The Cognitive Architecture Taxonomy
Ever since the irruption of AI in Search went from being a novelty to a flood of constant change (just think of Google Bard in 2023 and now Gemini 3), I've had a deja vu.
Today, like 20 years ago when I was starting my professional life as an SEO, I find myself with a thousand tabs open in my browser on pages I probably wouldn't have imagined consulting.
Part of my day is spent studying new terminology, mechanisms, strategies, and tactics.
Studying, understanding the mechanisms, experimenting, "playing," having clear concepts and immediately clarifying misunderstandings; this is the key to overcoming discouragement, to avoiding the temptation to give up, and above all, to avoiding taking the claims of so-called experts (in reality, snake oil salesmen) as true and/or accepting as correct what are hallucinations generated by AI itself.
It's been two years of collecting notes, jotting down definitions, and asking questions to colleagues far more knowledgeable than I am, creating a body of information I turn to whenever doubts grip my mind.
What you're about to see below is an attempt to summarise the most important information gathered over the past few months, presenting it in a comprehensible yet precise manner.
A guide I wrote for myself, but I believe it could be useful to everyone - SEOs, marketers, decision makers, and so on - who don't necessarily have a deep, native understanding of how AI works in search and how it has come to completely change the way we work.
Table of Contents
Chapter 1 - The Model: Architecture and Mechanics of LLMs
This chapter explains the inner mechanics of Large Language Models, from tokens and embeddings to attention, positional encoding, and inference optimizations. It clarifies how these systems reason over context, highlighting both their strengths and their practical limitations.Chapter 2 - The Agent: Reasoning, Planning, and Action
This chapter explores Agentic AI, showing how LLMs become goal-driven systems that plan, reason, and take action using tools. It covers how agents use memory, tools, and structured reasoning frameworks like Chain-of-Thought, Tree-of-Thoughts, and ReAct to execute complex tasks.Chapter 3 - The Library: Neural Search and Retrieval Architectures
This chapter examines how LLMs access external knowledge through neural search and retrieval architectures. It explains core RAG patterns—from vector and agentic retrieval to graph-based and hybrid search—and compares dense, sparse, and advanced retrieval strategies.Chapter 4 - The Strategy: SEO for AI Search
This chapter introduces SEO for AI Search, shifting optimization from rankings to being retrieved, reasoned over, and cited by AI systems. It outlines the core levers—retrieval, reasoning, agency, and authority—that drive visibility inside LLM-powered search and assistants.Chapter 5 - The Technical Foundation
This chapter explains why technical SEO is the foundation of visibility in AI-driven search and retrieval. It shows how crawlability, structure, performance, and machine-readable signals determine whether AI systems can find, trust, and use your content.Chapter 6 - Content and AI Search
This chapter outlines how to design content for AI search using an entity-first, hub-based strategy. It shows how clear ontologies, semantic chunks, and reasoning-friendly structures help LLMs retrieve, reuse, and cite your content across RAG and answer engines.Chapter 7 - Amplification
This chapter explains how amplification builds trust and visibility in AI search by creating consistent, entity-rich signals across the wider web. It shows how mentions, communities, multimedia, and structured profiles reinforce brand authority and increase the chance of being retrieved, cited, and remembered by LLM-powered systems.
Introduction: The Shift to Cognitive Computing
The emergence of Generative Artificial Intelligence has precipitated a fundamental shift in the architecture of software systems, moving from deterministic, rule-based logic to probabilistic, cognitive computing. This transition has introduced a lexicon that is as dense as it is critical; terms such as "monosemanticity," "speculative decoding," and "agentic reasoning" are no longer the sole preserve of research laboratories but have become the foundational vocabulary of modern enterprise architecture. To navigate this landscape effectively, stakeholders must move beyond surface-level definitions and develop a mechanistic understanding of how these systems represent knowledge, reason over ambiguity, and retrieve information.
This guide serves as an exhaustive taxonomy and research synthesis, deconstructing the complex nomenclature of Large Language Models (LLMs), Agentic workflows, and Neural Search.
It is structured to provide a precise technical dissection of each concept, grounded in recent literature, while maintaining a narrative clarity that elucidates the interdependencies between these technologies.
By categorising these concepts into distinct pillars - The Model, The Agent, The Library, and The Strategy - we reveal how isolated terms converge into a unified framework for intelligent systems.
What
This guide starts from the mechanics: how models represent language (The Model), how agents reason and act (The Agent), and how retrieval architectures feed them fresh information (The Library).
The strategy sections connect those mechanics to SEO for AI Search across all major environments:
Search-embedded AI features (e.g., AI Overviews, AI Mode, Web Guide and their analogues elsewhere).
