AI search ranking: what changes, what matters, what to do

Why classic SEO is no longer enough

Classic SEO rests on a simple principle: rank as high as possible in a list of results to generate clicks. This model is evolving.

AI engines ChatGPT, Perplexity, Gemini and Copilot short-circuit this logic. When a user asks an AI a question, they receive a direct answer. They do not browse ten links as they would on Google. They often click on none at all.

The data confirms this: the presence of an AI summary at the top of Google results is now correlated with a 58% drop in click-through rate on the first organic result. Meanwhile, AI referrals to websites jumped by 357% year-on-year by mid-2025.

Two opposing dynamics, but one shared conclusion: visibility now plays out on two distinct fields. Being absent from AI responses means becoming invisible to a growing share of your audience.

How AI engines select their sources

Unlike Google, which ranks pages, generative engines construct an answer. To do this, they rely on a mechanism called RAG (Retrieval-Augmented Generation): they query the web in real time, retrieve the most relevant content, project it into a vector space, then synthesize a response citing only 2 to 7 sources deemed reliable.

The selection criteria are not the same as in SEO. A site ranking first on Google may be completely ignored by an AI if its content is not structured, not factual, not synthesizable. Conversely, a lower-ranked page that is perfectly aligned with the intent of the query can become a regularly cited source.

AI referencing follows its own rules. Understanding them is the first step.

What changes compared to traditional SEO

The five pillars of AI referencing

There is no direct submission to an AI index the way one submits a sitemap to Google. Visibility in generative engines is built indirectly, through the quality and structure of your content. Here are the levers that have proven their effectiveness.

1. Structure for citability

LLMs break each page into semantic blocks (chunking) following the HTML hierarchy. A well-phrased H2 or H3 heading creates a usable block. Unstructured content generates incoherent chunks and a near-zero probability of being cited.

The optimal format identified by GEO research: blocks of 40 to 60 words beginning with a direct answer, short enough to be integrated into a generated response, dense enough to be informative.

2. Address the complete intent

When an LLM receives a prompt, it generates several derived queries to explore related dimensions (fan-out queries). Content that only covers the surface question will systematically be set aside in favor of a source that addresses the entire semantic field.

Anticipating these sub-questions and answering them within the same content is one of the most effective levers for being cited.

3. Provide factual and verifiable data

LLMs are trained to distinguish facts from opinions. Content loaded with marketing superlatives carries no weight. A precise, dated and sourced statistic becomes a preferred reference.

Original data (internal studies, exclusive figures, quantified feedback) is the content most naturally cited by AI, as it provides value that other sources cannot reproduce.

4. Strengthen E-E-A-T signals

Experience, Expertise, Authority, Trustworthiness: these four criteria defined by Google are precisely what generative engines look for to identify trustworthy sources.

• Identified author with recognized expertise in the field
• Information corroborated by other authoritative sources
• Clear editorial policy and thematic coherence of the site
• Mentions and citations in industry media or databases

5. Deploy structured data

Schema.org markup (FAQPage, HowTo, Article, Organization) facilitates information extraction by generative engines and significantly increases citability. It is one of the few levers directly readable by LLMs during crawling and synthesis.

Which AI platforms to target first?

The core principles apply to all these platforms. Optimizing for one naturally strengthens visibility on the others.

The most common mistakes

• Neglecting classic SEO. GEO does not replace SEO, it complements it. A poorly crawled or low-authority site will not be cited by AI. Technical foundations remain essential.
• Producing superficial content. LLMs favor depth. A 300-word article with no data or real structure stands no chance against well-organized expert content.
• Writing for the keyword, not the intent. Repeating a target term is not enough. You need to cover the entire semantic field associated with the question the user asks the AI.
• Ignoring external signals. Mentions of your brand in third-party sources matter more in GEO than in classic SEO. Press, studies, citations in reference content: these signals build the credibility perceived by LLMs.
• Thinking it is too early. In Europe, GEO competition is still low. Companies that position themselves now are building a lead that will be difficult to close in twelve to eighteen months.

How to measure your visibility in AI

There is no equivalent to Google Search Console for LLMs. AI visibility is probabilistic: the same page may be cited or ignored depending on the prompt, the platform and the model version.

Concrete methods do exist to track it, however:

• Simulated prompt audit: querying ChatGPT, Perplexity and Gemini on the priority queries of your sector and checking whether your domain appears in the responses.
• Conversational share of voice: frequency of citation across a corpus of prompts representative of your audience.
• Citability score: analysis of structure, direct answers, structured data and E-E-A-T signals present on each page.
• Adjacent intent coverage: identification of sub-questions not yet covered relative to the typical fan-out queries of your target queries.

These metrics form the basis of a structured GEO audit — the first step before any optimization.

SEO and GEO: complementary disciplines

One essential clarification before concluding: AI referencing does not replace classic referencing. It adds to it.

A well-optimized SEO site is a more credible source for AI. Content designed with GEO in mind is better understood by classic engines. The two disciplines feed each other.

What changes is the priority. The question is no longer solely about which keyword to rank for, but how to make each page of your site clear enough, structured enough and reliable enough to be selected as a source by a language model generating a response on behalf of your future customer.