The five graphs an AI assistant reads before it decides to cite you.

Our first piece on Total Graph Authority made a practical case. Build across three sources, first-party, third-party, and social, get them all describing the same brand the same way, and you stay in the answer whichever way a query resolves. That piece was about where to build. This one is about what you are building into.

Underneath the three sources sit the structures a model moves through when it decides who to trust. Search people have called them graphs for years, a link graph, a knowledge graph, and so on, and AI assistants lean on the same structures plus a couple more that matter more now than they used to. If the three sources are the strategy, the graphs are the machinery, and knowing the machinery tells you why the three-source model holds and where to push when your visibility stalls.

None of these graphs is a dashboard you can log into and read. They are ways of describing how signals connect, but they are real in the sense that models do resolve entities, weigh corroboration, and follow topical relationships. Total Graph Authority is the position you hold when you are strong across all of them at once.

What a graph means here

A graph is a set of nodes and the connections between them, where a node can be a page, a brand, a person, a product, or a topic. A connection can be a link, a mention, a shared attribute, or two names that keep turning up in the same sentence. Models reason over those connections rather than reading your site in isolation and scoring it, so they place it in a web of relationships and ask how well it connects to the things a given answer needs.

That reframing matters, because a backlink is only one kind of connection and for AI answers it is often not the deciding one. The graphs below are the other connections that determine whether you get named.

The link graph

This is the classic one, hyperlinks running internally and externally with the authority that flows along them. Google has leaned on it for two decades and it still works as a trust prior, though for AI answers it counts as one input among several rather than the thing that settles the question.

Two flows build it, the internal links where the structure of your own site tells a model which pages belong together, and the external links where other sites pass trust to you from sources a model may already rate. The internal half is yours to shape directly, while the external half has to be earned.

For Northwind, our made-up freight tool, internal links running from the freight-management pillar out to its comparison, how-to, and glossary pages tell a model those pages are one coherent body of work, and links from trade press then pass trust from a source the model already rates. First-party work builds the internal half and third-party work earns the external half, so when the first piece noted that third-party pages often get cited more than your own domain, this is that effect showing up, the link graph and the citation graph pulling in the same direction.

The entity graph, the spine

This is the knowledge-graph layer, the place where a model holds the relationships between entities. Northwind develops a product, was founded by someone, competes with a named rival, and operates in a country. It is also where a model works out that the Northwind on your site, the Northwind on LinkedIn, the Northwind on Crunchbase, and the Northwind on Wikidata are one company rather than four loosely related ones.

Consistent naming is what builds it, along with Organization schema, a full sameAs array, and a Wikidata item if you qualify for one, which anchors the whole thing.

Northwind Logistics
   ├── develops        → freight management software
   ├── operates in     → United States, Canada
   ├── competes with   → Alternative A, Alternative B
   └── same entity as  → linkedin.com/company/northwind
                         crunchbase.com/organization/northwind
                         wikidata.org/wiki/Q-northwind

When your schema, your LinkedIn, and your Wikidata all agree, signals from all three credit one node and that node gets strong. When they disagree, the signals scatter across two or three half-formed versions of you and none is strong enough to be the answer. This is why the build order in the first piece put the entity first, because every other graph credits back to it, and if you get it wrong everything you earn elsewhere quietly leaks out of the wrong node.

The citation graph

This graph is built from mentions, whether or not they carry a link, across news articles, Wikipedia, Reddit threads, forum answers, and video transcripts. A model treats a steady pattern of mentions from sources it trusts as evidence, even where there is no hyperlink to follow.

The things that build it are digital PR, original data, honest reviews, and real answers left in forums where your buyers already are, and the common thread across all of them is being talked about by other people rather than talking about yourself. Most brands under-build this graph because it does not show up in a backlink tool, so it feels like nothing is happening even while the mentions accumulate.

The Northwind data angle from the first piece is a citation-graph play. Say Northwind publishes that it looked at 40,000 LTL shipments and found freight misclassified on one in six, costing shippers an average of $214 a load. Every outlet that repeats that figure is a citation crediting Northwind as the source, link or no link, and the ones that never link still count toward how often the model has seen Northwind tied to that claim.

