to-paste JSON-LD blocks, each paired with the exact page it belongs on, plus instructions to validate with Google's Rich Results Test or the Schema Markup Validator. For big sites, we provide templates for repeated patterns, so the dev team can swap in the page URL and name instead of hand-writing 40 nearly identical blocks.
Here's a simplified version of what a sameAs block looks like:
{
"@context": "https://schema.org",
"@type": "Organization",
"@id": "https://www.example.com/#organization",
"name": "Example Co",
"url": "https://www.example.com/",
"sameAs": [
"https://www.linkedin.com/company/example-co/",
"https://www.wikidata.org/wiki/Q00000000",
"https://www.youtube.com/@exampleco"
]
}
It's pretty unexciting to look at. But it's the difference between not knowing which of the four Alyssas were talking about and knowing that obviously we mean short Alyssa on the second floor.
The context layer we all need
Entity mapping and knowledge graphs do for a brand the same thing we as people do when we provide context about a topic. Homer the poet vs Homer Simpson, short Alyssa who spells her name with a y vs twin Alyssa who is actually Alissa.
AI isn't going to stop running into brands with overlapping names, similar offerings, or messy histories of rebrands and acquisitions. What you can control is how much context you give it. If there's one thing I took home from brightonSEO, it's that AI visibility isn't only about being found anymore. It's about being understood and correctly understood, consistently, everywhere you show up. The brands that win won't always be the loudest. They'll be the least confusing.
