SEO. GEO. AEO. AI Visibility. It’s One System. Here’s the Map.

Every few weeks now somebody publishes a new framework for how brands should think about AI visibility. Agentic search. GEO. AEO. Pick the acronym… there’s a consultant somewhere ready to sell you a retainer around it.

Most of these frameworks are useful. Most of them are partial. And most of them are describing the same underlying thing from a different angle without quite saying what the underlying thing is.

The underlying thing is the retrieval system. Not a product. Not a platform. Not a discipline. The system that decides which entities get surfaced when any query runs through any evaluator. Google’s organic rank, ChatGPT’s citation layer, Perplexity’s source selection, Gemini’s AI Overviews, an agent doing multi-step research before it acts on your behalf. All of them reading the same system. Different evaluators pulling from the same structure.

Three frameworks showed up in the same week trying to describe parts of this system. They came from different starting points. They cover different ground. They aren’t equally developed. But the fact that three independent groups arrived at layered structures in the same window… that’s not coincidence. People aren’t inventing layers. They’re discovering them. Because the retrieval system actually has them.

Here’s how the three relate.

The retrieval architecture — L1 through L4

Posts 10 (“Retrieval Optimization: It’s Not Just SEO or GEO”) and 11 (“Retrieval Optimization at Each Layer“) laid out a four-layer model built from the data up. 255 research findings distilled from 151+ sources. The framework emerged from what the data showed before anyone named it.

Four layers. L1 Entity Establishment… does the system know you exist. L2 Entity Depth… does it know you at depth. L3 Category Citation… do you show up when someone asks a category question. L4 Content Optimization… can the system actually extract specific claims from your content.

Three mechanisms underneath all of it. K is what the model already knows from training. T is what’s accumulating in training data now and will become K at the next release. R is what gets retrieved live at query time. Day-to-day volatility is almost all R. Long-term position is K. T is the bridge between them.

The layers have a dependency order and this matters. L1 feeds L2. L2 feeds L3. L3 and L4 are where most observable optimization happens but they’re brittle without L1 and L2 underneath. A brand with strong content optimization and strong category presence but no training-layer depth is visible until retrieval shifts. And then it isn’t. Glenn Gabe’s case study from this week showed exactly that. A site with massive content at scale and zero entity depth collapsed completely when Google pulled the retrieval surface. Nothing underneath to hold it up.

Three evaluators read the same four layers. SEO reads them as rank-and-click. AI visibility reads them as cite-and-answer. Agentic reads them as evaluate-and-act. Same layers. Different thresholds. Agentic is the highest threshold because the agent acts instead of recommending. But it’s not a separate discipline. It’s the same work with a higher bar.

This framework describes the retrieval system itself. Where signals live. How they move. What mechanisms carry them. It’s the architecture.

The measurement framework — Aleyda Solis

Aleyda Solis published a three-layer measurement framework this week that does something different and does it well.

Her three layers. Presence… is the brand actually appearing in AI answers, how, where, on which platforms. Readiness… are the structural conditions for stronger visibility in place. Business Impact… is any of this translating into measurable value. And she keeps observed signals, proxy signals, and modelled signals separate instead of blending them into one number that overclaims attribution. That’s a level of epistemic discipline most frameworks skip.

The part that works best is the diagnostic chain. Each layer hands a hypothesis to the next. Presence tells you where the gaps are. Readiness tells you why the gaps exist. Business Impact tells you whether fixing them produces commercial value. One connected diagnostic, not three separate audits. That chain is genuinely more developed than anything else I’ve seen on the measurement side.

She gets other things right too. Platforms tracked separately because retrieval behavior diverges dramatically across ChatGPT, Gemini, Perplexity, AI Overviews, AI Mode. Prompt libraries built from real buyer behavior with actual constraints… price, team size, industry, integrations. Not keywords stretched into prompts. Real buyer questions.

Where this framework and retrieval optimization connect is in her Readiness layer. Her ten characteristics of AI search winning brands map to L1 through L4 work. But without the dependency order. They’re presented as a flat checklist. Retrieval optimization adds the sequence… entity establishment before entity depth before category citation before content optimization. Because each layer feeds the next. A flat checklist tells you what’s missing. A sequenced architecture tells you what to build first.

Her measurement methodology on top of the retrieval architecture would be a more complete system than either one alone. She built the measurement layer we haven’t. We built the dependency architecture she hasn’t. Complementary. Not competing.

The product stack — Semrush

Semrush published a three-layer stack recently. Traditional SEO as the foundation. AI Visibility as the second layer. Agent Readiness as the third. “Each layer builds on the last.” Clean visual. Clear product path.

The principle is correct. SEO is foundational. The visual is accessible. And they correctly identify that agent readiness requires the prior layers in place.

Here’s where it diverges. The framework is organized around Semrush’s product catalog. Traditional SEO maps to their core tools. AI Visibility maps to Semrush One and Brand Radar. Agent Readiness maps to WebMCP and whatever’s emerging. That’s smart commercial architecture. It gives buyers a natural decision path for which product to invest in next. Honest about what it is.

But it also means retrieval optimization work that doesn’t map to a Semrush product tends to disappear from the framework. Editorial placement in think-piece listicles… not a Semrush product. Conference speaking and journalist relationships that build long-term training-layer presence… not a Semrush product. Publishing primary research on third-party domains… not a Semrush product. All of these are core L2 and L3 work. None fit cleanly in the stack because the stack is organized around what Semrush sells.

Their own Venn diagram makes this visible. They published an SEO vs GEO comparison where the “GEO-only” circle contains technical SEO foundation, E-E-A-T and trust signals, domain authority, high-quality backlinks, brand mentions, structured data, clear page structure, topic authority. Every item in the GEO circle is SEO. You can’t fill the circle without borrowing from SEO fundamentals. The Venn diagram accidentally proves the point… GEO as a separate discipline doesn’t hold.

The framework isn’t wrong about what it covers. It’s bounded by what Semrush can measure and sell. Retrieval optimization isn’t organized around what anyone sells. The layers come from the mechanics, not from a product catalog.

Everyone keeps finding layers. here’s why.

This is the part that matters more than which framework is most complete.

Three independent groups. One working from a 255-entry research database. One working from client measurement needs. One working from a product suite. All arrived at layered structures in the same window. None of them looking at each other’s work. All of them looking at the same system.

That’s reverse engineering. The retrieval system has layers and anyone who looks at it carefully enough independently discovers some version of them. The convergence isn’t evidence that anyone invented the right labels. It’s evidence that the underlying structure is real.

The differences between the three tell you what each group was optimizing for. Data-first produces an architecture with dependency order and mechanism mapping. Measurement-first produces a diagnostic chain with KPI methodology. Product-first produces a commercial stack aligned to what the company sells. Each orientation is legitimate. Each produces a different shape. But they’re all describing rooms in the same building.

The next two years are going to produce a steady stream of these. Something from Ahrefs next quarter. HubSpot after that. Some AI-native tool nobody’s heard of yet. Each one will claim a framework. Each one will stake out vocabulary. Each one will have internal logic that works within its scope. Each one will be a slice of the retrieval system.

That’s fine. Multiple frameworks mean multiple vocabularies for the same underlying reality. More ways for different audiences to find their way in. But it also means the field needs a concept for the thing all those frameworks are slicing.

Call it what you want. SEO. GEO. AEO. AI visibility. The system underneath doesn’t care about the labels. The retrieval mechanisms operate whether or not anyone has a framework for them. Frameworks just help people see the system clearly enough to do the work. The question is how completely you map it.

Three groups looked. Three groups found layers. The layers are real.

The layers don’t care what you call them.