The content graph

This is the semantic map of what you cover and how deeply, topical authority described in structural terms. It comes down to which topics cluster around your brand, how thoroughly you answer them, and whether your coverage is coherent or scattered across a few thin posts. Hubs and clusters build it, glossaries thicken it, and depth with genuinely quotable passages is what makes a model reach for you over a shallower source.

Northwind's freight-management hub, a pillar page plus a comparison, a how-to, a checklist, and an entity-dense glossary, is a content-graph asset. The cluster tells a model that Northwind is a coherent source on freight rather than a brand that wrote one post and moved on, and of the five graphs this is the one you control most directly, because it lives almost entirely on your own domain.

The user interaction graph

This one is real-world interest, the branded searches, repeat visits, engagement, and sharing that show a brand is actually being used. It is harder to observe from the outside and less openly documented than the others, so treat it as a confirming signal rather than a lever you pull directly. Interest is difficult to fake at scale, which is part of why it carries weight when it shows up.

What moves it is brand demand, the branded searches that tend to follow good PR and good social, and communities that actually use the product. For Northwind, once the review video and the data study land, searches for the brand name climb, more people arrive direct, and a share of them come back, which reads as a real brand rather than a manufactured one. This is not a graph you build on its own, you earn it as the other four compound.

How the three sources map onto the graphs

The three sources from the first piece are not a separate model from the graphs, they are the channels that feed them. Line the two views up and the reason the three-source model works stops being a claim and starts looking structural.

Read down the columns and the logic falls out, because no single source feeds every graph. First-party barely touches the citation graph, third-party does little for the content graph on your own domain, and social is where the user interaction graph comes alive while almost nothing else reaches it. Build only one source and you leave whole graphs empty, which is the same as leaving retrieval paths that never reach you, whereas covering all three puts something into every graph.

The entity graph is the one that sits across all three columns, which is why it comes first in the sequence and why every other graph ends up crediting back to it.

How to tell which graph is holding you back

Once the graphs are visible, a stalled result gets easier to pin down, because "we are not showing up in AI" turns into the sharper question of which graph is thin. Four common patterns cover most of what we see.

• Named in branded prompts but missing from category ones usually means the entity graph works and the content graph is shallow. The brand is known, it just lacks the topical depth to win a "best X" question, so the hub is where the work goes.
• A strong site that stays absent from comparison answers points at a thin citation graph. The model wants corroboration you have not earned yet, which is PR, data, and genuine mentions.
• Mentioned in plenty of places but described inconsistently, or split across two spellings of your own name, is a broken entity graph. Naming, schema, and sameAs come first, because until they line up everything you earn keeps leaking away.
• Good coverage and good mentions but still flat tends to be a quiet user interaction graph. The demand is not there yet, and social utility plus PR are what move it.

Those four symptoms point to four different jobs, which is more useful than a single visibility score because it tells you the graph to work on instead of leaving you with a general worry.

A note on which graphs are actually documented

Two of these graphs are documented and three are inference, and it is fair to say which is which. The link graph and the knowledge graph are long-documented parts of how search works. The citation, content, and interaction graphs are reasonable descriptions of behaviour we can watch in the answers we track, not published algorithms with public weightings. No assistant exposes a "Total Graph Authority" score, and we are not claiming one sits behind the curtain somewhere.

Total Graph Authority is a lens rather than an algorithm, our name for the position where all of these structures point at one company at once. The value is not in the phrase, it is in what the phrase makes you do, which is to stop optimising one graph while ignoring the other four.

The first piece told you where to build, and this one is the layer beneath it, three sources feeding five graphs that all credit back to one entity. When your own content gives a model topical depth, other sites corroborate it, social shows the product working in practice, and every one of those signals resolves to the same brand, the graphs all point the same way. That is Total Graph Authority seen from the inside, and it is why the position keeps holding up under answers that keep moving.