SEO and AI: 8 Essential Strategies to Adapt to LLMs

At SMX Paris 2026, Carolyn Shelby, SEO expert at Yoast, discussed the major transformations that artificial intelligence (AI) imposes on SEO. AI systems, which do not rank pages but select fragments of information, are changing the game. Shelby presented eight recommendations for adapting SEO strategies to this new reality.

Structure Content for AI Extraction

Write to be Extracted, Not Just Ranked

While ranking on Google remains a prerequisite, it is no longer sufficient to be cited by AI systems. These systems do not take entire pages but select fragments such as paragraphs, lists, or definitions to construct their responses. Carolyn Shelby emphasized the importance of making content extractable: “The general rule is this: if your content cannot be extracted, it will not be cited. You can be first, you can be second. If your content cannot be extracted, it will not be used, and you will be invisible.” Now, it is the fragments of pages that compete, not the entire pages.

One Topic Per Paragraph

Language models (LLMs) break pages into segments and evaluate each separately. A paragraph that tries to explain, convince, and answer a question simultaneously dilutes the signal. Shelby recommends: “One master per paragraph. Don’t try to do too much with the space you have.” She reminds us that machines convert all text into numbers, and to achieve the best score, one must simplify and clarify.

Make Content Technically Accessible to LLMs

Client-side JavaScript rendering and content placed behind a paywall or login form are invisible to LLMs. “No one wants to spend computing power to translate this JavaScript,” Shelby warned, particularly targeting fashion and luxury sites. She cautions tech companies against the temptation to block access to their documentation: “If you do not provide this information, LLMs will get it somewhere else. And that somewhere will be Reddit or Stack Overflow. If you want Reddit users to answer your technical support questions, that’s your choice.”

Don’t Rely on schema.org for LLMs

Contrary to a common belief in the SEO community, LLMs do not read structured data like that from schema.org. Shelby states: “If someone tells you that LLMs can read schema, they are lying to you. They don’t want to do it. It costs them time and money. They don’t care.” What matters for LLMs is the visual structure of the content: short titles are interpreted as headings, and the hierarchy is deduced from the position on the page, not from HTML tags.

Build Authority and Control Brand Narrative

Become the Primary Source

AI systems favor first-party sources, such as official documentation, original research, and information published directly by the company in question. Shelby summarizes: “The systems will choose the source that creates the least risk of being wrong.” She illustrates this risk with the example of a Chanel advertisement seen at Charles de Gaulle Airport, which simply stated “I have two Chanels” without any product description. “If Chanel does not provide a structured description of its watch, the LLM will look for it at Bob’s Watch Shop. If the brand wants to control the narrative, it must provide structured content on its own site.”

Repeat the Same Message Everywhere

Consistency of messaging across all channels reinforces trust with AI systems. Shelby cites Disney, whose LLMs consistently use the phrase “magical experience”: “Why would the LLM say that Disneyland Paris is a magical experience? Because Disney says it everywhere: on its site, in its ads, in its texts, on its billboards. You too must have this discipline. Your message must be repeated in the same way wherever a user might see it, as all these signals are validation for search engines and LLMs.”

Invest in YouTube Video, Especially for Gemini

Google, via Alphabet, has access to the transcripts of all YouTube videos, making it a direct source of information for Gemini. Shelby advises: “If you are not making videos right now, you should really consider it, because the transcript will be used as a potential source of information. They are just fragments. It doesn’t matter that it’s not a web page.” She adds, pragmatically: “It all comes down to money. Google makes more when you watch videos, so it will push videos after every question asked by the user.”

Fill Information Gaps to Limit Hallucinations

LLM hallucinations are not random bugs: they occur when the available information is incomplete. Shelby explains: “LLMs do not invent things just for fun. They invent because they do not have enough information. It’s like your memory: when there are gaps, your brain fills them in to create a complete narrative. That’s all LLMs do.” The solution: provide systems with all the data they need, in a format they can directly exploit. The ability of content to be selected by LLMs directly depends on its clarity, structure, and completeness.

In conclusion, Carolyn Shelby summarized the issue in one sentence: “AI has not broken SEO. AI has simply revealed what SEO was always meant to be. Not tricks, not loopholes, not page factories. SEO has always been meant to make knowledge understandable, reliable, and findable.”