Stand-alone answer engines and assistants (ChatGPT, Gemini, Claude, Perplexity, etc.).
Private and domain-specific agents built by companies on top of foundation models.
The core idea is simple:
We are no longer optimising only for rankings and clicks, but for visibility, memory, and authority in the cognitive layer that all these systems share.
Why
As AI becomes the default interface for information discovery:
Retrieval decides whether you are even considered.
Reasoning decides whether your content is useful enough to be reused.
Agency decides whether systems can act through your tools and data.
Authority & Memory decide whether your brand is the one that models trust and recall.
This is true whether the user starts in Google, in a chat box in ChatGPT, in a Gemini “project,” in a Claude workspace, or in a Perplexity thread.
Action
Going forward, the practical implication is to treat your digital ecosystem as a knowledge system, not just a collection of pages:
Maintain a robust technical foundation (crawlability, structure, APIs).
Build entity-first content hubs with semantic chunks that models can retrieve and reason over.
Amplify your presence across the wider web so that your narrative is redundant and consistent wherever models look.
Continuously observe AI answers across platforms and use them as a feedback loop for both SEO and content strategy.
Do this, and you are not just chasing algorithms.
Do this, and you are designing for the cognitive architecture of modern AI across Google, ChatGPT, Gemini, Claude, Perplexity, and whatever comes next.
Curated selection of the sources used for creating this guide
What are AI agents? 🡪 https://www.ibm.com/think/topics/ai-agents
Awesome Papers for Understanding LLM Mechanism 🡪 https://github.com/zepingyu0512/awesome-llm-understanding-mechanism
Glossary of AI Terms (Plain Language) 🡪 https://umbrex.com/resources/ai-primer/glossary-of-ai-terms-plain-language/
What are word embeddings? 🡪 https://www.ibm.com/think/topics/word-embeddings
An intuitive introduction to text embeddings 🡪 https://stackoverflow.blog/2023/11/09/an-intuitive-introduction-to-text-embeddings/
Word Embedding models 🡪 https://dilipkumar.medium.com/word-embedding-models-38f667f752d9
Understanding Attention in Transformers: A Visual Guide 🡪 https://medium.com/@nitinmittapally/understanding-attention-in-transformers-a-visual-guide-df416bfe495a
Generalized Attention Mechanism: BigBird’s Theoretical Foundation and General Transformers Models 🡪 https://towardsdatascience.com/generalized-attention-mechanism-bigbirds-theoretical-foundation-and-general-transformers-models-9fb87bdac3b2/
Understanding LLMs Context Window and Working 🡪 https://www.matterai.so/blog/understanding-llm-context-window
What is positional encoding? 🡪 https://www.ibm.com/think/topics/positional-encoding
You could have designed state of the art positional encoding 🡪 https://huggingface.co/blog/designing-positional-encoding
Towards Monosemanticity: Decomposing Language Models With Dictionary Learning 🡪 https://www.anthropic.com/research/towards-monosemanticity-decomposing-language-models-with-dictionary-learning
Mapping the Mind of a Large Language Model 🡪 https://www.anthropic.com/research/mapping-mind-language-model
Unveiling Monosemanticity: Anthropic’s Groundbreaking Research on Large Language Models 🡪 https://wordlift.io/blog/en/unveiling-monosemanticity-anthropics/
An Introduction to Speculative Decoding for Reducing Latency in AI Inference 🡪 https://developer.nvidia.com/blog/an-introduction-to-speculative-decoding-for-reducing-latency-in-ai-inference/
Looking back at speculative decoding 🡪 https://research.google/blog/looking-back-at-speculative-decoding/
Caching 🡪 https://ai.google.dev/api/caching
Understanding the difference between context caching or prompt caching and semantic caching: A step toward optimizing RAG-based projects 🡪 https://medium.com/@wael-saideni/understanding-the-difference-between-context-caching-and-semantic-caching-a-step-toward-optimizing-1a2b44d25c12
Building effective agents 🡪 https://www.anthropic.com/engineering/building-effective-agents
What are AI agents? 🡪 https://www.ibm.com/think/topics/ai-agents
Handling Function Calls with Reasoning Models 🡪 https://cookbook.openai.com/examples/reasoning_function_calls
Writing effective tools for agents — with agents 🡪 https://www.anthropic.com/engineering/writing-tools-for-agents
Understanding How AI Agents Use Tools (with Working Example + Video) 🡪 https://community.retool.com/t/understanding-how-ai-agents-use-tools-with-working-example-video/58882
