AEO and GEO: how your business gets recommended in ChatGPT and Perplexity
More and more people ask the AI first and Google second. If you are not there, you lose enquiries without ever seeing it in a ranking report.
The three terms, sorted
SEO targets placement in the classic result list. AEO (answer engine optimisation) targets being served as the direct answer, for example in featured snippets or AI overviews. GEO (generative engine optimisation) targets being cited and recommended inside the answers of generative systems such as ChatGPT, Perplexity or Gemini.
Answer first, context second
Language models extract passages. A page that takes three paragraphs to get to the point is cited less often than one that answers in the first paragraph and explains afterwards. That order feels wrong while writing and still works better, for humans too.
Verifiability beats marketing language
"Leading agency" is worthless to a language model. "Self-employed since 2012, locations in Merzhausen and Heidenrod, projects between €3,000 and €15,000" is usable. Concrete numbers, places, timeframes and names make a provider tangible for AI systems.
Structured data is the translation layer
Schema.org markup for organisation, services, FAQs and articles tells machines explicitly what is on the page. It does not replace good writing, but it makes it easier to use. Important: the markup must match the visible content, otherwise it is ignored or penalised.
Do not lock the crawlers out by accident
Many websites block AI crawlers in robots.txt without knowing it. If you want to appear in AI answers, deliberately allow GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended. An llms.txt with a short company profile is an additional, cheap lever.
Mentions beyond your own site
Language models lean heavily on sources they have seen often: industry directories, trade portals, press, review platforms. Consistent name, address and phone details across all of them is an underrated GEO factor.
How to measure it
There is no ranking list. What works is a fixed set of test questions asked regularly across several systems and logged: is the company mentioned? With what description? Is it accurate? Inconveniently manual, but honest.