Notes on generative engine optimization

Read them in order. The notes cover what the GEO paper means by the name, how the public ChatGPT and Perplexity documents describe sources and crawlers, why an Arabic question has to be read on its own, what schema.org, JSON-LD, and llms.txt each do, how to read sample size, confidence intervals, and snapshots, and where these materials are easiest to misuse. Every figure comes from a linked paper or page. These notes do not publish brand results.

Published 2026-10-14

1. What generative engine optimization isThe name GEO comes from a 2024 KDD paper. It is about visibility inside a generated answer, not a classic search rank, and it promises no brand a citation.
2. What the public docs say about ChatGPT and Perplexity sourcesThe public docs describe citations, location or language parameters, and which crawler is for search or training. They do not publish a complete ranking formula.
3. Arabic questions need their own visibility readingArabic script is written right to left. Language and country are not one field. Common Crawl’s language shares are not AI citation rates.
4. What schema.org, JSON-LD, and llms.txt each coverStructured data is a vocabulary and a syntax. llms.txt is a proposal. The OpenAI and Perplexity documents checked here do not list it as a citation input.
5. How to read AI visibility with sample size, intervals, and snapshotsOne answer is not a percentage. A proportion needs a sample size and an interval. A snapshot records the question, language, country, date, and raw answer.
6. Where these materials are easiest to misuseCommon misuses include treating one answer as a percentage, pooling country and language, treating a paper’s gain as a promise, and treating markup or llms.txt as a citation requirement.