The existence of AI overviews, the latest interface changes, and whether ChatGPT is “the new Google” are all just surface stories of where search is going. 

The underlying shift to watch is that placement and influence, which used to move together, have come apart. You can now rank without being relied on and shape an answer without ever being clicked. And the systems that decide which sources get referenced in a response don’t care where you’re ranked on a list — they care whether your explanation still holds up when the question gets rephrased.

To catch the full picture of your site’s SEO, it’s important to first understand the distinction between positional and relational visibility.

Positional vs. relational visibility

Positional and relational visibility are directly explained by the difference between being ranked and being referenced.

Ranking is positional: it measures where a page appears relative to others for a specific query at a specific moment. It’s useful, well-defined, and easy to track — which is why it became the metric.

Referencing is relational: it measures whether a source is trusted enough to help form an answer. Not whether it appeared. Whether it was used.

Whether your content was used by an LLM is a different question that produces a different kind of visibility. 

Ranking (positional visibility) is momentary: it applies to one query, one interface, one snapshot. Referencing (relational visibility) persists across questions, contexts, and formats because it’s tied to how a topic is understood, not where a URL sits in a list. 

In other words, ranking can be won tactically and referencing has to be earned systemically.

This is why being referenced compounds growth. A page moves up or down with every update,but a brand that consistently explains its space well stays in the conversation regardless of what the interface looks like next quarter.

Why this distinction just became relevant

For most of SEO’s history, placement and influence were the same thing. If you ranked, you were clicked. If you were clicked, you were read. If you were read, you had a chance to be remembered. The chain held tightly enough that optimizing for the first link in it was a reasonable proxy for the whole.

That chain is now broken in both directions.

Users get oriented before they click, and sometimes instead of clicking. Synthesized answers pull from sources without surfacing them as links. The system can rely on your explanation while showing the user something else entirely. Conversely, a page can rank well for a query and contribute nothing to how the topic gets explained downstream.

Ranking didn’t get weaker. It got narrower. It still answers the question it always answered: where did we show up? It just stopped answering the larger question we used to fold into it: were we relied on?

How to build a system that produces LLM references 

Unfortunately, you can’t optimize for referencing the same way you optimize for a keyword. There’s no page-level lever. Instead, referencing emerges when your underlying system is built to produce it. 

To produce references, your system needs to have entity strength, well-planned internal architecture, and content that supports query fan-out.

1. Build entity strength

The system has to know what you are, what you’re authoritative in, and how your topic boundaries map to the rest of the web. Without that, you’re a string of pages, not a source. (More on this.)

2. Plan internal architecture that reflects how the topic actually works 

Internal linking isn’t a distribution mechanism for PageRank — it’s how you tell the system which ideas are foundational, which are extensions, and how understanding is supposed to deepen. When the architecture is coherent, the system can follow the structure of your thinking. When it isn’t, your content is just a pile. (More on this.)

3. Write content that supports query fan-out

When a question gets decomposed into sub-questions, the brands that get referenced are the ones whose content meets those sub-questions with the same clarity and consistency as the original. Gaps in coverage are gaps in reliability. (More on this.)

None of these produces referencing on its own. Together, they make a brand legible to the system as a source rather than a collection of URLs. At that point, being referenced isn’t something you ask for. It’s what the system does because it makes sense.

How LLM referencing affects measurement

If you only track rankings, you’ll miss every moment where your brand shaped understanding without being clicked. You’ll miss the language that got reused, the framing that carried forward, and the summaries your content informed. None of it shows up in a rank tracker because none of it is positional.

The signals that do show LLM referencing are split between two types

Direct signals

First, are you actually being cited? Tools like Profound, Knowatowa, and SEMRush track brand mentions inside AI answers across ChatGPT, Perplexity, Gemini, and Google’s AI Overviews. Manual prompt audits — running the queries that matter to your category and recording which sources get pulled in — fill the gaps. 

Knowledge Graph presence, Bing’s entity API, and Wikidata coverage tell you whether the system recognizes you as an entity in the first place, which is the precondition for being referenced at all.

Indirect signals

Branded search volume, direct traffic to deep pages, shorter sales cycles, conversion improving without traffic moving, prospects who arrive already informed. These aren’t noise — they’re what referencing looks like on the way to revenue. When referencing is strong, a ranking dip stops translating cleanly into a loss of influence, because reliance persists even when placement shifts.

Neither layer replaces traditional SEO metrics. They contextualize them. And together they tell you something a rank tracker can’t: whether the system is treating you as a source or just a result.

LLM referencing can’t be manipulated by a tactic

The common error right now is treating referencing as something to chase by adjusting prompts, adding schema, or imitating whatever pattern shows up in the latest AI overview screenshot.But referencing isn’t a tactic. It’s an output. And it gets harder to produce, not easier, the more you try to game it.

What actually moves the signals above — citation rates, entity recognition, branded search, share of voice in fan-out queries — is the same set of things every time: a recognizable entity, an internal structure that reflects how the topic works, and coverage that holds up when questions get rephrased. 

That’s not a content marketing trick. It’s the same architecture problem that determines whether your site holds together for anything else, and it’s the only thing that explains why some brands keep getting pulled into answers while others, with better rankings, don’t.

Ranking gets you noticed. Referencing means you’re trusted. In a search environment increasingly built on synthesis, trust is the signal that lasts — and it’s the one you can now actually measure.

Building an SEO system that earns trust instead of chasing position? Let’s talk.