I pulled up our server logs for 25 July to 8 August 2026. Requests classified as AI crawlers reached two content URLs on milcomp.hu: the homepage and this guide. The sitemap was the third URL reached; I count it separately. The raw request count was misleading, because it also included requests using fake bot names, so I track which content gets reached instead, and I verify who the crawler really is as a separate step. Being reached is only the start, though. GEO (generative engine optimization) is about becoming a source the AI answers quote.

How is it different from SEO?

GEO does not replace SEO, it builds on it. Without classic search optimization (a crawlable, fast, well-structured site and relevant content) there is no foundation for AI visibility either. The difference is the goal: in SEO you compete for a position on a list, in GEO you compete for the generated answer to cite YOU as its source. That takes a different way of writing, and a few technical conditions have to hold.

How AI systems choose

  • They must be able to reach your site. Search and model training are two different uses. At OpenAI, OAI-SearchBot supports ChatGPT search and GPTBot is the training crawler; Anthropic draws the same line between Claude-SearchBot and ClaudeBot. So you can let the search crawlers in without saying yes to every training use. And do not stop at robots.txt: check what your CDN and firewall answer too.
  • They read server-rendered text. If your content only appears after JavaScript runs, most AI systems see an empty page. This affects a large share of small-business websites.
  • They quote facts, not marketing copy. Concrete numbers, definitions, comparisons and step-by-step lists make it into answers; "market-leading, customer-centric solutions" never does.
  • They look for consensus. If consistent information about your company and your field appears in several independent places, the chance of being treated as a source grows.

The technical basics: what to ask your developer for

  • The crawlers that matter for your search goals have to be able to reach your public content. For Google's AI search features that means Googlebot access and the Search snippet controls; Google-Extended does not control them.
  • Server-side rendering or static generation: the full content should be in the HTML.
  • Structured data (schema.org): Organization, Service and Article markup with one consistent set of company facts.
  • On a multilingual site, correct hreflang, so the systems see the right language version.
  • Internal links a crawler can follow and an up-to-date sitemap; an RSS feed can help readers keep up with new articles. llms.txt is an optional extra: Google does not require it for its AI features.
  • Speed: a slow site gets crawled less often and less deeply.

Content that AI actually quotes

The most important principle: your own verifiable fact is worth more than any opinion. What works: precise definitions ("what is GEO"), claims backed by numbers, honest comparisons ("X or Y, and when"), step-by-step guides, and subheadings phrased as questions, because these systems answer questions. One topic per page, a clear heading hierarchy and short paragraphs that stand on their own all help extractability. And yes, writing honestly about who your product is NOT for works too; I think it is the strongest trust signal a page can send.

Our own example, with numbers

I applied all of this to our own site first. In the August 2026 rebuild of milcomp.hu, the machine-readable text on the 26 URLs we measured grew from 21,737 to 45,885 characters, with the right hreflang on 22 pages. The main text became readable without any JavaScript running in the browser. Those are dated measurements of what we changed about accessibility. They do not tell you how many AI answers will cite us.

How to measure whether it works

AI visibility has no mature measurement tool yet the way classic rankings do. What I do: I regularly ask ChatGPT, Perplexity and Google (AI Overviews) the questions our customers actually ask, and record whether the company appears and what gets cited. I watch Search Console for brand and question-shaped queries. And in the traffic sources I watch the visitors arriving from chatgpt.com and perplexity.ai. The trend matters, not the one-off snapshot.

When GEO is not worth paying for

I will say it, because few people do: if the foundation is missing, money spent on GEO is wasted. It is not worth it if your site is three service descriptions and a contact page, because there is nothing to quote. It is not worth it if your customers do not research online, or if the site is about to be rebuilt anyway; build that first. In those cases the next step is content and fundamentals, not AI visibility. :)

When to bring in help

If the technical part (rendering, structured data, hreflang, speed) is not in order, no amount of content writing makes up for it. My order of work: first an AI visibility audit that shows what Google, ChatGPT and Perplexity see of you today, then prioritised technical fixes, and only on top of those a content plan. See our AI SEO and GEO service for the details, or request an audit. If you want to read more before deciding, our AI invoice processing article is a good example of the kind of content AI systems like to cite.