GEO in 2026: why ChatGPT will never cite you (and the 4 levers to change that)

Google ranks, ChatGPT cites — and that changes the entire mechanic

When Google displays a results page, it offers a list — ten links, sometimes fifteen, from which the user chooses. Your place in the list matters, your relative position matters, but you exist. Even at 8th place, you are visible if someone scrolls far enough. That is the logic that structured SEO for twenty years: get into the list, climb the list.

An LLM does not work like that. When ChatGPT answers a commercial question — "who does external wall insulation in the Gironde", "who is the best carpenter in Le Haillan", "employment law specialist in Bordeaux" — it synthesises a response citing one or two sources maximum. No list of ten, no second page. You are the cited source, or you do not exist for that user. The shift from ten results to a single citation radically changes the rules of the game.

A second fundamental difference: Google indexes what it finds on the web and ranks it by relevance and authority. LLMs, on the other hand, have fixed knowledge from the moment of their training, which they have been supplementing for about a year with real-time web searches. This dual memory — initial training + live search — means you must exist in both. Appearing only on your own website is no longer enough.

My recommendation: above all, run the concrete test. Type into ChatGPT, Perplexity and Gemini the three commercial queries that bring customers to you. Note whether your name appears, which competitors are cited, and which sites the models reference as sources. That snapshot of your current absence is your starting point.

Why an LLM cites one source rather than another

LLMs have an implicit notion of citability: the ease with which a source can be attributed without risk. They favour content whose origin is clear, dated, signed by an identifiable entity, and rephrashable without ambiguity. A blog article signed by a named author, dated, with a short biography on the site, is ten times more citable than anonymous text published without context — even if the latter is technically of higher quality.

The concrete criteria that boost your citability are few but cumulative. A proper name identified as author (yours, your managing director, your spokesperson). A short, factual biography accessible on the site. Dated figures with their source. Perfect consistency between what is read on your site and what is read elsewhere (LinkedIn, press, trade directories). And above all, verifiable claims — an LLM prefers to cite a source it can cross-reference with two other mentions on the web over an isolated source.

What tanks your citability, on the other hand: purely marketing content ("we are sector leaders"), superlatives without proof ("best provider in the Gironde"), corporate pages with no author, figures without a source, commercial promises that look copy-pasted from a competitor. Anything an LLM cannot verify or properly attribute gets ignored.

My recommendation: audit your key pages with one simple question. "If a journalist needed to quote a passage from this page, could they attribute it without risk?". If the answer is no — text too generic, no author, no concrete data — the page has no chance of being cited by an LLM. Rewrite it to make it citable, even at the cost of a less commercial tone.

ChatGPT does not only read your website — and that changes everything

Here is the mechanic that most SEO providers have not yet grasped. When an LLM builds its response, it does not rely solely on your website. It cross-references mentions of your business across the entire indexable web: Wikipedia, local and national press, trade directories, LinkedIn profiles of your team, guest articles, podcasts, YouTube transcripts, specialist forums, press releases, citations in other blogs. It is this multi-platform footprint that determines whether it has enough signal to cite you.

Practical consequence: even an excellent website, if it stands alone, is no longer sufficient. If you appear nowhere else on the web — no press article, no mention in a serious trade directory, no active LinkedIn profile for your managing director, no interview on a sector blog — LLMs have no way to verify your existence. They then prefer to cite a less competent competitor who is better distributed in their knowledge corpus. It is harsh, but it is mathematical.

Building a multi-platform footprint requires patience and a coherent strategy. Not a presence everywhere, but a consistent presence on the right channels. For a small business, this means concretely: a properly managed director LinkedIn profile, two or three mentions per year in local business press, a presence in the two or three serious trade directories in your sector, and ideally guest articles on recognised sector blogs. That web of mentions is what makes your name citable.

My recommendation: list the five places where your business should appear on the web but does not. A leading trade directory, local press, your director's LinkedIn profile, a recognised sector blog, a specialist collaborative encyclopaedia. Those are your five multi-platform footprint priorities for the next twelve months. It is slow, it is patient work — and it is exactly what LLMs value.

Write so an LLM can extract your paragraphs — not so a human reads your entire page

The fourth lever, the one nobody states directly, is the structure of the content itself. When an LLM answers a question, it does not read your entire page — it extracts two or three passages to rephrase. The more extractible your content is, the greater its chances of being cited. This is a different writing logic from classic SEO, which primarily aimed at length and keyword optimisation.

Concretely, content designed to be dissected has five characteristics. Titles that answer specific questions rather than marketing headlines. Short, self-contained paragraphs — each paragraph must be readable in isolation without the context of the preceding ones. Dated, quantified answers that can be extracted verbatim. A structured FAQ section with clear questions and answers. And a vocabulary from your trade that matches what a user types into a conversational engine, not your internal jargon.

The classic mistake is publishing well-written long-form articles that are hard to extract — where each paragraph depends on the previous one, factual information is diluted in prose, and figures appear mid-sentence without clear context. A human can read that content with pleasure. An LLM, however, moves on and prefers to extract a more structured text from a competitor — even if the quality is lower. Structure trumps beautiful prose in the GEO world.

My recommendation: test it yourself. Take one of your best articles, paste its content into ChatGPT and ask it to extract the three main factual pieces of information. If the model struggles or reformulates off-target, your content is not extractible. Rewrite it with more self-contained paragraphs, dated figures, and a FAQ. You will see the difference in the very next test.

Four levers, a new playing field.

GEO does not replace SEO — it stacks on top of it. A site invisible on Google will not be any more visible in ChatGPT. Technical foundations remain the base; content and popularity remain the two floors above.

If you are starting from scratch, read the SEO Holy Trinity to understand the order of work. GEO is the floor you add once the three below are already solid.

END · 18.05.26 ↑ Back to top