It all starts from this statistic: only 10 to 12% of AI results come from Google's top 10.
Because it was possible to manipulate Google (that's changing). That is no longer the case with LLMs, which don't have that bias: they prioritise what comes up often in quality "contexts".
Almost everyone has tried typing a specific query into ChatGPT hoping to find their site.
But while we were in the top 3 on our best keyword, our site isn't cited once in the AIs. So should we conclude that SEO and AI don't work the same way?
Towards the end of the 3 pillars of SEO
For a long time, you were told over and over that there were 3 pillars. Easier to sell, easier to package, it was a way of getting the client to understand how SEO works.
To keep it simple, SEO based on the 3 pillars looks like this:
- 1. Content: create as many pages as possible on as many keywords as possible
- 2. Technical: boils down mainly to having an optimised sitemap and robots.txt
- 3. Popularity: buy backlinks, everywhere, all the time, on every query, without being fussy about where the links come from
The problem that surfaced very quickly was an ideology of hacks over quality. You had to produce more and more, on more and more keywords, whatever the method.
The keywords weren't always the most interesting ones, just the ones with the biggest volume.
When people talked about authority, one of the ranking criteria, they reduced it to buying backlinks, forgetting how important it is to build a brand community to, for example, increase the number of citations and reviews on social networks.
And then LLMs arrived.
For many, it was seen as a catastrophe. Everything had to be rethought, processes changed.
The reason? LLMs prioritise the most-cited sources. A site's authority is determined by how often a fact or brand name appears in its training corpus.
Through optimisation, your site may have taken first place on a keyword while being completely invisible to LLMs.
A shift has been under way for a few weeks, and it could speed up very quickly with a new way of approaching these 3 pillars:
- 1. Content: work on search micro-intents + adopt a different editorial angle + expert content within a semantic framework
- 2. Brand building: add value to your topic so that people share your work (forums, social networks, etc.)
- 3. Technical: focus mainly on indexing URLs and setting up structured data (a pillar that therefore becomes more accessible)
OK, but do I need to optimise my site for AI?
To generate conversions in the short term, the answer is no. But…
But what's happening through AI ranking shows a new way of approaching SEO as a whole.
It's not simply an algorithm change, it's a transformation of search.
- We are moving from keyword search to natural language.
- We are moving from backlinks to citations.
- We are moving from optimised content to content that aims to engage and is ultra-niche.
Two charts to back up the case for AI:
- Blue (Traditional Organic Search): the value of classic organic search is gradually declining.
- Red (LLMs, incl. Google AI): the share of value generated by AI rises sharply, especially after 2026.
- Yellow (Untracked/Misattributed LLM Impressions): many impressions via AI are not measured or are wrongly attributed, but they are rising too.
- Green (Total LLM Value): the total value generated by LLMs grows exponentially and far exceeds organic search from 2026–2027.
A new SEO criterion
Google's model is now moving closer to that of an LLM. That is the main reason behind the arrival of Muvera, the American search engine's new algorithm (and a signal consistent with the updates that penalise sites stuffed with generated content).
Going beyond simple keyword matching to focus on the vectors close to the main keyword:
King vector: [0.8, 0.2, 0.7, 0.1, 0.9]
Queen vector: [0.7, 0.3, 0.6, 0.2, 0.8]
Green bean: [0.9, 0.3, 0.2, 0.1, 0.1]
The last one has a vector far from the first two (low cosine similarity), so it won't be considered in the answer.
The goal? To find the best information using the widest set of data (documents, images, videos, etc.), which are called "items".
Google has clearly understood that it had fallen behind, and is aligning with what LLMs already do. We are seeing a move from "word matching" to "understanding meaning". The search engine no longer looks only for matching keywords, but for the overall meaning and the intent.
The operational answer comes down to one approach: splitting the market into exploitable semantic micro-territories, the method I use on all my sites so I no longer depend on volume.
Two main areas of work to appear in LLM results:
1. Build your topical authority
Google prefers specialists to generalists. Choose one or two main keywords, and become a specialist. Work on FAQs split by decision-stage keyword. Create AI tools / simulators to offer something other than purely static content and to rank on action keywords.
By action keywords, we mean this type of query, ideal for LLMs:
- how to do [task]
- how to use [tool]
- how to install [software]
- how to create [site, document, campaign]
- learn [skill]
- step by step [action]
2. Identify bottom-of-funnel long-tail keywords
Before tackling this step, read my method for spotting the queries LLMs rephrase. It's the prerequisite that makes the difference.
Forget "payroll software", and do everything you can to rank on "payroll software for small construction firms". What counts isn't search volume, but:
- Purchase / decision-making intent
- The level of specificity
- A direct link with your product/service
- The presence of a CPC
We'll talk about micro-intent to define very specific needs that are either transactional or bottom-of-funnel informational.
E.g.:
- How to move from "solution X" to "solution Y"
- "Buying guide" pages: to rank on "how to choose {product}"
- "Comparison" pages: to rank on "comparison / vs {product A vs product B}"
- "Solutions" pages: to rank on "{problem} + solution"
What is happening to SEO is going in the right direction: quality content that talks to users to answer a real need. To start this shift without reinventing the wheel, start from my library of AEO prompts.
The classic SEO vs AI SEO question isn't really interesting, because in the end the two will become one.