The migration from traditional SEO to Generative Engine Optimization (GEO) is no longer theoretical for small businesses. NavetteForYou, a private airport transfer service based in Marrakech, Morocco, has published a detailed breakdown of how they implemented the llms.txt specification to ensure their services are correctly indexed by AI-powered search engines like ChatGPT, Perplexity, and Google AI Overviews. The case study highlights a pragmatic approach to making unstructured web content machine-readable for large language models (LLMs).

The Technical Implementation of llms.txt

At the core of their strategy is the creation of a plain-text file named llms.txt, served from the domain root. Unlike robots.txt, which manages crawler permissions, llms.txt provides a structured summary of the website’s offerings. The file explicitly defines the entity 'Marrakech Transport,' its operator 'Serenity Morocco Tours,' and specific services such as airport transfers from Marrakech Menara Airport and day trips to Essaouira or the Sahara. The team emphasizes keeping the file under 2,000 tokens to prevent truncation by AI crawlers with limited context windows during ingestion.

Beyond the Text File: A Full GEO Stack

The article argues that llms.txt is insufficient in isolation. NavetteForYou implements a broader technical stack including Schema.org entity graphs, where the operator is defined as an Organization with a canonical @id linked to over 15 verified profiles on platforms like Wikidata and LinkedIn. Furthermore, each route page features passage-level citability, starting with self-contained paragraphs that directly answer user queries (e.g., 'How do I get from A to B?'). This structure allows LLMs to extract clean, quotable text without navigating complex HTML navigation or scripts. The site also utilizes FAQPage structured data, as LLMs favor Q&A pairs that map directly to their query-answer format.

Measuring Success in Generative Search

Tracking the impact of GEO requires new metrics beyond traditional organic traffic. The team monitors brand mentions within AI responses by querying ChatGPT, Perplexity, and Copilot about local service categories. They also analyze referral traffic specifically from domains like chat.openai.com and perplexity.ai. Validation of structured data using tools like Google's Rich Results Test remains critical to ensure that the underlying markup supports the AI's ability to parse and cite the information correctly.

Key Takeaways

  • llms.txt acts as a lightweight, structured summary for AI crawlers, distinct from robots.txt.
  • Keeping llms.txt under 2,000 tokens ensures full processing by LLM context windows.
  • Schema.org entity graphs and multi-platform consistency increase AI confidence in citing a source.
  • Passage-level content design facilitates easier quoting by generative engines.

The Bottom Line

SEO is not dead β€” it is evolving. The sites that win in 2027 will be the ones that speak both Googlebot and GPT. An llms.txt file takes ten minutes to create and costs nothing to maintain. If you run a local service business, start there.