What Is Chain of Thought Prompting? 🡪 https://www.nvidia.com/en-us/glossary/cot-prompting/
Demystifying Chains, Trees, and Graphs of Thoughts 🡪https://arxiv.org/html/2401.14295v3
What is tree of thoughts prompting? 🡪 https://www.ibm.com/think/topics/tree-of-thoughts
ReAct Prompting 🡪 https://www.promptingguide.ai/techniques/react
Retrieval Augmented Generation (RAG) in Azure AI Search 🡪 https://learn.microsoft.com/en-us/azure/search/retrieval-augmented-generation-overview?tabs=docs
Graph RAG vs RAG: Which One Is Truly Smarter for AI Retrieval? 🡪 https://datasciencedojo.com/blog/graph-rag-vs-rag/
Traditional RAG vs. Agentic RAG—Why AI Agents Need Dynamic Knowledge to Get Smarter 🡪 https://developer.nvidia.com/blog/traditional-rag-vs-agentic-rag-why-ai-agents-need-dynamic-knowledge-to-get-smarter/
RAG vs. GraphRAG: A Systematic Evaluation and Key Insights 🡪 https://arxiv.org/html/2502.11371v1
Advanced RAG Techniques for High-Performance LLM Applications 🡪 https://neo4j.com/blog/genai/advanced-rag-techniques/
BM25 vs Sparse vs Hybrid Search in RAG — From Layman to Pro 🡪 https://medium.com/@dewasheesh.rana/bm25-vs-sparse-vs-hybrid-search-in-rag-from-layman-to-pro-e34ff21c4ada
Advanced RAG Techniques for High-Performance LLM Applications 🡪 https://neo4j.com/blog/genai/advanced-rag-techniques/
Hypothetical Document Embedding (HyDE) – A Smarter RAG method to Search Documents 🡪 https://www.machinelearningplus.com/gen-ai/hypothetical-document-embedding-hyde-a-smarter-rag-method-to-search-documents/
Advanced RAG: Query Expansion 🡪 https://haystack.deepset.ai/blog/query-expansion
Fuzzy Matching and Semantic Search: Improving Visibility in AI Results https://ipullrank.com/fuzzy-matching-semantic-search
Technical Foundations and Setup for AI Search https://ipullrank.com/technical-seo-for-ai-search
How AI Search Platforms Leverage Entity Recognition and Why It Matters https://ipullrank.com/ai-search-entity-recognition
GEO, AEO, LLMO: Separating Fact from Fiction & How to Win in AI Search by Lily Ray https://www.amsive.com/insights/seo/geo-aeo-llmo-separating-fact-from-fiction-how-to-win-in-ai-search/
Updated: The 5 Key AI Search vs Traditional Search Differences – A Visual Comparison https://www.aleydasolis.com/en/ai-search/key-traditional-search-ai-comparison/
The 10 Steps AI Search Content Optimization Checklist [With Examples + Google Sheets] https://www.aleydasolis.com/en/ai-search/ai-search-optimization-checklist/
Google AI Mode’s Query Fan-Out Technique: What is it and How Does it Mean for SEO? https://www.aleydasolis.com/en/ai-search/google-query-fan-out/
The Hidden Entity Layer of ChatGPT: From Named Entities to Products https://wordlift.io/blog/en/chatgpt-named-entities/
Why SEO Success Now Depends on Entity Architecture, Not Volume https://wordlift.io/blog/en/entity-architecture-seo/
From Retrieval to Reasoning: The Architectural Evolution of Information Systems for Large Language Models https://wordlift.io/blog/en/retrieval-evolution-for-large-language-models/
The Architecture of Authority: An 8-Task Workflow for Engineering a High-Performance Content Hub https://www.iloveseo.net/the-architecture-of-authority-an-8-task-workflow-for-engineering-a-high-performance-content-hub/
What framework to use for increasing visibility in AI Search https://www.iloveseo.net/what-framework-to-use-for-increasing-visibility-in-ai-search/
Why Informational Content Still Matters in the AI Search Era https://www.advancedwebranking.com/blog/informational-content-still-matters-in-ai-search-era
Web Guide, AI Mode/Overviews, and the Rise of AI Search SEO: How Query Fan-Out is Reshaping Google Search https://www.advancedwebranking.com/blog/ai-search-seo-web-guide-ai-mode-overviews
Article by
Gianluca Fiorelli
With almost 20 years of experience in web marketing, Gianluca Fiorelli is a Strategic and International SEO Consultant who helps businesses improve their visibility and performance on organic search. Gianluca collaborated with clients from various industries and regions, such as Glassdoor, Idealista, Rastreator.com, Outsystems, Chess.com, SIXT Ride, Vegetables by Bayer, Visit California, Gamepix, James Edition and many others.
A very active member of the SEO community, Gianluca daily shares his insights and best practices on SEO, content, Search marketing strategy and the evolution of Search on social media channels such as X, Bluesky and LinkedIn and through the blog on his website: IloveSEO.net.
