No public ranking formula describes every AI provider’s source selection. Focus on qualities you can verify rather than invented weights or a guaranteed citation recipe.
Make information accessible
Keep important content in readable page HTML, use stable URLs, and check successful responses. Review crawl preferences and relevant search indexing requirements. JavaScript-only shells, access restrictions, and broken links can prevent a system from obtaining the information.
Answer the actual question
Use clear headings, a direct answer, and enough detail to explain conditions and exceptions. Publish current product facts, policies, prices, and contact details. Link related pages when the reader needs a comparison or a next step.
Make claims verifiable
Identify authors and organizations clearly. Cite primary sources, distinguish measurements from opinions, and state the scope and date of statistics. Accurate structured data can describe visible content, but it does not guarantee selection.
Understand different retrieval paths
Models may answer from existing knowledge or use search, fetched pages, uploaded files, and connected sources. Some retrieval systems split documents into passages; others use larger sections or full documents. There is no universal 100–300-word limit or single provider-independent pipeline.
Google’s AI features documentation explains its search eligibility and use of query fan-out. The llms.txt proposal provides a complementary content guide, not a substitute for accessible pages.
Measure what changed
Compare crawler access, human referrals, controlled answer tests, and your own chat questions separately. Avoid treating a crawler hit as a citation. See the metrics guide and content practices.