Why Do Senior SEOs Say There Is No Such Thing as LLM Optimisation? Search Engineering Masterclass, Day 4

Key Takeaways

1. AI optimisation and SEO are not opposites. The inputs are largely the same. The evaluation layer sitting on top of those inputs is what changed.
2. SEO stopped being keyword matching long before AI arrived. Modern SEO has been about meaning, intent, and entities for over a decade.
3. AI does not randomly predict words. It first runs a situation analysis, then uses probability to choose what to say within the bounds of that reasoning.
4. Bad SEO will never win in AI search. Good SEO is necessary but no longer sufficient.
5. The shift is not from SEO to something new. It is another evaluation layer added on top of an already evolved discipline.

Here is the YouTube link to the search engineering masterclass, day 4

What is the biggest misconception about AI search?

A few weeks ago, FTA's team was pitching when one of the client-side gentlemen made a comment that seemed reasonable on the surface. He said AI is just a predictor. It learns from the web and keeps predicting word by word.

He was not entirely wrong. AI works on word-level probabilities at the generation layer. The problem is what people conclude from that statement. The leap most teams make is that if AI is just predicting words, then AI optimisation must be a completely different discipline from SEO. That conclusion is where the entire industry's confusion starts.

Here is the uncomfortable truth that breaks the assumption. If AI optimisation were genuinely a separate discipline, then good SEO should already be failing across the board. It is not. 

Many of the people who understand AI search best today are senior SEOs who have been in the industry for over a decade. They are quietly succeeding in AI search using the same fundamentals they have been refining since 2010.

Why do senior SEOs say there is no such thing as LLM optimisation?

Many experienced SEOs have started saying it openly: there is no AEO, there is no LLM optimisation, it is just SEO. They are not wrong, and the reason they are not wrong requires fixing a common misunderstanding about what SEO actually is.

SEO stopped being simple keyword matching a very long time ago. In the late 1990s, when Yahoo was leading the search market, ranking was indeed about keyword density, stuffing, and basic word-to-word matching. 

The black-hat era of cloaking and white-on-white text emerged during this period. By 1999, when Google was getting into full throttle, the game had already moved on. 

Google introduced PageRank, which evaluated pages based on their relationships rather than on keyword density alone. That is what backlinks were always about.

The 2010 and 2011 Panda and Penguin updates pushed evolution further. Google started understanding the meaning behind queries, the intent behind searches, knowledge graphs, entities, topics, and content context. 

By the mid-2010s, SEO was already a meaning-based discipline, not a keyword-based one. Useful content, clear page structure, internal linking, topical authority, digital PR, and brand mentions were the standard playbook long before AI search emerged.

When senior SEOs look at LLM optimisation tactics, they see the same inputs they have been working with for years. They are right to call it out. The inputs really are largely the same.

Where does the actual difference between SEO and AI search show up?

The confusion is not about the inputs. The confusion is about assuming the goal stayed the same.

Traditional SEO, even at its most advanced level, was always optimising for retrieval. The questions being asked were straightforward. 

Can the search engine find this page? 

Can it rank this page? 

Will the user click on the result and land on it? 

AI search changes this last step entirely. AI does not stop at showing options. It decides the answer itself.

Here is how the goal shifts across the two systems.

This table shows what each system aims to achieve once your content is found:

The inputs may be familiar. The output decision is no longer being made by the user. It is being made by the model.

Does AI actually predict words randomly, or is something else happening?

This is where the most damaging assumption needs to be corrected. AI does work on probability, but not in the way most people imagine.

If models were genuinely guessing words at random, ChatGPT would not be able to answer complex questions consistently. The reality is that before generating any answer, AI runs an internal situation analysis. 

The model considers what the user is actually trying to solve, which constraints matter for the specific question being asked, and which explanation would make the most sense given the conversation and session context.

Only after that internal understanding does the generation layer start. Even then, the model has two priorities. 

For Wikipedia-style factual questions, such as "What is a DMAT account?", the model retrieves the answer directly from its training data without needing to consult real-time sources. 

The moment the question shifts to something like, " Which is the best DMAT platform to open an account in 2026, the model goes to the live web because the answer requires current information.

Probability is how AI chooses what to say. Reasoning is how it decides what makes sense for the situation. Both layers are running. Most teams optimise for the first and forget the second exists.

Why does SEO no longer guarantee a mention in AI answers?

Because the building blocks look familiar, but the evaluation has changed.

SEO has historically asked one question of every page. Is this page relevant and authoritative enough to rank for the keyword someone just searched? For the DMAT account, the SEO playbook is well understood. 

Most top-ranking pages will be titled along the lines of " What is a DMAT account or definition of a DMAT account. SEO teams structure the content to match the informational intent Google has identified for that query.

AI asks a fundamentally different question. Can I use this source to explain the answer clearly and confidently? 

This source might be the model's internal training data for a stable definitional query, or a live web source for a real-time question. In either case, the evaluation is no longer about ranking position. It is about whether the content fits cleanly into the reasoning path the model is building.

Here is where the strange thing starts happening. A page can have great SEO. It can rank well. The brand can have genuine authority in the industry. And the page can still never appear in AI answers. The SEO is not bad. 

The content simply does not fit the reasoning path the AI is constructing for that specific query at that specific moment.

AI is constantly evaluating two things. Does this help me explain the situation accurately, and is this source trustworthy in the current context? The trustworthiness layer is often solved automatically for established brands with Wikipedia presence, consistent press coverage, and authoritative third-party mentions. 

Where most brands fail is the situation analysis layer. Does this brand actually belong in the answer being constructed right now? If the answer is no, the brand gets ignored, regardless of how strong the SEO foundation is.

What is the right way to think about SEO in the AI era?

Stop framing it as SEO versus LLM optimisation. The framing itself is the mistake.

The accurate way to see it is this. SEO evolved beyond keywords a long time ago. AI search is the next evaluation layer added on top of an already-evolved discipline. Bad SEO will never win in AI search. Good SEO is necessary, but it is no longer sufficient on its own.

Once the picture is reframed this way, the path forward becomes clearer. Continue investing in everything that has been working. Useful content, clear structure, internal linking, topical authority, digital PR, and brand mentions. None of that is optional. 

What needs to be added on top is a layer of work focused on whether your content fits the reasoning path AI is building, not just the ranking criteria Google is applying.