How do you make Google and the AI models agree on who you are?
This site is the demo: one canonical Person ID, schema on every page, a machine layer for the crawlers, and the same bio everywhere. The build, documented.
You pick one canonical identifier for yourself, declare it in structured data on every page you control, point every profile you own at it, and repeat the same facts about yourself everywhere — byte-identical, boring, relentless. Search engines and language models don't "know" who anyone is; they resolve entities from evidence. Your job is to make the evidence unanimous.
This site is my working example. I built carlfarrington.com as an entity-resolution exercise as much as a personal site, and since I write about getting businesses found in AI search, it's only fair to document the build on myself. Everything below is live — view source and check.
One ID to rule the record
Every claim about me on this site hangs off a single canonical identifier: https://carlfarrington.com/#carl. That's the @id of a Person object in JSON-LD markup, present in the head of every page — name, location, what I do, what I run.
The @id is the move people miss. Most sites that bother with Person schema mint an anonymous blob of facts per page — five pages, five disconnected Carls. With one stable @id, every page, every article byline, and eventually every property I operate refers to the same node. The dispatches here carry BlogPosting schema whose author field is a reference to that @id — not a name string. Names are ambiguous. References aren't.
Next step, in progress across my network: each property I run points its founder/Organization schema back at that same canonical ID, with a plain visible link alongside. One entity, referenced from many independent places, is how a knowledge-graph node accretes. Many disconnected mentions of the same name accrete nothing.
Say it the same way everywhere
Machines resolve identity the way a skeptical clerk does: by checking whether the documents agree. So the name string, the title, and the bio are kept consistent — deliberately byte-identical wherever I can control the field — across this site, LinkedIn, and GitHub. Every "creative variation" in a bio is a resolution risk: not fatal alone, but each one is a reason for a merging algorithm to hedge, and hedged entities get described vaguely or not at all.
The same discipline applies inside the site. The homepage opens with a 40–60 word definition paragraph — who I am, in prose written to be lifted verbatim by an answer engine — and that exact text doubles as the description in the Person schema. What a machine quotes and what a machine parses are the same words.
The machine layer
Alongside the human pages, this site ships a set of files that exist purely for crawlers and models: llms.txt and llms-full.txt (a curated markdown map of who I am and where the canonical pages live — full write-up here), sitemap.xml, an RSS feed, robots.txt, and humans.txt. Plus the unglamorous fundamentals on every page: one <h1>, real <time> elements, canonical link tags, per-page meta descriptions, and content that's present in the HTML without JavaScript. None of this is exotic. All of it is the difference between being parseable and being a guess.
The last piece is telling the referee: the site is registered in Google Search Console with the sitemap submitted, so indexing is observable rather than assumed. Within a day of launch, every page was discovered. You can't manage resolution you can't see.
Does it work?
The honest answer for a fresh build: entity resolution is cumulative, and I'll report results as they come in — this site is weeks old as I write, and the test is simply asking Google and the assistants "who is Carl Farrington?" and watching the answers sharpen. But I'm not guessing at the method. It's the same one I've applied across decades of SEO work and now run across my own properties: machines reward the identity that's easiest to verify.
That's the part anyone can copy, this week, for themselves or their business. One canonical ID. Schema on every page that references it. Profiles that agree with each other to the byte. A quotable definition of who you are, sitting where machines look first. None of it requires a platform or a budget — just the willingness to be consistent about yourself in public, which turns out to be rarer than it should be.