August 3, 2026
This month we sat down with Dixon Jones to talk about EntityMap, the open standard he’s been working on with his co-founder Fred Laurent. Dixon co-founded InLinks and Waikay, spent a decade as marketing director at Majestic, and wrote Entity SEO: Moving from Strings to Things back in 2021, so he’s been thinking about entities rather than keywords for longer than most of the industry has had a word for it.
The starting point is a gap most of us have worked around without naming it. Schema markup is good at stating facts, but it does that page by page and organisation by organisation. It doesn’t describe how the things a business talks about connect to each other across a whole site. In traditional search, internal linking did some of that work, giving crawlers a sense of what sits near what and what matters most. There’s no equivalent when a language model is assembling an answer. Dixon and Fred’s response was to strip a website back to its component parts and publish the connections directly, rather than hoping a model infers them.
In practice an entity map is two files. A JSON file at a predictable root-level URL, which is the format models and retrieval pipelines read comfortably, and a human-readable HTML version alongside it. The HTML matters more than it first appears: every AI crawler already follows ordinary links, so a footer link means the file gets found and parsed even by systems that have never heard of the standard. Inside, three things are declared. The entities a business covers, such as products, services, people, locations and areas of expertise. The typed relationships between them. And evidence, meaning short supporting passages pulled from the site, each carrying its source URL and publisher attribution.
Dixon’s case for it rests on three things. Fewer hallucinations, because the model has a stated set of facts rather than an inferred one. Better attribution, because every claim is tied to a source and a publisher. And better visibility, because a brand gets represented more fully. This is also where llms.txt comes up. Dixon’s view is that a plain content index tells a machine what exists without telling it how anything relates or where the proof sits, and that relationships plus evidence are the parts that actually change an answer.
The most interesting finding from their early case study wasn’t a bigger volume of mentions. It was that responses shifted further down the funnel. The brand started being picked out for longer-tail, decision-stage questions, which suggests the effect is about granularity rather than reach. Worth saying clearly: this isn’t a keyword play. Dixon’s point is that an entity map crystallises what a brand genuinely is, and stuffing it would degrade the result rather than improve it.
Before launching, they took the idea to a few people whose verdict would tell them whether it was worth pursuing. One was Alex Moss, Principal SEO at Yoast. Another was R.V. Guha, who created schema.org, RSS and RDF, and who reviewed the project and called it “a good thing for the world”. Dixon’s own read is that it’s a technology that was waiting to happen
Self-declared accuracy is the obvious risk, and one Dixon raised himself. Every file carries a verification status, running from generator-draft through self-declared to third-party-verified, with a registry to handle that last tier due this quarter. There’s a validator at entitymap.org/validate that checks technical validity. Disambiguation is where these files go wrong in practice, so the value is in being unambiguous rather than comprehensive. His hardest design problem, incidentally, was reducing every possible relationship verb to a workable finite set.
We’ve started adding entity maps to sites we look after, with the first live for under two weeks. There’s some new bot activity showing up, but it’s far too early to draw conclusions from it. We’ll keep testing across a few more sites and share what we learn once there’s a clearer picture.