Lesson Transcript
Expand the Complete Lesson Transcript
[00:00:00] Welcome to AI Search Optimization and Agentic SEO course. I’m Beatrice Gamba, an expert in innovation and semantic search. I work at WordLift, and I’ve been building and testing AI-driven SEO systems for years, from knowledge graph introductions to RAG pipelines and entity recognition at scale. Before we get into the tactics or tools, I want to be precise about the problem we are actually solving because most of what is being currently called AI SEO right now is conceptually wrong.
People are applying the old mental models to systems that work in fundamentally different ways, and this course aims to correct that
Let’s define the term we will use throughout every lesson because precision here matters
Agentic search in discourse refers to the search systems that autonomously retrieve, evaluate, and synthesize information from multiple sources to generate a response. Let’s notice the main [00:01:00] concepts: autonomously, retrieve, evaluate, synthesize. None of these words appear in the description of a traditional search engine because a traditional search engine retrieves and ranks, of course, but then the human evaluates and the human synthesizes by reading multiple pages.
In an agentic system, the machine does all the three steps before the user even sees anything, and that single architectural reference and difference is why everything you learn about optimizing for clicks, rankings, and bluelinks requires rethinking, which doesn’t mean that what we’ve learned until now has to be forgotten, but just rethinking
There are four properties that define whether a system qualifies as agentic search. The autonomous decision-making. The system selects which sources to read and which claims to extract without human input [00:02:00] at each step. Second, multi-source synthesis. It does not return one document, but it combines information across multiple documents into a single generated response.
Third, claim-level evaluation. This is the one that changes SEO most fundamentally. The system does not assess your page as a whole. It assesses the individual statements, the individual facts. A page that ranks well overall can still have its specific claims ignored or overridden. Fourth, selection before visibility.
By the time a user reads an answer, the system has already decided whether your content contributed to it or not. You never appear in a list for the user to choose from. You are either selected or you are not selected
These are not hypothetical architectures. They are deployed today. ChatGPT with web browsing and source citations, Perplexity’s [00:03:00] answer engine, the Google AI Overviews, the Microsoft Copilot, each of these implements the four properties I just described through their underlying retrieval and ranking mechanism differ in documented ways we will examine in later lessons
Equally important is what does not qualify as agentic search
Traditional search engines for starters like classic Google that returns a current, a ranked list. The human reads and decides it’s not agentic. Simple chatbots. They follow scripted response trees. They do not retrieve or synthesize. They’re not agentic. Recommendation systems like Netflix or Amazon, they filter and rank existing items, so they do not generate new synthesized claims.
Not agentic. Basic voice assistants, they execute single fact lookups or device commands. Not agentic. The test is simple. Does the system generate a new synthesized response [00:04:00] by combining claims from multiple sources without human intervention at each step? If not, it falls outside our definition
One sentence captures what this entire course is built around. If the system just retrieves and ranks existing content for humans to evaluate, it is not agentic search
The primary systems we focus on are LLM-based search products, meaning ChatGPT with search, Perplexity, Claude with search, Gemini. Secondary focus covers search engine AI features that are integrated into traditional engines and enterprise AI assistants that operate over internal knowledge bases. All three categories apply the same four properties but with different retrieval architectures and different signals that they respond to
We also cover our secondary systems production RAG implementations, AI agents with web access and tool calling, and knowledge graph enhanced search. [00:05:00] These are the infrastructure layers that power the primary systems we just saw. Understanding them gives you a diagnostic capability rather than just surface level tactics, which is not what this course is aimed at.
Every agentic search system runs through four decision stages.
Query understanding, it’s where the system classifies intent and decomposes complex queries into subtasks. Retrieval, where it decides which sources to access, how many candidates to pool, and applies initial relevance filtering. Evaluation, where it scores source credibility and checks claim consistency across sources.
And synthesis, where it decides which claims enter the final response, how to handle conflicting information, and whether to attribute sources explicitly. Each stage involves an autonomous decision. Each decision is a point where your content can be selected or excluded. This course teaches you to diagnose which stage is the problem and how to address [00:06:00] it precisely
So let’s dive into what you will be able to do after taking this course And these are four concrete capabilities. The diagnosis, you will be able to identify whether your content is falling at retrieval, evaluation, synthesis, and you will understand the difference between being indexed and being selected.
Content design, structuring information as atomic verifiable claims that agentic search can extract and reuse, including entity markup that supports multi-hop reasoning across sources. Measurement, we will build test query sets to track the selection rate across different agent systems, and you will be able to measure entity recognition’s completeness.
Being able to take strategic decision, you will be able to evaluate any new SEO claim or vendor recommendation against how these systems actually work architecturally. And you will separate durable optimization principles from tactics that depend on current product configurations
[00:07:00] At the end of the day, you want to be able to explain how agent systems select information, identify why your content is being excluded, designing content for direct agent consumption, measure performance in an agent-mediated search, and distinguish what matters from what is noise. Let’s get to work and see you in the next lesson
Lesson Slides
What’s Covered In This Lesson
- Definition of “agentic search” and why it differs fundamentally from traditional search engines
- The four properties that qualify a system as agentic search (autonomous decision-making, multi-source synthesis, claim-level evaluation, and selection before visibility)
- Examples of systems that are agentic search versus those that are not
- The distinction between primary systems (LLM-based search products) and the secondary infrastructure layers that power them
- The four decision stages every agentic search system runs through, from query understanding to synthesis
- An overview of the concrete skills and capabilities the course will build
Key Takeaways
- Agentic search systems retrieve, evaluate, and synthesize information autonomously—doing work that a human previously did when reading through ranked results
- Optimization approaches built around clicks, rankings, and blue links need to be rethought rather than discarded
- Evaluation happens at the level of individual claims, not whole pages, so a well-ranking page can still have its specific facts ignored or overridden
- Your content is either selected or excluded before the user ever sees an answer—there’s no list to appear in
- Each decision stage is a point where content can be included or left out, so diagnosing where the breakdown occurs matters more than applying generic tactics
- The goal is durable, architecture-based understanding that lets you evaluate any new tool or SEO claim critically