Your website ranks well, your campaigns are running, and yet a new customer tells you they found you via "something on ChatGPT". Or that the chatbot actually sent them to a competitor. More and more searches no longer end in a list of blue links, but in a single AI-generated answer with a few names in it. If you are not in it, you don't get the click either. What can you do about it? The answer is generative engine optimization, or GEO for short. What does it involve, and where do you start?

What is generative engine optimization?

Generative engine optimization (GEO) aims to make a brand, product or organisation appear accurately and often in the answers of generative AI systems such as ChatGPT, Gemini, Claude, Perplexity or Google AI Mode. Where search engine optimization aims for a high position in a list of results, GEO aims for a mention, an accurate description and a source citation in the answer itself. It is sometimes also referred to as AI visibility or AI search visibility.

Note, though: GEO does not replace SEO. An AI assistant often gets its information from the same web pages that search engines look at. Only the result differs: with SEO, it is about the ranking of your page; with GEO, it is about what the model writes about you.

What GEO is not

A new term like GEO quickly creates expectations that don't always match reality. Four sober observations keep your feet firmly on the ground.

  • This is not a trick. There are no hidden settings that force an AI assistant to mention you. The model decides based on what it finds and reads about you. Full stop.
  • Measuring once is not enough. Answers differ per assistant, per region, per wording and per day. So it is best to measure continuously, to evaluate properly whether your GEO efforts are paying off.
  • You won't get certainty. Nobody knows the exact rules a model uses to decide whom to mention. Treat every piece of advice, even good advice, as a hypothesis you test against your own measurements.
  • Good content remains the foundation. An assistant can only cite what there is to cite. Clear, factual web pages remain the starting point.

What you can measure

You can't improve what you can't see. Fortunately, AI answers are measurable, provided you ask the question the same way every time. Four data points get you a long way:

  • Mention. Does your brand appear in the answer to a question a customer asks?
  • Share. What share of all mentions of the tracked brands goes to you over a period, and what share to your competitors? That is your Share of Voice.
  • Description. How are you mentioned: as first choice, as an alternative or with a caveat?
  • Sources. Which pages does the assistant base its answer on? Those pages deserve your attention, even if they are not yours.

In GripIQ, this happens in Prompts. There you record questions the way a customer asks them, optionally linked to a persona. With every scan, GripIQ retrieves the answers from ChatGPT, Gemini and Claude, with web search enabled just like an ordinary user, and from Google AI Mode. For every answer, you see the sources it relies on. You filter per assistant, and the region of a Prompt determines which market the answer is retrieved for. The analysis extracts the mentioned brands, the sentiment and the tone from every answer.

What a first measurement looks like

Say you are the marketing manager at a provider of accounting software for freelancers. You write down ten questions a customer would ask, such as "which accounting software suits a freelancer in Belgium?" or "what is an alternative to this or that package?". You enter them as Prompts. Under Brands, you add your own brand and three competitors.

After the first scan, you see three things: in how many answers your brand appears, who is listed next to you and which sources the assistant cites. That last one often yields the most. If a competitor keeps showing up via a comparison site where you are missing, you know straight away where your next step lies.

Where to start this week

You don't need to wait for a big project. You can apply this sequence right now:

  • Start with what customers really ask. Take your questions from sales conversations and from customer service, not from your own product names.
  • Measure first, then write your content. Without a baseline, you never know whether a change makes a difference.
  • Read the sources. The pages that show up in the answers are the ones that make the difference. Make sure your pages are among them: clear, factual and up to date.
  • Close the gaps. If an answer cites another brand, or stays vague, you have a topic for a new article or a better page. How to turn that into a calendar is explained in How to build a content strategy on the questions AI answers leave open.
  • Track each assistant separately. Google AI Mode behaves differently from a chat assistant. You can read what that changes in Google AI Mode: how it changes the way your brand is found.

Don't expect fixed rules

GEO is a young discipline. What works today may be outdated in a year, because the models and their sources change. That is why your own measurements carry more weight than any advice, including the advice in this article. The cycle stays the same: record the question, measure the answer, read the source, improve the page, measure again. That way, you don't depend on a tip that will be outdated tomorrow.

What you can do now

  • Write down ten questions today that a customer would ask an AI assistant about your category.
  • Put them to two assistants and note who is mentioned and which sources appear.
  • Pick one page that belongs among those sources, and improve it this week.

Would you rather measure this continuously, for several assistants and regions at once? Book a demo: we'll set up your first Prompts together with you.