GEO (Generative Engine Optimization): get cited by AI search engines

What is GEO (Generative Engine Optimization), in practice?

A procurement manager in Lyon or Dubai asks ChatGPT which Tunisian manufacturers can supply food packaging to their market. The assistant does not return ten links: it writes an answer, names three or four companies and cites a few sources. If your company is not among them, that buyer will never visit your website.

GEO, short for Generative Engine Optimization, works on that exact moment. The goal is twofold: for your company to be among the sources used by ChatGPT, Gemini, Perplexity, Copilot or Google's AI Overviews, and to be described accurately there. You will also hear it called AI search optimisation or answer engine visibility.

The term comes from academic research. Researchers from Princeton and other institutions found that citing sources and adding verifiable data helps content appear in generated answers, while keyword stuffing does nothing for it (source: Aggarwal et al., arXiv:2311.09735). That is good news: GEO rewards clarity and evidence, not tricks.

How does GEO differ from traditional SEO?

Traditional search optimisation, covered by our SEO, GEO & local search practice, aims to rank your pages in a list of results. GEO aims to get your company into a written answer. The foundations are shared; the final stretch is not.

This is why a company that ranks well on Google can be missing from AI answers. The assistant does not copy the first page of results: it assembles a summary from sources it considers reliable, often comparison pages, directories or press articles. If your website says one thing and your LinkedIn page another, it will more readily pick a competitor whose facts line up.

GEO does not replace SEO. A poorly indexed or slow site has little chance of being cited; if that is your situation, technical & on-page SEO comes first.

How do you measure visibility in AI answers?

GEO has no Search Console, so we build the measurement with you, before any other work.

1. Choose the questions. Together we write the list of questions your clients actually ask: "who manufactures… in Tunisia?", "which firm should I consult for…?", "what is the difference between…?". The list stays the same throughout, otherwise reports stop being comparable.
2. Query several assistants. Each question goes to the main assistants, in a fresh session, in each of your languages.
3. Record what matters. Are you cited? In what position, with what description, with or without a link? Which competitors appear, and which sources does the assistant mention?
4. Calculate a citation share. The proportion of answers in which you appear, sometimes called Share of Model, becomes the baseline.

Answers vary between sessions: a single screenshot proves nothing, only the trend across several reports counts. The report also reveals errors to correct, such as an outdated address or a service you no longer offer.

What do we change on your site to get it cited?

The work follows a simple order: be readable, then be understood, then be quotable.

• Let the right crawlers in. Your robots.txt file decides which AI crawlers can read your pages. Major providers often separate training crawls from live search: OpenAI, for example, distinguishes GPTBot from OAI-SearchBot. You can refuse one and allow the other; it is a business decision, and we document it with you.
• Serve the text in the HTML. Some AI crawlers read the page exactly as the server returns it, without running scripts. Content loaded afterwards by JavaScript can stay invisible to them.
• Lead with the answer. Under a heading phrased like the client's question, the answer comes first and the detail follows. Tables, lists and short definitions can be extracted without distortion.
• Back up what you claim. Dated pages, a named author, sourced figures, explicit service terms: an assistant more readily quotes a statement it can verify.
• Describe the company for machines. Schema.org structured data (Organization, Service, FAQPage) links your name, services, contact details and profiles.

What about llms.txt? It is a proposed convention for guiding language models through a site. Its effect on citations has not been demonstrated so far: we can publish one, but we do not sell it as a lever.

The team that measures is also the team that builds sites in Next.js. Where access allows, it applies the fixes itself, from robots.txt to structured data; otherwise it writes precise tickets for your developer. The page you are reading follows these rules: a short answer up top, question headings, a marked-up FAQ.

Why do mentions on other sites matter so much?

An assistant that recommends a supplier cross-checks what you say about yourself against what others say about you. Your own site is not enough.

We start with the entity: the same name, description, services and contact details on your site, your Google Business Profile, LinkedIn and professional directories. A company fact sheet becomes the reference for anyone publishing on your behalf.

Then come credible mentions: business press, trade associations, partners, trade fairs, client reviews and genuine contributions to the forums of your industry. We do not buy links or publish mass-generated text: these practices invite Google penalties. A Wikipedia page is only worth pursuing if the notability criteria are genuinely met.

How does GEO adapt to Tunisia, Europe and the Gulf?

In Tunisia, business searches happen mostly in French, and naming a city or the country changes the answer: "in Sfax", "in Tunisia". For exporters, the decisive questions are often asked by foreign buyers, in French or English.

In Europe, starting with France, then Belgium or Switzerland, questions are more specific and competition within answers is denser. We work on Paris time in winter, one hour apart in summer.

In the Middle East, in the UAE, Saudi Arabia or Qatar, your clients switch between Arabic and English. We measure each language separately, because an assistant can cite you in English and ignore you in Arabic. Arabic content is written natively, displayed right to left and aligned with the English version; for the site itself, see our multilingual & RTL websites. Dubai is only three hours ahead of Tunisia.

What can we not promise?

Nobody can guarantee a citation by ChatGPT or Gemini. Their providers do not publish the detail of their criteria, models change, and part of what they know comes from training data frozen at a given date. A fix on your site shows up faster in answers based on a live web search than in a model's internal memory, which only renews with new versions.

Part of the outcome depends on you: time from your experts to review pages, information kept up to date, a review request after each satisfied client, permission to name a partner. Your accounts stay in your name, and the baseline report belongs to you whether or not you continue with us. Our GEO guide explains these mechanisms in more depth.

How do you get started?

It all starts with the free SEO & GEO audit: a first report of your citations on a few key questions, the state of AI crawler access to your site and the priority fixes. Request it through the project form. We come back with a clear diagnosis and a scope proposal, quoted after the audit; monthly tracking plans are described on the Pricing page.