The four layers
Crawler access comes first: robots.txt has to allow the agents that matter, among them GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and CCBot, and the content has to survive with JavaScript disabled, because several of those agents do not execute it.
Extractability is the second. A passage that answers the question in its first sentence and stands on its own, without needing the three paragraphs above it, is the unit a model can lift. Tables, glossaries and FAQ blocks are the formats that survive extraction best.
External consensus is the third, and the one a site cannot fix on its own: what independent sources say about the entity. Wikidata, review platforms, sector press, forums. A model that finds one description on your site and nothing anywhere else has no reason to repeat it.
Citable format is the fourth: structured data that matches the visible content, dates the reader can see, figures with a source.
How it differs from SEO
Classic SEO optimises for position in a ranked list of links. GEO optimises for being the passage a model quotes inside an answer where no list is shown. The technical groundwork overlaps almost completely, the priorities and the metrics do not.
On measurement
No public API today gives a reliable share of voice inside LLM answers. Measurement is periodic manual prompting of a fixed question set, recorded over time so the trend means something. A supplier quoting an exact visibility percentage is quoting an estimate, and the method behind it is worth asking for.