More and more people look for information by asking AI. AI answers in search engines, conversational AI assistants and AI search services compile an answer from several sources and cite some of them. For a business this raises a new question: when a customer asks AI about services in your industry, is your company mentioned in the answer?
GEO (Generative Engine Optimization) means developing a website's content and structure so that AI-based services find, understand and mention it. The term became established following a research paper published in 2023.
How does AI search choose its sources?
The sources of an AI answer come, roughly speaking, from two places:
- The model's training data: the model has learned information from the public web before its training ended.
- Real-time search: at the time of answering, the service retrieves information from a search index or the web and compiles the answer from the pages it finds.
The latter matters more for maintenance, because you can influence it quickly. Real-time search often relies on search indexes, so a page must first be findable and indexable. That is why GEO starts from the same basics as technical SEO. Indexability, structure, speed and mobile-friendliness are the same in both, as eLuotsi's article SEO vs. GEO – myths busted (in Finnish) points out.
In real-time search, the language model splits the question into several background searches (query fan-out) and builds the answer from the sources it finds with them. According to research findings compiled by eLuotsi, these background searches are often made in English even when the user asks in another language, such as Finnish. English-language content can therefore improve a non-English business's chances of appearing in answers. The topic is covered in more detail in the article AI searches in English in the background (in Finnish).
Maintenance foundations for GEO
Indexability and bot access
Check that robots.txt or the server firewall does not block the search bots of AI services, if you want to appear in their answers. Some services use different bots for model training and for real-time search, in which case you can allow search and disallow training use. Check the bot names and their purposes in each service's own documentation, because they change.
Content in HTML
Many bots do not execute JavaScript as thoroughly as a browser. If the most important content of a page only loads with JavaScript, the bot may not see it. Make sure the text content is included in the HTML returned by the server.
Speed and availability
A slow site, or one that is often down, is more easily left out when a service retrieves sources in real time. Read more in Site speed and Core Web Vitals.
Structured data
Structured data (schema.org) states unambiguously what the company is, what it offers, where it operates and who wrote the content. Organization or LocalBusiness data, author and date information on articles, and FAQ structures in particular help machines interpret content.
Content that AI cites
Based on research and practical observation, AI answers favour content that:
- Answers the question directly. A concise answer right below the heading, with reasoning and details after it. That is why this article starts with an In brief section.
- Backs up its claims. Figures, sources, dates and concrete examples. The original research paper found that adding citations, statistics and quotations improved the visibility of content in generative answers.
- Is up to date. A visible update date and content that has genuinely been updated.
- Uses a clear structure. Descriptive subheadings, tables, lists and definitions from which a single point can be lifted out on its own.
- Is unambiguous about the company's details. Name, services, area of operation and contact details in the same form on the website, in business profiles and in directories.
Mentions outside your website
AI also forms its picture of a business from other sources: news, industry publications, review services, directories, Wikipedia and discussion forums. Your own website alone is therefore not enough. Consistent and up-to-date information about the company across different sources reinforces that AI identifies the business correctly and connects it with the right topics. The importance of external visibility and mentions is discussed in eLuotsi's article What matters most in AI optimisation? (in Finnish).
Tracking visibility
Visibility in AI search cannot be measured with rankings alone, because answers vary with the person asking, the wording and the time. A practical tracking model:
- Draw up a list of 20–50 questions your customers are likely to ask.
- Track monthly whether your business is mentioned in the answers and whether your site is cited.
- Also track competitors' mentions for the same questions.
- Measure referral traffic from AI services in your analytics.
Commercial tools have also been developed for tracking, running questions automatically across several services.
For Google's own AI features, data is also available in Search Console. The Generative AI report, added in August 2026, shows how many impressions a site has received in features such as AI Overviews and which pages appear in them. The report does not include search queries or clicks, as explained in eLuotsi's article Google Search Console now offers new data on AI visibility (in Finnish).
GEO checklist for maintenance
robots.txtand the firewall do not block the AI bots in whose answers you want to appear.- The most important content of each page is in the HTML, not dependent on JavaScript alone.
- Structured data states the company's details and the authors and dates of articles.
- The most important pages start with a direct answer and use clear subheadings.
- Claims are backed up with figures and sources, and the update date is visible.
- The company's basic details are consistent across all sources.
- Visibility is tracked regularly with a chosen list of questions.
Frequently asked questions
Does GEO replace SEO?
No. GEO builds on SEO. Many AI search services retrieve information through search indexes, so a page must first be findable and indexable. GEO widens the goal: besides ranking, you aim for a mention and a citation in the AI's answer.
Should AI bots be allowed in robots.txt?
That is a business decision. If the goal is to appear in the answers of AI services, the bots that serve their search features should be allowed. Some services use different names for the bots that collect model training data and the bots that serve real-time search, so you can decide on them separately. Check each service's documentation for the current bot names.
What is llms.txt?
llms.txt is a proposed file format that gathers a site's key content in a form that is easy for language models to read. It is not an official standard, and not all services use it. Adding one is little work, but there is no established evidence of its effect on visibility.
How is visibility in AI search measured?
Choose a set of questions your customers ask and track regularly whether your business is mentioned in the answers and whether your site is cited as a source. Commercial tracking tools have also been developed for this. Impressions in Google's AI features can be followed in the Generative AI report in Search Console. You can also track referral traffic from AI services in your analytics.
Sources
- Aggarwal et al.: GEO – Generative Engine Optimization (arXiv)
- Google Search Central: AI features and your website
- OpenAI: Overview of OpenAI Crawlers
- The llms.txt proposal
- eLuotsi: SEO vs. GEO – myths busted (in Finnish)
- eLuotsi: AI searches in English in the background (in Finnish)
- eLuotsi: What matters most in AI optimisation? (in Finnish)
- eLuotsi: Google Search Console now offers new data on AI visibility (in Finnish)
- GEO-opas.fi – a Finnish-language guide to GEO