AI stopped reading your pages and started reading your entity graph.

For most of search history, the thing a system cared about was the page. That is changing. The models now deciding whether your brand shows up in an AI answer increasingly retrieve information as a graph of entities and the relationships between them, and a lot of well-written sites turn out to be hard to read in that format. The content is fine. The identity behind it is a mess of loose ends the machine cannot tie together.

On July 1, 2026, Search Engine Land published a full explainer on what entity-first retrieval means for SEO, and it became one of the most shared pieces in the industry that week. The idea underneath it is uncomfortable for anyone who has spent years perfecting content. The factor that decides whether you get cited is shifting away from how good your writing is and toward how legible your identity is to a system building a knowledge graph.

What GraphRAG actually is

GraphRAG is Microsoft Research's 2024 extension of retrieval-augmented generation, and there is a full ecosystem built on it now. Ordinary RAG pulls flat chunks of text that look similar to a query and hands them to the model, which then does its best to stitch an answer together from scraps. GraphRAG builds a knowledge graph first. Nodes are the entities, so your company, your products, the people who work for you, your certifications. Edges are the relationships that connect them, things like "offers", "is certified by", and "authored by".

When a model retrieves against a graph, it can follow a verified path between entities instead of guessing from raw text similarity. On the multi-hop questions where plain vector search falls apart, that produces more accurate answers with cleaner attribution and fewer invented facts. Building the graph is the expensive part. By Microsoft's own estimate, graph extraction accounts for roughly 75% of indexing cost, which is a big reason this sat as a research curiosity in 2024 and turned into a mainstream SEO conversation only now that the cost curve has bent far enough for production systems to run it.

Why entity-first retrieval became an SEO problem

A page can be accurate, original, well-sourced, and still lose the citation because the retrieval system could not work out who published it or how it connects to the claim it carries. That is the trapdoor under a lot of "our content is great and we still never get mentioned" complaints, where the writing was never the weak link and the graph around it was the thing that failed.

This reframes what optimisation even means. For a decade the job was to make a page the best answer to a query. In an entity-first world the job also includes making your brand a resolvable, connected thing that a machine can place with confidence. If the model cannot decide whether the "Search Agency" that wrote a methodology is the same "Search Agency" a client credited in a case study, it will hedge, and hedging usually means leaving your name out.

The three ways a brand goes missing in a graph

Search Engine Land's explainer lines GraphRAG's strengths up against three failures that most brands already live with, whether they have named them or not. The first is disambiguation. When the same entity shows up under different names, a graph counts them as separate, weaker signals rather than one strong one. If "the firm", "the agency", and your actual brand name never resolve to a single node, you have split your own authority three ways and handed two of them to nobody.

The second is attribution, and it stings more because you did the work. Your content gets blended into an AI answer, the fact survives, and your identity evaporates on the way through. Worse is when the model extracts your data point but credits a competitor who presented the same thing in a form that was easier to attribute. You supplied the value and someone else collected the recognition.

The third is relationships. The connections that give your expertise its meaning, the line from an author to their area of expertise to a methodology to a real client outcome, usually sit buried in prose. A human reader infers the chain. A retrieval graph does not infer, it reads declared relationships, and if yours are never stated in a machine-readable way, the graph treats your expertise as a pile of unconnected facts. All the proof of competence is there and none of it is wired together.

How to audit your own entity disambiguation

Start by searching your brand the way a graph would encounter it, across your own site, your Google Business Profile, your LinkedIn, Crunchbase, Wikidata, and any directories you appear in. Write down every variant of your name that shows up, every founder and staff name, and every way your services are labelled. If the same company appears as three slightly different strings with no signal that ties them together, that is your disambiguation problem in plain sight.

Then check whether those variants point back to one canonical identity. Consistent naming across profiles, a sameAs set in your Organization schema that links your site to your authoritative external profiles, and author markup that ties every article to a real, described person are the levers that collapse three weak nodes into one strong one. This is unglamorous cleanup work, and it is the highest-impact move most brands can make before chasing anything newer, because a graph cannot cite an entity it cannot pin down.

EntityMap, a sitemap for what your site knows

There is movement one floor up from the models too. On June 1, 2026, a new open standard called EntityMap opened a 33-day public consultation ahead of a July 1 launch. It was started by Fred Laurent, CTO of InLinks and Waikay, with backing from Dixon Jones, names this audience already associates with entity SEO. The framing is deliberately familiar. Where sitemap.xml tells search engines which pages exist, an entitymap.json file is meant to tell AI systems what an organisation actually knows, which entities it covers, how they relate, and where the supporting evidence lives, with every declared relationship able to carry its own source URL, publisher, and timestamp.

Whether you adopt it yet is a judgement call rather than an obvious yes. It is open-licensed and ships with a human-readable companion file and a working reference implementation, so the cost of publishing one is low. The honest caveat is that no major AI lab has committed to consuming entitymap.json, so today it is a bet on where retrieval is heading, not a switch that turns on citations. For brands already doing the schema and disambiguation work, adding an entity map is a small, low-risk extension of the same idea. For brands that have not fixed their basic entity consistency, it is a roof on a house with no walls.

What the evidence actually shows

The mechanism is more than theory at this point. In a Schema App case study, connecting schema markup with proper entity linking produced a 19.72% increase in AI Overview visibility, and in a Fortune 500 implementation the same approach moved average AI Overview market share from 27.5% to 36%. A separate cross-platform empirical study on SSRN tested whether structured data predicts AI citation across engines and found structured, entity-rich markup correlating with getting picked up in generative answers.

None of these are enormous single-move wins, which is the part to hold onto. Entity work compounds rather than spikes, because each cleaned-up relationship makes the next one easier for a graph to trust. The brands seeing gains got there by making their existing expertise legible as connected facts a machine can follow, not by shipping one more article on top of the pile.

Where to start

If you only do one thing this quarter, make your brand resolve to a single entity everywhere it appears, then declare the relationships that prove your expertise instead of leaving them implied in prose. That is cheaper than a content sprint and it addresses the failure that is actually costing you citations. Everything past that, from richer schema to an entity map, is an extension of the same discipline.

We run this exact audit for clients through our free AI Visibility Audit, checking how AI systems currently resolve your brand and where your entities are splitting or getting misattributed. If you want to see how your brand reads to a graph before it costs you another answer, get in touch.