AI readability is the answer on the page
LLM visibility scores promise to measure whether ChatGPT will mention you. Most of them measure a prompt the vendor chose. Read the page instead.
What answer engines can cite
A stable URL, a clear answer in the first screen, and facts that match the visible text. Schema and llms.txt help them find the source. They do not rescue a page that is a slogan, a clone, or a noindex.
Why the score misleads
A readability formula counts sentences. An LLM visibility score counts whether a private prompt returned your brand. Change the prompt and the score moves. You cannot file a ticket against that. You can file a ticket against a missing explanation.
The check to run
Put the answer above the fold. Name the product, the price, and the limit in text. Add only the schema you can defend. Publish llms.txt that points at the canonical pages. Then run a crawl so those pages are actually linked.
Answer first
The citation is a sentence you already wrote.
One URL
Do not make the model choose among ten clones.
llms.txt
A map for assistants, not a keyword list.