Why GEO, and why now?
When a prospect asks ChatGPT "which accountant should a small business in Nice work with?" or reads the generated answer at the top of Google (AI Overviews), they no longer scan a list of ten blue links: they read a written answer that cites two or three businesses. Either yours is one of them, or this channel is invisible to you, and it is the one growing fastest.
The good news for small businesses: unlike classic SEO where big budgets dominate, GEO mostly rewards clarity. A well-structured ten-page website can get cited ahead of a competitor ten times bigger whose site is unreadable for AI.
Does GEO replace SEO?
No, and that shortcut is an expensive one. AI engines do not start from a blank page: to answer, they look for candidate pages in a search index, then assemble extracts from them. Google says it explicitly about its own AI features: the same crawling and indexing rules apply as in ordinary search. A page Google has not indexed has no chance of being cited in a generated answer.
So the foundation is shared: an accessible website, indexed pages, clear titles, content that really exists in the HTML. What changes is the goal:
- SEO aims for the click. You are one line in a list of ten, and you win when the visitor picks you.
- GEO aims for the citation. You are one sentence inside an answer, and a prospect can reach you already knowing your specialty and your price range, without ever having opened your website.
The practical consequence is about measurement. Growing AI visibility does not show up first in your visitor count: it shows up in calls and form submissions from people who quote details back at you that they never read on your site, or who open with "ChatGPT told me about you". Write those down somewhere. In a small business, that log is usually the first usable GEO indicator you will have.
Condition 1: be readable by AI
Before an AI can cite you, it has to be able to read your website. This is the prerequisite that fails most often, and the silliest one to fail:
- Your robots.txt must not block AI crawlers. GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot and Google-Extended must be allowed in your robots.txt. Many websites block them without knowing it, often because of a default setting. Our dedicated guide: block or welcome AI crawlers?
- Your content must exist in the HTML. If your text only appears after JavaScript runs, some crawlers will never see it.
- The technical fundamentals still matter: a website served over HTTPS, decent response times, an up-to-date sitemap.
Condition 2: be understandable
Reading is not enough, the AI has to understand who you are, what you sell, and where. It is an ambiguity problem: "Martin & Fils" could be a bakery, a law office or a plumber. Three tools remove the doubt:
- Structured data (JSON-LD): a few lines of code that state in black and white your name, your business, your address, your opening hours, your offers, in the format machines read natively.
- The llms.txt file: a factual summary of your business placed at the root of your website, designed for AI. Our complete llms.txt guide explains what to put in it, and our free generator writes it for you.
- Consistent signals: same name, same address and same description on your website, your Google listing and your social profiles. Every inconsistency plants a doubt in the AI's mind.
Condition 3: be citable
AI systems build their answers by assembling passages. A passage only gets picked up if it stands on its own: a clear, complete statement that makes sense out of context. In practice:
- Answer questions in the very first sentence of each section, then elaborate. A section that opens with three paragraphs of context will never be cited.
- Structure content as questions: a well-built FAQ is ready-to-cite material for AI, on top of being useful to your visitors.
- Give numbers and facts: "on-site within 48 hours", "120 clients served since 2019", "from €90". AI prefers citing verifiable figures over slogans.
Condition 4: be credible
With equivalent passages, the AI cites the source showing the most trust signals: an identifiable author or company (an about page, complete legal notice), proof of real-world experience (case studies, reviews, dates), and mentions of your brand somewhere other than your own website. It is Google's E-E-A-T criteria transposed to the AI world, and it is the condition that takes longest to build: start with the first three.
How do you know if AI already mentions you?
You do not need a paid tool to find out. Two checks are enough, and they fit in a morning.
Ask your customers' questions, not your own name. Searching for "[your company]" proves nothing: give an AI your name and it will find you eventually. Ask instead the five questions a prospect actually asks, along the lines of "which plumber can fix a leak on a Sunday in Nice?" or "how much does bookkeeping cost for a small limited company?", in a fresh conversation and while logged out, on ChatGPT, Perplexity and Google. For each answer, note three things: are you cited, which companies are cited instead of you, and which pages are given as sources. That third point is the most useful one: the cited sources are often directories, comparison sites or the local press, in other words places where your absence can be fixed in a few hours.
A single test proves nothing: answers vary from one session to the next and from one user to another. It is the repetition, a month apart and on the same written-down questions, that reveals a trend.
Then check whether the crawlers actually come. Your host's access logs list automated visitors. Look for GPTBot and OAI-SearchBot (OpenAI), ChatGPT-User (the live fetch when someone opens your page from an answer), ClaudeBot (Anthropic) and PerplexityBot. One detail catches a lot of people out: Google-Extended is declared in robots.txt but never appears in your logs, because Googlebot does the actual crawling on its behalf. If none of these crawlers has been through in months, your problem is not how your pages are written, it is further upstream, back at condition 1: nothing gets cited that never gets read.
Where to start?
In order, because each condition depends on the previous one:
- Check that AI crawlers are allowed on your website (10 minutes).
- Add an llms.txt file and basic structured data (1 hour).
- Rewrite your key pages answer-first, with a FAQ and hard numbers (1 day).
- Keep strengthening your trust signals over time.
The first two points can be checked automatically: our audit devotes 23 checks to AI visibility (crawler access, llms.txt, structured data, citable passages, hard numbers, E-E-A-T signals) and tells you exactly where you stand.