Context window
A context window is how much text an AI model can take in at once: the question, the conversation so far and any pages it reads.
A context window is how much text an AI model can work with at once. Everything the model uses to write an answer has to fit in it: the question, the conversation so far, its instructions, and any pages or passages it pulled from the web.
Context windows are measured in tokens, which are pieces of words. They've grown large, but they're still a limit, and a platform answering a question usually fills its window with parts of many pages rather than the whole of one.
Why it matters for your pages
When a platform searches the web before answering, it rarely reads every word of every page it finds. It picks out the passages that look most relevant and puts those into the window, next to passages from other sites.
So the parts of your page that do the work are the parts that make it in. A clear passage that answers the question has a good chance. A key fact buried in paragraph twelve, after the company history, may never be read.
For example, Fernhill's pricing page opens with 800 words about where its beans come from, and the prices sit in a table at the bottom. Kestrel's page opens with "Plans start at £12 a month for 250g, delivered weekly or fortnightly." Ask a platform what the two subscriptions cost, and Kestrel's price is easy to find, while Fernhill's may be missing from what the platform reads.
Conversations use up the window too
In a long chat, earlier messages take up room. Details from the start of a conversation can carry less weight later, or fall out. That's one reason the same question can get a different answer in a long chat than as a fresh question.
In AnswerOrbit
One of the eight Page checks in Recommendations is an opening that answers: whether your page answers the prompt in its first lines. Another checks that the prompt appears in a heading.